<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Donald Harmitt on Medium]]></title>
        <description><![CDATA[Stories by Donald Harmitt on Medium]]></description>
        <link>https://medium.com/@donaldharmitt?source=rss-d85d57a8baaf------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*GjmhUT5cw9uPv-RoGI_URA.jpeg</url>
            <title>Stories by Donald Harmitt on Medium</title>
            <link>https://medium.com/@donaldharmitt?source=rss-d85d57a8baaf------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 06:52:31 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@donaldharmitt/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Benchmarks Are the New Battlefield: What Quantum Research Tells Us About the Next Frontier]]></title>
            <link>https://medium.com/@donaldharmitt/benchmarks-are-the-new-battlefield-what-quantum-research-tells-us-about-the-next-frontier-8e79db972dbc?source=rss-d85d57a8baaf------2</link>
            <guid isPermaLink="false">https://medium.com/p/8e79db972dbc</guid>
            <category><![CDATA[research]]></category>
            <category><![CDATA[quantum-computing]]></category>
            <category><![CDATA[quantum]]></category>
            <category><![CDATA[investing]]></category>
            <dc:creator><![CDATA[Donald Harmitt]]></dc:creator>
            <pubDate>Fri, 15 Aug 2025 13:35:56 GMT</pubDate>
            <atom:updated>2025-08-15T13:35:56.242Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/860/0*zq5h82E2IqnaofLO.jpg" /><figcaption>Image credits: <a href="https://www.nextgov.com/emerging-tech/2024/09/so-you-want-build-quantum-computer/399478/">Nextgov</a></figcaption></figure><p><em>Imagine a race where every runner is on a different kind of terrain. Some dash across smooth, paved roads, others climb icy slopes, and a few navigate sandy deserts. This is what the quantum hardware landscape looks like today. Each modality of quantum computing hardware has its own terrain, advantages, and obstacles. While private R&amp;D progress remain opaque, academic and preprint literature provides a useful lens into how these systems are performing in the open. Peer-reviewed studies and experimental benchmarks offer early signals on which hardware types are showing credible promise across experimental and real-world applications.</em></p><p><em>Whether you’re an investor trying to find your niche, a researcher tracking breakthroughs, or a strategist exploring quantum’s next frontier, this deep dive aims to provide a comprehensive overview of the research progress made across four of the most mature modalities: </em><strong><em>superconducting circuits</em></strong><em>, </em><strong><em>trapped ions</em></strong><em>, </em><strong><em>neutral atoms</em></strong><em>, </em><strong><em>and photonics</em></strong><em>, by focusing on four high-impact use cases: </em><strong><em>optimisation</em></strong><em>, </em><strong><em>simulation</em></strong><em>, </em><strong><em>machine learning (ML)</em></strong><em>, and </em><strong><em>cryptography</em></strong><em>. Drawing from a wide range of recent papers (peer-reviewed and pre-prints), this analysis identifies the technical progress made within each modality and its relevance to specific industry use cases</em><strong><em>.</em></strong><em> The goal is not to predict a single winner, but to help researchers, enthusiast, and investors understand how performance profiles align with distinct application verticals. For investors, this research view serves as a practical indicator of where differentiated value may start to emerge.</em></p><p><em>There are also emerging modalities such as topological qubits, with an example being Microsoft’s recent </em><a href="https://news.microsoft.com/azure-quantum/"><em>Majorana 1 chip</em></a><em> (however, this modality and others will not be explored further in this deep dive due to the level of tech maturity). Similarly, minimal publicly available research has been carried out with silicon-qubits (another major modality) across the four use cases, hence, this modality is not included in the deep dive.</em></p><h4>📖<strong>TL;DR:</strong></h4><ul><li>IBM’s <strong>superconducting</strong> quantum chips has solved complex optimisation problems using 127 qubits, beating previous results by a factor of 1500 and outperforming competing systems such as trapped ions in both speed and accuracy.</li><li>Quantinuum’s <strong>trapped ion</strong> computer delivered the most accurate results on certain optimisation tasks and became the first to simulate a hydrogen molecule using error corrected logical qubits, marking a major milestone toward practical quantum chemistry.</li><li><strong>Neutral atom</strong> systems tackled a notoriously difficult problem using 289 qubits, showing a rare case where quantum computing clearly outperformed classical methods and provide a strong early signal of real quantum advantage.</li><li><strong>Photonic</strong> quantum chips performed especially well in tasks involving smooth and continuous variables, such as optimising fuel use and classifying images, by using light-based processors and quantum inspired versions of neural networks.</li></ul><p><strong>Quantum computing is gearing up to move on from the noisy era</strong></p><p>To make sense of recent research, we need to understand the current quantum computing technological landscape. Today’s quantum computers are error-prone, and incapable of continuous quantum error correction required to achieve fault-tolerance. Even though progress is rapid and the technology has come a long way, we are still in what is known as the<strong><em> </em></strong><a href="https://thequantuminsider.com/2023/03/13/what-is-nisq-quantum-computing/"><strong><em>noisy intermediate-scale quantum</em></strong></a><strong><em> (NISQ) era</em></strong>. Noisy, referring to the error-prone nature caused by external factors, e.g. temperature change, and intermediate scale, referring to the volume of qubits which limits computational power.</p><p>Quantum bits or qubits, which are the building blocks of quantum computers, are very unstable and highly susceptible to noise which manifest in many forms such as vibrations, temperature changes, and so forth. Hence, one of the key focus areas regarding the long-term viability of the various modalities, is their coherence capabilities. Simply put, <a href="https://www.findlight.net/blog/quantum-coherence-simple-guide/#What%20is%20Quantum%20Coherence?"><strong><em>coherence</em></strong></a> is how long a qubit retains its ‘useful’ information (quantum mechanical state) without being interrupted by external noise (leading to decoherence). All computational steps need to occur during this coherence timeframe, hence, the longer coherence time, the better the system will be at avoiding errors, and the longer it has to perform its operations in a useful state. As a result, different hardware developers are working on extending and enhancing coherence capabilities in their respective systems using techniques such as quantum error correction and decoherence suppression.</p><p>Both <a href="https://www.bcg.com/publications/2019/quantum-computers-create-value-when">Boston Consulting Group</a> (BCG) and <a href="https://azure.microsoft.com/en-us/blog/quantum/2023/06/21/microsoft-achieves-first-milestone-towards-a-quantum-supercomputer/">Microsoft</a> have defined the quantum computer technology growth trajectory in 3 eras, growing from the NISQ era to large-scale, universal, fully fault-tolerant computers, and even quantum super computers. In order to make the transition to the next 2 eras, improvements in error-correction, qubit performance enhancement, and noise reduction, are some of the pivotal milestones required to achieve full commercialisation.</p><p>According to Microsoft, the quantum hardware development will occur in these 3 stages:</p><ul><li><strong>Level 1</strong>: Noisy-Intermediate Scale Quantum (NISQ) systems which are <strong><em>highly error-prone and made of “noisy” qubits</em></strong>. In this era, progress is measured by the number of qubits.</li><li><strong>Level 2</strong>: Quantum systems that <strong><em>operate primarily on reliable logical qubits</em></strong>. In this era, scaling becomes more feasible due to the transition from noisy physical qubits to reliable logical qubits, and measurement will be based on logical qubit error rates. <em>Note: logical qubits are a “form of quantum error correction leveraging multiple physical qubits to create a more stable single qubit unit, protecting against errors”.</em></li><li><strong>Level 3</strong>: Quantum computers can solve <strong><em>impactful </em></strong>problems which classical supercomputers are incapable of. In this era, progress is measured on what is known as reliable <em>Quantum Operations Per Second (rQOPS)</em>, which is effectively how many reliable operations can be executed in a second.</li></ul><p><strong>A quick guide to modalities: pros, cons and practical implications for investors</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tipIvxALpjOc6i5QzNIksQ.png" /><figcaption>Source: Image created by author (sources are available at the end of the article)</figcaption></figure><p>Quantum computing holds the potential to revolutionise numerous specific domains, leveraging its superior capabilities compared to classical computing approaches. However, each hardware modality is a trade-off. Understanding these trade-offs is essential for matching them to real-world use cases. For instance, trapped-ion systems show great promise in coherence times but difficulty in scalability due to entanglement challenges, while photonics can perform at room temperature (which is cheaper as no cryogenic equipment is required), with great computational speed, but has relatively lower fidelity (which is the accuracy and reliability of a qubit) due to photon loss.</p><p><a href="https://www.techtarget.com/searchcio/tip/The-6-different-types-of-quantum-computing-technology#:~:text=2">Superconducting qubits</a> have made headlines with their rapid gate speeds, high qubit counts, and integration with semiconductor fabrication, making them strong contenders for scalable quantum systems. Their nanosecond gate times suit high-speed applications such as Monte Carlo simulations, and leading platforms like IBM and Google are pushing boundaries with hardware scale and gate fidelity. However, superconducting systems face hurdles in maintaining coherence and managing crosstalk, as well as significant cryogenic and wiring demands that complicate large-scale deployments.</p><p><a href="https://thequantuminsider.com/2022/02/19/6-quantum-computing-companies-working-with-trapped-ions/">Trapped ion systems</a>, in contrast, offer exquisite gate fidelity and remarkably long coherence times, supporting complex algorithms with deeper circuits and all-to-all connectivity. Yet their slow gate speeds and operational complexity, particularly due to laser control requirements, present bottlenecks in real-time computing tasks and practical scalability.</p><p><a href="https://thequantuminsider.com/2024/02/22/harnessing-the-power-of-neutrality-comparing-neutral-atom-quantum-computing-with-other-modalities/">Neutral atoms</a> bring room-temperature operation, long coherence, and potential for highly parallel architectures, as evidenced by <a href="https://www.newscientist.com/article/2399246-record-breaking-quantum-computer-has-more-than-1000-qubits/">recent scaling milestones</a>. But their weaker fidelity, gate speed limitations, and the delicate optical controls required for precise operations remain technical hurdles for mainstream use.</p><p><a href="https://thequantuminsider.com/2024/02/22/harnessing-the-power-of-neutrality-comparing-neutral-atom-quantum-computing-with-other-modalities/">Photonic systems</a> flip the quantum script entirely by leveraging light as the information carrier, allowing operation at room temperature and enabling long-distance quantum communication. They excel in quantum key distribution and modular connectivity but suffer from the probabilistic nature of entanglement and the challenge of photon loss, making fault-tolerant scaling a steep climb.</p><p>Still, hybrid approaches are emerging to bridge strengths across platforms. Startups like <a href="https://photonic.com/">Photonic</a>, backed by the likes of <a href="https://www.amadeuscapital.com/">Amadeus Capital Partners</a> and other promiment deep tech investors, are combining silicon-spin qubits with photonic links to unlock scalable, networked quantum computing, signalling that the future may lie not in choosing one modality, but in strategically blending them.</p><p>Ultimately, when analysing the various quantum hardware modalities, investors should factor in the following checklist diagram to inform their decision-making:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*A4R4-lE3v51qW6O5ZhUC8g.png" /></figure><p>Now onto the use cases….</p><p>🎯<strong>Use case 1: Optimisation: where qubits meet the travelling salesman</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*NdjTWNCubtW2iov6.jpg" /><figcaption>Image credit: <a href="https://physicsworld.com/a/conquering-the-challenge-of-quantum-optimization/">Physics World</a></figcaption></figure><p>Let’s dive into the first and perhaps most commercially tantalising use case: optimisation. Here’s where quantum might deliver speed and accuracy over classical approaches, particularly for areas such as finance and logistics.</p><p>Combinatorial optimisation problems are seen as a promising area for quantum computers due to their broad applicability and the potential for speedups over classical methods. Quantum algorithms like the <a href="https://arxiv.org/abs/2306.09198">Quantum Approximate Optimisation Algorithm (QAOA)</a>, and quantum annealing systems, have been developing overtime to enhance optimisation-focused tasks. Trapped-ion systems have gone as far as showing specific application to finance-related tasks (portfolio optimisation) <a href="https://arxiv.org/abs/2305.03857">using 32 qubits</a>. This offers significant promise that quantum computing could potentially revolutionise financially related optimisation use cases, alongside other subsectors, by significantly improving computational speeds and accuracy as the systems scale.</p><p><strong>Superconducting processors </strong>have been demonstrated for binary combinatorial optimisation problems, essentially finding the best “yes-or-no” choices to solve a complex problem, successfully solving problems with up to 127 qubits. The research by <a href="https://arxiv.org/abs/2406.01743">Sachdeva et al</a>. used a Q-CTRL solver (tool that improves quantum algorithms by reducing errors and boosting performance) on IBM superconducting processors to successfully demonstrate the ability to outperform published results from quantum annealers for certain non-trivial binary optimisation problems, e.g. <a href="https://blog.algoanalytics.com/2023/04/20/business-applications-of-the-maximum-cut-problem/">Max-Cut</a>. The IBM superconducting solver consistently found the known optimal solution for problems up to 127 qubits, representing the largest quantum optimisation successfully solved on hardware to date (in the public domain), outperforming D Wave’s quantum annealers by 1,500x for the same application. Not only that, but the demonstrations delivered up to 9x increased likelihood of success for identical problems up to the 32-qubit range, compared to 4x smaller implementations carried out on a trapped-ion computer.</p><p>Another study by <a href="https://arxiv.org/abs/2502.06471">Montañez-Barrera et al</a>. compared 5 quantum computing systems (superconducting and trapped-ions) and found the largest implementation (1-D chain problem) on a IMB fez (superconducting system) with p =10,000 (number of layers) and almost 1 million 2-qubit gates. However, the best performance (accuracy) were obtained on the Quantinuum H2–1 (trapped-ion system) on a fully connected graph problem. Although superconducting technology offers more rapid execution times, it is plagued by considerable 2-qubit gate errors, illustrating the ongoing trade-off between speed and accuracy within quantum hardware choices.</p><p><strong>Trapped ions </strong>are suitable for implementing Variational Quantum Algorithms (VQAs) which are hybrid algorithms designed to make the most of the combined classical and quantum computing power in the NISQ era, across optimisation problems. <a href="https://arxiv.org/abs/2502.06471">Montañez-Barrera et al</a>.’s benchmarking study leveraging an optimisation algorithm (LR-QAOA) used Quantinuum H2–1 and IonQ Aria (both trapped-ion systems), demonstrating the best performance using the Quantinuum system, as previously mentioned. However, despite the more promising 2-qubit gate error rates, and full connectivity capabilities, the trapped-ion technology is plagued by considerably slower operations. In a hypothetical problem highlighted in the study, it was denoted that the IonQ Aria system would need ~18,000 seconds to execute vs. 0.51 seconds in the IBM fez system (superconducting system), based on 2-qubit gate speeds. This highlights a key-drawback in trapped ion systems which leads to higher shot costs. Therefore, the practical adoption of trapped-ions systems may currently be limited to highly accuracy-sensitive tasks due to cost concerns.</p><p><strong>Neutral atoms (Rydberg atom arrays) </strong>trapped in optical tweezers<strong> </strong>have been used for solving the <a href="https://www.quera.com/glossary/maximum-independent-set#:~:text=The%20Maximum%20Independent%20Set%20(MIS,independent%20set%20of%20maximum%20size.">Maximum Independent Set (MIS) problem</a>, a notoriously difficult optimisation problem. The experiments conducted by <a href="https://arxiv.org/abs/2202.09372">Ebadi et al.</a> used up to 289 qubits in 2-D planes, with performance benchmarked against classical simulated annealing. The results showed that on the challenging graphs, a “superlinear quantum speedup” (performance improvements increase faster than proportionally expected, as problem complexity grows) in finding exact solutions in the deep circuit was demonstrated, highlighting the possibility of “quantum advantage” in runtime. Despite this, only a limited depth was achieved, and improvements in error-correction, qubit count, and coherence need to be improved to increase circuit depth.</p><p>When it comes to optimisation problems, <strong>photonic quantum computers</strong> encode continuous optimisation problems rather than discrete optimisation problems implemented by other modalities, e.g. trapped ions systems. Photonic quantum computers leverage information encoded in “qumodes”, which are continuous variables that carry quantum information using properties of light. Where this is advantageous is in use cases where continuous optimisation problems are costly to encode using qubits e.g. minimising fuel use in a flight path. A <a href="https://arxiv.org/abs/2307.14853">study</a> tested a hybrid quantum-classical approximate optimisation algorithm, Photonic Counterdiabatic Quantum Optimisation (PCQO), on a nanophotonic quantum chip for continuous-variable optimisation. The results successfully showed outperformance against existing hybrid adiabatic quantum algorithms <strong>illustrating superior implementation ease and convergence</strong> (the algorithm reaches optimal solutions faster and more reliably). This experiment is instrumental in providing supporting evidence that photonic quantum computing could become highly influential in fields requiring complex continuous optimisation, through the use of the PQCO algorithm on photonic chips in the near-future.</p><p><strong>💡Takeaway:</strong></p><p><em>Each modality showcases distinctive strengths in solving optimisation problems, reflecting a broader theme across the landscape: success depends on matching hardware capability to problem structure. Superconducting systems are proving adept at scaling to higher qubit counts with competitive speed, while trapped ions offer high precision for accuracy-sensitive tasks despite slower execution. Neutral atoms are beginning to show runtime advantages in specific graph problems, and photonic systems demonstrate unique potential for continuous-variable tasks that are otherwise cumbersome for traditional qubit encoding. Together, these findings suggest that commercial traction in optimisation will likely be driven by nuanced modality selection aligned with the nature of the computational challenge</em></p><p>🧬<strong>Use case 2: Simulation: modelling what can’t be modelled today</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ElsxWhEl5VQCiW4z.jpg" /><figcaption>Image credit: <a href="https://www.livescience.com/technology/computing/quantum-simulation-breakthrough-will-lead-to-discoveries-impossible-in-todays-fastest-supercomputers-google-scientists-claim">Live Science</a></figcaption></figure><p>Now we shift focus to simulation, a natural fit for quantum computers, especially for modelling molecules, reactions, and materials.</p><p>Simulating quantum-many body systems, which are systems involving interactions between multiple quantum particles like electrons, atoms, or photons, is a core application area for quantum computers, such as chemistry and material sciences problems. The exploration of molecular vibronic spectra (studying molecular vibrations to understand chemical structures) is pivotal in understanding molecular properties in analytical chemistry. Fields such as drug discovery stand to benefit significantly from quantum computing due to the need for molecular modelling and predictive analytics. Effective quantum simulation could substantially accelerate pharmaceutical and materials development processes.</p><p><strong>Superconducting qubits</strong> have been used for simulating molecular vibronic spectra <a href="https://arxiv.org/abs/2405.17385">using a 69-qubit superconducting processer</a> serving as a hybrid analogue-digital quantum simulator to study thermalisation (how quantum systems reach thermal equilibrium) and criticality (how properties change dramatically under certain conditions) in the XY model. This study importantly demonstrated experiments above the reach of classical simulation within cross-entropy benchmarking experiments. With the aim to bridge the gap from experimentation to real world applications, leveraging a superconducting computer, <a href="https://arxiv.org/abs/2401.03759">researchers have also developed a hybrid quantum computing pipeline</a> to address genuine drug design challenges. Their work focused on two outcomes: 1) accurately determine Gibbs free energy profiles for pro drug activation and 2) precise simulation of covalent bond interactions for drug-target interactions. Not only did the results help to verify that the VQE algorithm can be leveraged in “near-term” quantum computers, but it showcased the effectiveness of quantum computing (superconducting systems in this instance) within these real-world applications.</p><p>Similarly, <strong>trapped-ion systems </strong>have been used for molecular vibronic spectra simulation. Previous experiments have been carried out, showing the first <a href="https://arxiv.org/abs/1702.04859">experimental demonstration of molecular vibrational spectroscopy</a> of sulphur dioxide using a trapped-ion system. Stepping outside of the qubit realm, researchers (<a href="https://arxiv.org/abs/2310.12110">Meth et al.</a>), have used a trapped-ion <strong><em>qudit</em></strong><em> quantum processor </em>(encoding information in multiple states, more than two per ion) to demonstrate the ability to observe the effect of how matter can influence the behaviour of a magnetic field that is broken down into discrete, or “quantised” units. Alongside this, they studied the dynamics of pair creation and magnetic energy, which is effectively how pairs of particles can spontaneously appear and how energy is stored and moved around in the magnetic field. This approach simplifies experimental complexity relative to a qubit-based approach, highlighting the versatility of trapped-ion platforms. On the road to full-scale commercialisation, <a href="https://www.quantinuum.com/press-releases/quantum-chemistry-progresses-meaningfully-towards-a-fault-tolerant-regime-using-logical-qubits#:~:text=%E2%80%8DTokyo%2C%20Japan%2C%20Cambridge%2C%20UK%2C%20and,quantum%20processor%20using%20logical%20qubits">Quantinuum’s H-1 trapped-ion system</a> was the first to simulate the hydrogen molecule using 3 logical qubits and a partially fault-tolerant algorithm.</p><p>When it comes to Rydberg atoms arrays (<strong>Neutral atoms</strong>), these can function as programmable quantum simulators with examples shown in simulation of 2D antiferromagnets (materials with magnetically opposing atomic spins), using hundreds of Rydberg atoms (<a href="https://arxiv.org/abs/2012.12268">Scholl et al.</a>). They have also been used to simulate lattice gauge theories (mathematical frameworks describing particle interactions, crucial in understanding fundamental physics) and strong dynamics. Other successful simulations include <a href="https://arxiv.org/abs/2312.02265">Maskara et al.’s</a> experiment using a hybrid digital-analogue simulation toolbox and is applicable to other platforms like quantum charge-coupled device (QCCD) architecture (like a conveyor belt for trapped-ion qubits). A key achievement is the ability to extract chemically relevant spectral properties, such as excitation energies and finite-temperature susceptibilities (how materials respond to external influences at different temperatures) from these simulations.</p><p>Shifting to photonics, a large-scale <strong>photonic network</strong> using squeezed vacuum states has been used to simulate molecular vibronic spectra, using an integrated quantum photonic microprocessor chip with 16 modes. <a href="https://www.nature.com/articles/s41467-024-50060-2">Researchers</a> simulated the molecular vibronic spectra of complex molecules (formic acid and thymine) with relatively high reconstructed fidelity above 92%. Though experimental, these kinds of demonstrations pioneer the use of photonic quantum computers beyond classical computers towards commercialisation in real-world applications such as quantum chemistry.</p><p><strong>💡Takeaway:</strong></p><p><em>As these simulation studies demonstrate, quantum computers are beginning to move beyond theoretical benchmarks and into use cases with tangible impact. Superconducting systems are building hybrid pipelines for chemical simulation and early-stage drug design, while trapped-ion platforms bring versatility through both qubit and qudit architectures. Neutral atoms are enabling high-resolution modelling of material behaviours and particle dynamics, and photonic systems are delivering high-fidelity quantum chemistry simulations. These advances collectively point to simulation as one of the most viable near-term applications where quantum hardware can deliver differentiated value across industries such as pharmaceuticals, materials science, and energy.</em></p><p>👨🏻‍💻<strong>Use case 3: Machine learning: teaching quantum systems to learn</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*cL8443uzfaq3ubed" /><figcaption>Image credit: <a href="https://www.informationweek.com/machine-learning-ai/how-quantum-machine-learning-works">Information Week</a></figcaption></figure><p>What if quantum systems could learn? While still nascent, early research shows machine learning (ML) may be one of the most creative and dynamic use cases.</p><p>Experimental realisation of quantum end-to-end ML has been reported on a <strong>superconducting processor</strong>. <a href="https://arxiv.org/abs/2203.09080">Pan et al.’s</a> trained model achieved 98% accuracy for two handwritten digits and 98% for four digits, using 2 and 3 qubits, respectively. This was carried out using the Mixed National Institute of Standard and Technology (MNIST) database (a collection of handwritten digit images used to train and test ML models). The results act as an indicator for the future quantum end-to-end learning potential in solving challenging ML-related tasks, once qubits begin to truly scale.</p><p>Natural language processing (NLP) is a very promising area for quantum computing. NLP experiments have been carried out on <strong>trapped ion</strong> quantum computers with various applications such as classification and ambiguity resolution.<strong> </strong><a href="https://arxiv.org/abs/2206.02171">Widdows et al.’s</a> experiment was important in showing NLP can be performed on NISQ-era quantum computers, even though the dataset size is limited. Approaches included using single/2-qubit rotations and entangled gates for linguistic tasks, achieving a 62% average accuracy on classification tasks involving over 10,000 words. The researchers conducted an experimental demonstration of a quantum ML application with “provable guarantees” for its performance and efficiency on a 11-qubit trapped ion computer, showing promising results with matching accuracy compared to classical nearest centroid classifiers (an algorithm that labels data by finding which group’s average (centroid) it is closest to) for MNIST digits. Alongside this, up to 100% accuracy was shown for 8-D synthetic data. Even with today’s hardware limitations, quantum NLP can start to compete with classical models on specific linguistic tasks, opening the door to highly efficient language-based AI systems.</p><p><strong>Photonic systems</strong> are being developed for probabilistic ML using quantum vacuum noise (pure randomness). A <a href="https://arxiv.org/abs/2403.04731">photonic probabilistic computer using controllable photonic elements</a> (photonic probabilistic neurons, PPNs) has been implemented in a bistable optical parametric oscillator (a specialised optical device which basically acts a “neuron” in a neural network). The platform leverages quantum vacuum noise as a random seed and demonstrations included probabilistic inference for MNIST-handwritten digits using a probabilistic neural network and image generation for MNIST. Very recently, Photonic Quantum Convolutional Neural Networks (PQCNNs), which are essentially quantum-inspired versions of deep learning models tailored for image analysis, have been designed and <a href="https://arxiv.org/abs/2504.20989">experimentally implemented on integrated photonic interferometers</a> (8 and 12 modes), chip scale optical circuits that manipulate light waves as they pass through different paths, for binary image classification. This setup showed how photonic hardware can be applied to real-world ML tasks. QCNN offers potential advantages in scalability, trainability, and expressivity over classical CNNs, and the photonic implementation shows promising scaling potential.</p><p><strong>💡Takeaway:</strong></p><p><em>These early demonstrations in quantum machine learning highlight a diverse and growing toolkit across hardware platforms. Superconducting processors are showing how small-scale systems can already perform image classification with high accuracy, and trapped-ion devices are opening the door to quantum-enhanced natural language processing. Photonic systems, meanwhile, are taking a different path by exploiting quantum noise and optical circuits to build novel probabilistic learning models. While still in a formative stage, the convergence of quantum hardware with machine learning suggests strong potential for breakthroughs in data processing, model expressiveness, and training efficiency over the coming decade.</em></p><p>🔐<strong>Use case 4: Cryptography: securing the future with quantum keys</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*8R34KG_rF9Z37klv.jpeg" /><figcaption>Image credit: <a href="https://www.sigarch.org/post-quantum-cryptography/">ACM Sigarch</a></figcaption></figure><p>Finally, we explore security — where quantum could either crack encryption or secure the digital future through certified randomness and entangled protocols.</p><p><a href="https://arxiv.org/abs/2212.12372"><strong>Superconducting processors </strong>have been used to investigate quantum algorithms for factoring integers</a>, a problem with cryptographic significance. The SQIF algorithm, which leverages QAOA to optimise a “size-reduce” procedure in a classical lattice reduction algorithm (ultimately speeding up a step in breaking down complex number problems), has been applied to challenging RSA numbers (numbers used in encryption to protect sensitive information). The researchers’ estimates are that to break the RSA-2048 encryption, 372 qubits would be required with thousands of gate depth when using a superconducting system.</p><p>While <strong>trapped-ions systems</strong> have demonstrated limited cryptography-related applications so far, <a href="https://arxiv.org/abs/2407.00835">distributed quantum computing (DQC) architecture based on trapped-ions</a> and <strong>photonic </strong>interconnects enables the control of distributed quantum systems, which has applications in multipartite secret sharing (secure division of information among multiple parties). Additionally, <a href="https://arxiv.org/abs/2503.20498">Quantinuum’s H2–1 trapped ion quantum computer (via online access) has been used to demonstrate “certified randomness</a>”, which has multiple applications including information security, and is a feat which is deemed impossible to achieve with classical computations. This recent work is seen as another step towards real-world applicability of present-day quantum computers.</p><p><strong>💡Takeaway:</strong></p><p><em>With encryption and security increasingly under scrutiny in the quantum era, these efforts show that progress is being made on both sides of the equation. Superconducting systems are helping to define what it might take to break RSA encryption with quantum algorithms, bringing more clarity to the threat landscape. On the other hand, trapped-ion systems, in combination with photonic links, are laying the foundation for more secure quantum communication. The ability to generate certified randomness and enable distributed key-sharing protocols marks a key shift, indicating that quantum hardware could ultimately strengthen digital infrastructure, not just challenge it.</em></p><p><strong>Final Thoughts</strong></p><p>In summary, various quantum hardware modalities are actively being developed and applied to problems across optimisation, simulation, ML, and security. Superconducting and trapped-ion platforms show breadth across multiple areas including simulation (quantum chemistry, electron transfer), ML (learning tasks, NLP), and security (certified randomness, factoring investigation). Neutral atom arrays are demonstrating strong results in optimisation (Max Independent Set) and simulation (molecules/materials). Photonic processors are making strides in ML (PQCNNs) and exploring continuous variable algorithms for optimisation.</p><p>So, what should investors and strategists take away from this research-based overview?</p><p><strong>First</strong>, there is no one-size-fits-all quantum modality, each has a niche, and hybrid approaches may dominate.</p><p><strong>Second</strong>, technical performance is only part of the puzzle; practical deployment, cost, and ecosystem maturity matter too.</p><p>And <strong>third</strong>, research traction is a crucial indicator, but not a complete proxy for commercial readiness.</p><p>The real winners in quantum may be those who best understand how to match modality strengths to the right problem spaces, and who move early when public progress signals future commercial dominance.</p><h4>GLOSSARY (alphabetical order):</h4><ul><li><strong>Certified Randomness: </strong>randomness generated and verifiably proven to be truly unpredictable using quantum processes, important for secure cryptographic protocols.</li><li><strong>Coherence Time: </strong>the duration a qubit can maintain its quantum state before decohering due to environmental noise. Longer coherence times are critical for executing accurate quantum computations.</li><li><strong>Decoherence: </strong>the loss of quantum information from a qubit, typically caused by external disturbances such as temperature fluctuations or electromagnetic interference.</li><li><strong>Fault-tolerant quantum computing: </strong>a future stage in quantum technology where systems can run large-scale, reliable algorithms without being significantly disrupted by noise or errors.</li><li><strong>Logical Qubit:</strong> a more stable unit of quantum information created by combining several physical qubits using quantum error correction, allowing for longer coherence and fewer errors.</li><li><strong>NISQ Era (Noisy Intermediate-Scale Quantum): </strong>the current stage of quantum computing, characterised by quantum devices with tens to hundreds of noisy qubits that are not yet error-corrected or fully scalable.</li><li><strong>Quantum advantage: </strong>the point at which a quantum computer can solve a problem that is infeasible or significantly faster than the best classical alternative.</li><li><strong>QAOA (Quantum Approximate Optimisation Algorithm): </strong>a quantum algorithm designed to solve combinatorial optimisation problems by approximating solutions efficiently on NISQ devices.</li><li><strong>Quantum annealing: </strong>a quantum computing technique that solves optimisation problems by mimicking the physical process of annealing in metallurgy, used prominently by D-Wave.</li><li><strong>Quantum Convolutional Neural Networks (QCNNs or PQCNNs): </strong>quantum equivalents of classical convolutional neural networks, used for tasks like image recognition but designed to leverage quantum properties.</li><li><strong>Quantum error correction (QEC): </strong>techniques used to detect and correct errors in quantum systems by encoding information across multiple qubits, enabling fault-tolerant computation.</li><li><strong>Qubit (Quantum bit): </strong>the fundamental unit of quantum information, analogous to a bit in classical computing, but capable of being in multiple states simultaneously through superposition.</li><li><strong>rQOPS (Reliable Quantum Operations Per Second): </strong>a proposed benchmark for measuring how many dependable quantum operations a system can execute per second, used to assess scalability and utility in the future.</li><li><strong>VQAs (Variational Quantum Algorithms): </strong>hybrid quantum-classical methods where a quantum computer suggests solutions refined by a classical optimiser, suitable for NISQ devices.</li></ul><h4>SOURCES EXPLORED (<a href="https://near-geranium-5c9.notion.site/Quantum-hardware-modalities-research-papers-use-case-focused-1f2cbb5dbca1801abe0ae9bc9a760825">HERE</a>)</h4><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8e79db972dbc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cracking the Climate Code: Why Deeptech Software is the Sweet Spot for VC Returns]]></title>
            <link>https://medium.com/included-vc/cracking-the-climate-code-why-deeptech-software-is-the-sweet-spot-for-vc-returns-0b9c8795271b?source=rss-d85d57a8baaf------2</link>
            <guid isPermaLink="false">https://medium.com/p/0b9c8795271b</guid>
            <category><![CDATA[venture-capital]]></category>
            <category><![CDATA[climate-technology]]></category>
            <category><![CDATA[deep-tech-venture-capital]]></category>
            <category><![CDATA[software]]></category>
            <category><![CDATA[deeptech]]></category>
            <dc:creator><![CDATA[Donald Harmitt]]></dc:creator>
            <pubDate>Fri, 07 Feb 2025 12:20:40 GMT</pubDate>
            <atom:updated>2025-03-20T10:11:59.016Z</atom:updated>
            <content:encoded><![CDATA[<p><em>The escalating climate crisis presents both a profound challenge and a significant opportunity. As the world intensifies its search for sustainable solutions, the convergence of deep technology and climate technology emerges as a fertile arena for transformative advancements. Traditionally, deep tech ventures, particularly those centred on hardware, have deterred venture capitalists (VCs) due to lengthy development timelines and substantial capital demands. However, the anticipated surge in hardware deployments over the next few years presents a compelling case for software solutions that can optimise and enhance these infrastructures. Deep tech software, therefore, stands out as an attractive investment avenue, blending the transformative power of advanced technologies with the scalability and operational efficiency of software, making it an appealing proposition for venture capitalists seeking high-impact yet manageable risk profiles.</em></p><p>📖 <strong>TL;DR</strong></p><ul><li><strong>Deep tech investment has been on the rise</strong>, increasing 16x from 2013 to 2023.</li><li><strong>Climate tech is growing within funded deep technologies</strong> (7% in 2016 to 38% in 2022), as investors increasingly recognise that cutting-edge scientific innovations are essential for driving the climate transition.</li><li><strong>Deep tech software offers an enticing equilibrium for hardware-averse investors</strong>, blending technological disruption with scalable business models that resonate with high-tech venture capital expectations.</li><li><strong>Emerging startups in novel AI, advanced computing, and space technology are unlocking lucrative software opportunities </strong>at the intersection of deep tech and climate tech, promising both impact and investment returns.</li></ul><p>🛠️<strong>Deep tech is predominantly hardware focused, but continues to open up software opportunities for early visionaries</strong></p><p>When you hear the term “deep tech”, what is the first thing that comes to mind? The term deep tech often evokes images of artificial intelligence, quantum computing, or space exploration, yet it also encompasses climate-related innovations such as carbon capture and advanced renewable energy systems. Deep tech is often described as <em>“innovations based on advanced and novel scientific research and engineering”</em>. Typically characterised by long development cycles, deep tech spends an extended period of time in the laboratory phase before reaching commercialisation, that is if it manages to overcome the looming “Valley of Death” faced by many.</p><p>Notably, what is considered deep tech today may not be considered the same in a few years’ time as the technology matures and becomes commonplace, losing its technological “je nais se quois”. Many investors, especially VCs, are more reluctant to invest in hardware-concentrated sectors such as climate and deep tech. 20% of non-deep tech (known as high-tech) VC funding is towards hardware while <a href="https://dealroom.co/guides/deep-tech-europe">the remaining 80% is software</a>, demonstrating a clear affinity towards software across the VC ecosystem. Nevertheless, the projected increase in hardware-derived revenues is expected to catalyse a parallel surge in software investments within the deep tech and climate tech sectors. The proliferation of AI-driven software solutions is not only democratising deep technologies but also accelerating their path to market, presenting investors with scalable business models underpinned by robust technological foundations.</p><p>🕓<strong>Deep technologies come and go in waves</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/902/1*5mdO0UDQ90CFHpt1uPRyKA.png" /><figcaption>Source: <a href="https://lakestar.com/news/the-2023-european-deep-tech-report">European Deep Tech Report</a></figcaption></figure><p>The evolution of deep technologies unfolds in waves, each bringing paradigm-shifting innovations that redefine societal norms. Although the term deep tech is a 21st-century construct, its essence has been evident in historical breakthroughs such as the internal combustion engine (ICE), electricity, and the internet. The ICE automobile changed the way we get from A to B, electricity and the light bulb transformed domestic and industrial processes, and the internet increased globalisation beyond our wildest imaginations. These innovations fundamentally altered transportation, industrial processes, and global connectivity, giving rise to industry titans like Mercedes-Benz and Apple. Now that these technologies are ubiquitous, they are taken for granted and technically are no longer in the deep tech realm. Today, as technologies such as AI, quantum computing, and space tech approach commercial readiness, they signal the advent of another transformative wave within the deep tech ecosystem.</p><p>Decade after decade, we see new mega-waves of advanced technological innovation which significantly disrupts the way we live and work. A great recent example is the effect of AI and its disruption across all types of industries with its countless use cases. AI, quantum, and space tech are a few current deep technologies which have either achieved commercialisation or is close to achieving this feat. This advancement in technological readiness in these areas has been one of the key drivers behind the recent growth in the deep tech sector.</p><p>The current global challenges, ranging from cybersecurity threats to climate change, are spurring a new era of deep tech innovations aimed at providing sustainable and scalable solutions. As the saturated SaaS market diminishes investor returns, there is a noticeable pivot towards hardware-centric deep tech investments, driven by the potential for substantial long-term returns from disruptive technologies addressing critical global issues.</p><p>Within this context, the intersection of deep tech and climate tech presents a particularly compelling investment landscape. The anticipated proliferation of hardware infrastructure, from renewable energy systems to electric vehicle batteries, creates a fertile ground for software solutions that can orchestrate, optimise, and enhance these physical assets. As we delve into this dynamic intersection, it becomes evident that software will serve as the linchpin for maximising the efficacy and profitability of climate tech deployments, offering investors a promising avenue for engaging with deep tech innovations while mitigating inherent risks.</p><p>🚀<strong>Deeptech investment growth has catapulted by more than 3x over 8 years</strong></p><p>Deep tech is a hard sector to ignore, and no doubt investors have been taking interest. Global VC investment in deep tech has been growing rapidly since 2016 (increasing over 3x by 2022), peaking at $20.6 billion in 2021. This influx of capital spans diverse domain across 7 primary deep tech industries:</p><ol><li><strong>Novel AI/ML</strong> — cutting-edge artificial intelligence and machine learning algorithms redefining how systems perceive, predict, and optimise complex problems.</li><li><strong>Quantum technology</strong> — Harnessing quantum mechanics to revolutionise computing, sensing, and communication beyond classical capabilities.</li><li><strong>Biotech</strong> — Breakthroughs in biology and technology to transform healthcare, agriculture, and environmental sustainability.</li><li><strong>Robotics</strong> — Advanced automation and intelligent machines reshaping industries and daily life through precision and autonomy.</li><li><strong>Novel energy</strong> — Innovative energy generation, storage, and efficiency technologies driving sustainable and scalable power solutions.</li><li><strong>Space tech</strong> — Frontier technologies enabling exploration, commercialisation, and colonisation of outer space.</li><li><strong>Advanced materials</strong> — Next-generation materials engineered to deliver unprecedented performance, resilience, and versatility.</li></ol><p>Public funding has played a pivotal role in driving growth in deep tech with European funds such as <a href="https://lakestar.com/news/the-2023-european-deep-tech-report">EIC Fund (over $1.7bn in EU deep tech startups)</a>, and the European Investment fund which invest heavily as a limited partner (LP) in many deep tech funds, acting as early backers in cutting-edge technologies. Unsurprisingly, most VC deep tech investment was concentrated into the novel AI and biotech subsectors in 2023, which were the key reasons for the sector being the single largest VC funding category in Europe in 2023. Would you like to guess which sector was second? If you guessed energy, then you’ve hit the nail on the head. It comes as no surprise that there is significant overlap between climate tech and deep tech. This convergence is characterised by hardware-intensive solutions with long R&amp;D cycles and high technological risks, making software a critical enabler for operational efficiency, performance optimisation, and user engagement at the application layer.</p><blockquote>“In a world where you have a hardware paradigm shift, you then start to say, how can I orchestrate and optimise that hardware? Ultimately, that’s a software problem.” — <strong>Sebastian Hunte, Deep tech investor at Albion VC</strong></blockquote><p>🌍<strong>Deep tech has been playing a pivotal role in the climate phenomenon</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Rw3P9PeQeqb7atJ-EjQx_Q.png" /></figure><p><a href="https://www.imf.org/en/Blogs/Articles/2023/11/27/world-needs-more-policy-ambition-private-funds-and-innovation-to-meet-climate-goals#:~:text=The%20path%20to%20net%20zero,%2C%20credit%20enhancements%2C%20or%20guarantees.">$5 trillion dollars per year</a>. That is the ginormous figure required for low-carbon investments annually by 2030, in order to meet global net zero targets by 2050. In 2023, roughly $1.8 trillion was spent on clean energy, so the gap is somewhat daunting to close. The good news is, investment in the energy transition (including deploying physical assets) has been growing and hitting record highs since 2014, so there is some light at the end of the tunnel. For climate technology, specifically from a VC perspective, annual deal value has been on a rising trajectory since 2017, <a href="https://files.pitchbook.com/website/files/pdf/Q1_2024_PitchBook_Analyst_Note_VC_Investment_in_Climate_Tech_Preview.pdf">peaking at the highs of 2022</a> ($51 billion).</p><p>Deep tech is a sector that has been crucial in tackling the climate challenge so far, with new scientific and engineering breakthroughs within areas from smart grid solutions to new material development. Likewise, the deep tech and climate tech intersection (Deep Tech x Climate Tech) has been on the rise, with climate tech growing within VC funded deep technologies, jumping from only <a href="https://dealroom.co/guides/deep-tech-europe">7% in 2016 and rising to 38% in 2022</a> ($7 billion funding) — more than a 5x increase. In the UK for instance, the most popular subsector within deep tech is cleantech with 517 deep tech startups, and deep tech equity investments has soared from <a href="https://tech.eu/2024/01/05/the-uk-is-home-to-nearly-3500-active-deep-tech-companies/">£174 million to £5 billion</a> over the last 10 years, reflection growing investor confidence in this intersection.</p><p>Despite this promising outlook, many investors are <a href="https://www.wartsila.com/insights/article/deep-tech-may-just-save-the-planet">averse to investing in climate and deep tech</a> due to perceived returns and scaling potential not being as immediate as software-dominant sectors. R&amp;D and product development is a long play in these sectors, and not to mention, they are highly capital intensive. When it comes to minimising climate change, software alone is not going to cut it, especially when you think about high impact potential technologies such as carbon capture, batteries, and renewable energy. The good news for software investors is, the shift towards a new hardware paradigm unlocks a plethora of software white space to enhance the hardware deployments.</p><blockquote>“Every hardware opportunity leads to many more software opportunities that you can build because of it.” — Sebastian Hunte, Deep tech investor at Albion VC</blockquote><p>⚔️<strong>Deep and climate tech startups face interesting challenges, but the favourable upside might just outweigh them</strong></p><p>Northvolt, a European green energy battery developer, poised to be the poster child of climate tech by taking on Chinese battery producers, recently went bankrupt. This will undoubtedly cast a wave of doubt over climate and deep tech companies, especially hardware startups, as $13 billion had already been pumped into Northvolt at the time of bankruptcy. The Valley of Death scenario is certainly not uncommon, and disproportionately affects startups with long R&amp;D timelines, which happen to be primarily hardware companies. It’s no secret that climate and deep tech are plagued by a few challenges on the road to commercialisation:</p><p>💸<strong>Capital intensive</strong> — due to the capital expenditure required to build large-scale hardware, many startups in these sectors burn through a lot of cash during R&amp;D, and even into commercialisation to scale before achieving revenue generation.</p><p>🔬<strong>Long R&amp;D timelines</strong> — the advanced technological innovations involved in the underlying technology leads to extensive research and development required in identifying and testing the ideal product/solution to take to market.</p><p>💻<strong>Higher technology risk</strong> — technologies within these sectors are novel and typically are yet to be proven on a commercial stage, which could see years of investment not yield any return in the end.</p><p>👥<strong>Talent</strong> — the strong domain expertise required in developing advanced scientific solutions is relatively scarce compared to high-tech sectors, making scouting talent and hiring a challenging process.</p><p>🚧<strong>Regulatory hurdle</strong> — governmental policies are a key driver of climate and scientific-related innovation, but can also act as a headwind when there is a shift to unfavourable legislations or lack of incentives.</p><p><strong>However, there is a reason why deep tech is one of the most funded sectors:</strong></p><p>🤼<strong>Low competitive risk</strong> — many companies within these sectors have a strong technology moat and are difficult to replicate, creating a high barrier to entry and replication.</p><p>🚨<strong>Relatively lower market risk</strong> — problems being solved with these technologies are typically on a global scale and have the potential to disrupt sectors, technologies, and/or business models, hence, market demand is high for these startups</p><p>📈<strong>Typically higher valuation potential in the long run </strong>— despite the slower ramp up in the earlier stages of deep tech startups compared to high-tech startups, deep tech companies tend to have higher valuation potential with deep tech focused funds producing <a href="https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/european-deep-tech-what-investors-and-corporations-need-to-know">7% higher percentage points net IRR</a> than traditional tech funds.</p><p>📡<strong>Software could be the silver bullet for averse deep tech investors</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LN3vcwo44L75kQqR_tfGMg.png" /></figure><blockquote><em>Over the next five years, 60% of revenue in “technology” will come from hardware, with only 40% coming from software —</em><a href="https://dealroom.co/guides/deep-tech-europe"><em> Dealroom</em></a></blockquote><p>The deep tech sector sees 60% of VC funding going towards hardware, compared to the rest of tech which sees only 20% of investment within this modality. Even though hardware is predicted to be the main technology revenue driver over the next few years (generating 60% of technology revenue compared to 40% from software), there are two questions here to be explored.</p><ol><li><em>What are the software opportunities that the 40% of deep tech investors are investing in?</em></li><li><em>What do these opportunities look like when we narrow the focus towards the intersection of climate and deep tech, in a bid to accelerate climate action?</em></li></ol><p>In a space where hardware dominates, software emerges as a strategic lever for investors seeking exposure to deep tech and climate tech without the associated capital burdens. “Software is eating the world”, the famous quote by Marc Andressen at a16z VC firm, highlighted the technological shift from a hardware-focused economy to a software-enabled economy back in the early 2010s. We have seen how software has disrupted industries and transformed global economies by “eating” traditional business models and players, creating outsized returns for investors.</p><p>With equity stakes at some point <a href="https://eqvista.com/top-investments-andreessen-horowitz/">by a16z in Facebook, Twitter, Stripe, Airbnb, and many more</a>, it is no wonder that they are big advocates for software focused business models. The wider VC ecosystem has followed suit, largely due to the scalable business models compared to hardware. As hardware makes its way back into fashion, it will be good to explore where deep tech software opportunities lie for hardware-averse investors to engage with transformative technologies, while mitigating common deep tech investment risks, alongside tackling climate change.</p><p>🛰️<strong>Novel AI, Future of Compute, and Space Tech startups are already making waves in climate tech</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-dtGGlKGTGKrj6EB1lBp9w.png" /></figure><p>There are 7 main deep tech sectors, however, in this deep dive we will be focusing on 3 of these sectors when looking at the software opportunities at the convergence of climate and deep tech: <strong>Future of Compute</strong>, <strong>novel AI</strong>, and <strong>space tech</strong>.</p><p><strong>1. Novel AI</strong></p><p><strong><em>Smarter renewable energy deployment and grid integration</em></strong></p><p>The rise in electrification is putting a strain on old, legacy grids used today. By 2030, <a href="https://www.iea.org/news/massive-global-growth-of-renewables-to-2030-is-set-to-match-entire-power-capacity-of-major-economies-today-moving-world-closer-to-tripling-goal">renewables is expected to meet half of the global electricity demand, and wind, solar, and PV generation set to double to 30%</a>. This is an incredible amount of hardware being installed around the world to meet the transition to clean energy. However, there are an array of problems with this. Firstly, renewable energy solutions are intermittent, relying heavily on external conditions such as weather and temperature. This is interlinked with a second challenge — integrating renewables with the grid to meticulously change in line with demand and supply. The intermittent nature of distributed energy resources means there are instances where enough power is not available to meet supply, or too much demand above the infrastructure capabilities lead to what is known as grid congestion. There is an imminent need for solutions which enable the grid to allocate power where and when it is needed, based on rapidly fluctuating demands.</p><p>Decentralised AI has the potential to play a critical role in this use case through making the grid more “intelligent” and flexible, creating smart grids. This means accurately forecasting demand fluctuations based on historical data to seamlessly balance the interchange between generation and distribution. AI plays several roles, including energy flow optimisation, demand forecasting, demand energy response coordination and integration, and many other use cases. The change in demand and supply has opened the door for <a href="https://www.exro.com/industry-insights/peak-shaving">peak shaving</a> and <a href="https://energy.drax.com/insights/what-is-load-balancing/">load balancing</a> mechanisms acting as operational cost savers, and capitalising on energy tariffs by shifting electricity usage to off-peak times.</p><p>Startups have already identified key problems in this space and are steadily emerging to capitalise with innovative solutions. Early players are building software capabilities around energy &amp; grid management, and energy trading, to integrate with the deployment of renewable hardware and energy storage applications.</p><p>Ogre AI, is an energy tech startup focused on enhancing grid efficiency, by providing AI-based forecasting solutions. Other notable trends include startups combining AI with other technologies such as internet of things (IoT), and blockchain, to differentiate themselves in the space, specifically within grid management and energy trading.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Io2muKA7aaGFR8OWcZt_xQ.png" /></figure><p><strong><em>Renewable energy predictive maintenance and monitoring software</em></strong></p><p>The lifetime of a wind turbine is 20–25 years and the lifetime of a solar panel is 30–35 years dependent on material type. However, this can be extended if sufficient predictive maintenance is used to prevent equipment failure, which happen to be the main cause of wind turbine downtime. AI can be leveraged to apply machine learning and data analytics, to not only reduce operational and maintenance costs across the asset lifetime, but also proactively increase the reliability and efficiency of distributed energy systems.</p><p>ANNEA is a startup which automates predictive maintenance and enhances energy production through software (digital twin). The analysis of vibrations and acoustics can be used on wind turbines to predict upcoming failures before they happen and detect cracks and physical defects, respectively. Another use case is within battery energy storage systems (BESS) to forecast the remaining useful life and identify defective cells through the analysis of charge and discharge cycles.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x7Fv2-JfOIoConeEa1EQmA.png" /></figure><p><strong><em>Battery recycling</em></strong></p><p>With the imminent growth in EV production and battery energy storage deployment, we are seeing a sharp increase in battery production. Batteries do not last forever, and their disposal is not as straightforward as other items due to their pollutant potential, which lead to environmental challenges. Majority of batteries are recycled, and valuable rare earth materials are often recovered from end-of-life batteries to be reused in the production of new batteries. Additionally, crucial minerals for battery production such as lithium, cobalt, and nickel are finite, and battery recycling helps to slow down the depletion of these resources through enabling this circular economy. The global demand for lithium is forecasted to <a href="https://techcrunch.com/2024/11/11/as-demand-for-lithium-explodes-battery-recycling-startup-tozero-sprints-to-scale-with-11-7m-seed/">increase by 4x by 2030</a> (3 million metric tonnes), highlighting the need for recycling as lithium production in 2023 yielded less than 6% of that quantity.</p><p>With the rise in geopolitical tensions, Western countries are increasing onshoring capabilities to reduce supply chain vulnerabilities. In order for regions such as Europe to be able to compete with China in the battery production market, significant efficiency gains are required to compete on a cost basis. This is both within the production of batteries and the recycling of batteries to extract recyclable materials.</p><p>There are two main opportunities for novel AI within the battery recycling sector to increase efficiency and reduce cost. Firstly, during the <a href="https://cicenergigune.com/en/blog/opportunities-offered-artificial-intelligence-battery-recycling">sorting of waste, and secondly, during the recovery of critical materials</a>. AI’s pattern recognition and categorisation capabilities leverage neural networks to group waste for similar recovery routes. During the recovery phase, AI can automate the process to increase efficiency by analysing the degradation levels of the waste and suggesting the appropriate action, minimising human intervention. There are several early-stage companies in this space, not only focused on recycling, but also upcycling (repurposing waste materials into other products) e.g. Circu Li-ion.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ANpAoQXwE0foOy3fL6ViLA.png" /></figure><p><strong>2. Future of Compute</strong></p><p><strong><em>Quantum cryptography in energy grids — securing decentralised renewable energy systems</em></strong></p><p>Quantum technologies are making promising progress in their advancements. Quantum technologies has significant potential to disrupt a wide array of industries due to the 100s of use case applications, once the computers are powerful enough to carry out more complex calculations error-free. One of these use cases is in line with energy grids, as a form of security to shield against cybersecurity attacks. With electrical grids becoming “smarter” and an increasing volume of data flowing between utility companies and end customers, this opens up opportunities for serious adversarial attacks (fibre tapping) on the grid network. Not to mention, many of the existing grid <a href="https://www.forbes.com/councils/forbestechcouncil/2023/08/11/is-quantum-key-distribution-the-key-to-power-grid-modernization/">is decades old</a>, making it more susceptible to these kinds of attacks. For example, bad actors could directly disrupt grid communications leading to a change in behaviour of distributed energy resources or even causing power outages.</p><p>The need for a secure grid offers opportunities for quantum cryptography algorithms to enable a “quantum-resistant” grid, which makes the grid resistant to future advanced quantum computers. Companies are developing quantum key distribution (QKD) devices with algorithms which can be deployed between utilities and end customers, by using a photon light though a fibre optic cable to generate a random key known only to the stakeholders. There are startups already in the commercialisation phase, making QKD one of the most mature areas of quantum technologies today. Startups can innovate on the middleware and software layer to optimise the devices, enabling larger algorithms to be ran on less power and bandwidth. For instance, Phasecraft is a quantum software startup applying quantum algorithms to the discovery of new materials, focused on the energy transition.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tIfOdgLI1ABKHZ44dmclHQ.png" /></figure><p><strong><em>Quantum simulations for material discovery</em></strong></p><p>Quantum computing is deemed as the next frontier, following the immense wave of AI. Quantum computing has a vast array of future promises which can be overwhelming to dissect. There are 4 main use cases for quantum computing: simulation, optimisation, machine learning, and cryptography (which we touched on earlier regarding energy grids).</p><p>Simulation is one of the highly promising use cases and is deemed to have strong potential for commercialisation in the nearer term, starting with simulating the properties of sections of complex molecules. As natural processes are typically quantum in nature, they lend themselves readily to the capabilities of quantum computers to mimic or simulate molecular interactions and structures on the atomic level. Additionally, simulation use cases offers a number of synergies between quantum computing and AI. Supercomputers and AI have made significant strides in the simulation of complex molecules, however, quantum adds an extra layer of precision at the subatomic level, ultimately speeding up the discovery process for end users. This removes the need for physical experiments in the lab, offering precise “virtual” simulations.</p><p>The good news is smaller-scale quantum computers can be used for simulations especially in laboratory environments within pharma and other scientific industries. Startups such as Quantistry are leveraging AI and quantum simulations to achieve quantum-enhanced drug discovery and design.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QOm3xE3L5bhaZdWq9ydgsA.png" /></figure><p><strong>3. Space Tech</strong></p><p><strong><em>Satellite monitoring and data intelligence targeting climate and energy related sensors</em></strong></p><p>Space technology might not be the first thing that comes to mind when you think of climate technology. However, the juncture between space tech and climate tech has been growing as a number of startups are finding lucrative opportunities as the convergence matures. This has largely been driven by an increasing interest in space travel and decreasing costs to launch a satellite and rocket into space, driving the growth of <a href="https://ellipsis-drive.com/blog/satellite-as-a-service/">satellite-as-a-service</a>. Thanks to startups like Space X, the <a href="https://www.bvp.com/atlas/state-of-deep-tech#Introducing-The-XB100">costs to launch satellites decreased</a> by 10x from $50K/kg to just $5K/kg, enabling more satellite launches and driving the software market with it. Coupled with the pressing need for climate action, this has unlocked new markets for space tech entrepreneurs.</p><p>Within space tech, startups are exploring climate-related software solutions through leveraging satellite data, to provide services such as advanced weather monitoring, emission monitoring, natural disaster monitoring, and solar power output forecasting. Companies in this space can secure large long-term government contracts, as this data will be largely useful to policy makers in relation to the growing ESG mandates. Other entities such as financial institutions offering green finance, commodity &amp; carbon traders, insurance enterprises, and energy companies, will all be target customers for these spacetech-based climate solutions. This highlights the budding market opportunity here, as the environmental landscape continues to be under threat as we have seen with increasing wildfires and floodings.</p><p>With the growth in satellite-as-a-service, this has reduced the barrier to entry for launching new satellites into space. The more satellites launched into the space, the more data can be collected, which drives the software market with vertical applications for end-users. Kayrros is a startup focused on Earth observation from aggregating satellite data to generate insights for companies e.g. insurance entities.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*flV8tHFR0JRkYZCroHK-WQ.png" /></figure><p>🗝️️<strong>The hardware paradigm shift should be welcomed with open arms by software investors</strong></p><p>As hardware reclaims its prominence, software remains an indispensable pillar for investors, offering scalable models and proven returns in the modern economy. The deep tech domain is increasingly intertwined with climate technology, driven by the recognition that advanced scientific breakthroughs are essential to meet ambitious climate targets. With anticipated growth in hardware across renewables, carbon capture, and other climate solutions, deep tech software stands poised as a compelling investment frontier. The expanding landscape of hardware deployment creates fertile ground for complementary software innovations, opening new avenues for forward-thinking startups.</p><p>For investors venturing into deep tech, success hinges on cultivating expertise, fostering patience, and forming strategic alliances with academic institutions and industry leaders. The accelerating convergence of deep tech and climate tech presents unparalleled opportunities, particularly in areas like novel AI, advanced computing, and space technologies. Emerging startups are already reshaping critical domains such as grid management, battery recycling, and satellite communications. This is a transformative moment in the pursuit of net zero, and investors would do well to craft thoughtful strategies now to harness the immense potential of deep tech software. The intersection of deep tech and climate innovation is evolving rapidly, and it would be wise for investors to keep an eye on this space, ready to seize the opportunities that lie ahead.</p><p><em>Thank you for making it the end! I would like to thank Nikita, the Included VC team, and my fellow cohort members, for their inspiration in writing this piece. Additionally, I would like to give a massive thanks to Albion VC deep tech investor, Sebastian Hunte, for his incredible knowledge and insight which contributed to this deep dive.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0b9c8795271b" width="1" height="1" alt=""><hr><p><a href="https://medium.com/included-vc/cracking-the-climate-code-why-deeptech-software-is-the-sweet-spot-for-vc-returns-0b9c8795271b">Cracking the Climate Code: Why Deeptech Software is the Sweet Spot for VC Returns</a> was originally published in <a href="https://medium.com/included-vc">Included VC</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cracking AI’s Competitive Code: What It Takes to Win at the Application Layer]]></title>
            <link>https://medium.com/@donaldharmitt/cracking-ais-competitive-code-what-it-takes-to-win-at-the-application-layer-834dab144775?source=rss-d85d57a8baaf------2</link>
            <guid isPermaLink="false">https://medium.com/p/834dab144775</guid>
            <category><![CDATA[deeptech]]></category>
            <category><![CDATA[business-strategy]]></category>
            <category><![CDATA[aritficial-intelligence]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[venture-capital]]></category>
            <dc:creator><![CDATA[Donald Harmitt]]></dc:creator>
            <pubDate>Fri, 31 Jan 2025 12:21:12 GMT</pubDate>
            <atom:updated>2025-02-01T15:13:09.357Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*aI6XtvfrRY4icLYf" /><figcaption>Image source: <a href="https://alpenglo.digital/digital-marketings-ai-defense-strategy/">Alpenglo Digital</a></figcaption></figure><h4><strong>Setting the scene</strong></h4><p>There has been a lot of focus on AI’s foundational model with incumbents taking up the lion share of this market segment and flexing their muscles to create the winning large language model (LLM). Training these LLMs come with a significant dent in one’s pocket, as the models you know and love such as Google Gemini and Open AI’s GPT-4 cost almost <a href="https://fortune.com/2024/04/18/google-gemini-cost-191-million-to-train-stanford-university-report-estimates/">$200 million</a> and <a href="https://www.techradar.com/pro/openai-spent-usd80m-to-usd100m-training-gpt-4-chinese-firm-claims-it-trained-its-rival-ai-model-for-usd3-million-using-just-2-000-gpus">$100 million</a> to train, respectively. Startups lack the capital and data access to compete on this scale to outcompete incumbents. However, we have seen startups start to shift downstream to where they know best…the application layer.</p><p>Not only that, but open-source models are dampening the competitive advantage underpinning the foundational layer, making the market potential less attractive for incumbent and emerging players. Companies competing at the application layer will emerge from the AI race as the real winners in the long run, as enterprises will spend the big bucks to integrate AI tools to optimise their cost basis and decision making, while driving growth in their top line. However, there is one key challenge startups need to consider in order to make it to the finish line. The key question: <strong>How can AI startups at the application layer achieve defensibility? </strong>I have several thoughts on this that I will dive into later, but first let us dive into the AI tech stack.</p><h4>📉<strong>Incumbents dominate the foundational layer in a race to the bottom</strong></h4><p>A few years ago, AI’s battleground was at the foundational layer. Big names like OpenAI, Anthropic, and Google DeepMind were in an arms race, pouring billions into training ever-larger language models. The hope? That whoever controlled the biggest, most powerful models would capture the lion’s share of value. And for a while, that logic held up. But fast forward to today, and things are looking a little different. Foundational models are quickly becoming commodities, APIs have cracked open access to cutting-edge AI, and suddenly, the real action is happening somewhere else: the application layer.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6-ZwJ2EEQF8LhqFeUzEljg.png" /></figure><p>Before we can identify opportunities and challenges within the AI landscape, we need to understand the AI tech stack. The AI tech stack is a multi-layered architecture that facilitates the development, deployment, and operation of AI systems. There are traditionally<strong> four broad layers</strong> across the AI stack:</p><p><strong>1.</strong> <strong>Infrastructure layer</strong>: Foundational layer comprising of hardware and computing resources, including GPUs, CPUs, and specialised accelerators, providing the computing resources necessary for training and running AI models.</p><p><strong>2.</strong> <strong>Data layer</strong>: This layer involves the collection, storage, and management of data.</p><p><strong>3.</strong> <strong>Model layer</strong>: Focused on the development and training of AI models using ML frameworks and algorithms.</p><p><strong>4.</strong> <strong>Application layer</strong>: The application layer where AI interfaces with end-users e.g., web applications, APIs, and user interfaces that deliver AI functionalities into practical applications that address specific user needs.</p><p>Incumbents dominate the infrastructure layer primarily due to the astronomical capital expenditure needed to build the hardware and compute power, alongside requiring access to large masses of data. With Open AI raising $6.6 billion, going primarily towards increasing compute capacity, the billions of dollars spent training and running these models will become unsustainable without efficiency improvements. These players risk being undercut on a cost basis by rivals brave enough to compete. Sounds familiar? In the past week, everyone and their dog has been talking about the DeepSeek saga. Just in case you have been living under a rock, I will briefly explain what has been going on.</p><h4>⛩️<strong>China’s efficiency strikes again</strong></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2AANLoco2l7Qbi4gllKmqw.jpeg" /><figcaption>Image adapted from <a href="https://www.tensorops.ai/post/breaking-down-the-cost-of-ai-for-organizations">TensorOps</a></figcaption></figure><p>China’s DeepSeek has ruffled some feathers in the LLM ecosystem by unveiling it’s cheaper (only $6 million to train according to DeepSeek), open source LLM (R1) which is said to have on par performance when it comes to niche tasks including data retrieval and search accuracy, compared to GPT-4o. If there was one signal to take away from the DeepSeek saga, it’s that LLMs are swiftly losing its unique differentiator and heading towards being commoditised. The true value will be gained from the opposite end of the spectrum at the application layer, where business-specific solutions solving genuine painpoints will be an AI company’s USP. The AI revolution is entering its next phase, and it’s not just about bigger models or more powerful chips, it’s about making AI cheaper, more accessible, and more tailored.</p><p><strong>What is Jevons Paradox and why does it matter for AI?</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*BOgtITcJo9Bk7vhh.jpeg" /><figcaption>Image source: <a href="https://www.hottakes.space/p/jevons-paradox-and-a-world-of-spiraling">Jevon’s Paradox and a world of spiraling complexity</a></figcaption></figure><p>Jevons Paradox, an economic principle dating back to the 19th century, states that when efficiency improves, overall consumption increases rather than decreases. We’ve seen this in countless industries:</p><ul><li><strong>More fuel-efficient cars?</strong> People drive more.</li><li><strong>Cheaper flights?</strong> Travel surges.</li><li><strong>Cheaper AI compute?</strong> AI-powered applications explode.</li></ul><p>This is exactly what’s about to happen in AI. As models become more efficient and computing costs drop, the real winners will be the AI application layer i.e. the startups, enterprises, and developers that will now be able to integrate AI at scale.</p><p>🧠<strong>Why this is a game-changer for AI applications?</strong></p><p>For years, AI innovation has been bottlenecked by cost. Training large-scale AI models has required billions in investment, restricting AI breakthroughs to a handful of elite companies. But with cost barriers falling, we’re about to see a massive expansion in AI applications, where businesses can experiment, deploy, and refine AI at unprecedented scale.</p><p>Here’s why this shift will be transformative:</p><p>● <strong>AI moves from being exclusive to ubiquitous</strong></p><p>High costs have meant that only the biggest players, such as OpenAI, Google DeepMind, and Anthropic, could afford to train cutting-edge models. But as AI efficiency improves, startups and enterprises will have access to state-of-the-art AI capabilities without massive upfront investment.</p><p>● <strong>Explosion of industry-specific AI applications</strong></p><p>AI in enterprise software will move from simple chatbots to fully AI-driven workflows as we seeing with the rise of agentic AI. Verticalisation of AI will continue to increase as startups will build their defensibility through domain specialisation. For instance, AI in healthcare will continue to be embedded into daily patient care as health database becomes less unstructured — amid the regulatory challenges of course, while in finance it will automate decision-making at an entirely new level.</p><p>● <strong>Lower costs mean more experimentation</strong></p><p>When AI costs thousands per query, businesses limit their usage. When it’s pennies per query, AI can be embedded into every customer interaction<strong>, </strong>every decision, and every workflow.</p><p>● <strong>AI becomes an infrastructure layer, not a luxury</strong></p><p>Just as cloud computing transitioned from an expensive innovation to an everyday business necessity, AI is now heading in the same direction. The shift from scarcity to abundance means that soon AI won’t be a feature, it’ll be a fundamental layer powering every industry. As “AI eats software”, this will come with a shift in business model and how businesses approach software. The way software is built, utilised, and integrated will divert away from the typical SaaS model and will require significantly less human intervention.</p><h4>🔐 <strong>Building defensibility in AI applications: what actually works?</strong></h4><p>So where does this leave startups? The answer is clear. Startups will need to build defensible, application-layer AI solutions that go beyond just wrapping an LLM in a sleek interface. The next generation of AI winners will be those who carve out moats (used to describe a company’s competitive advantage), embedding themselves so deeply into workflows that switching becomes near impossible. The challenge, however, is that defensibility in AI applications is not a given. Without clear differentiation, companies risk getting lost in a sea of competitors who are leveraging the same models and similar data pipelines.</p><p>Defensibility is a tricky puzzle. In traditional software, moats often come from network effects, regulatory advantages, or sheer market dominance. But AI applications? A different ballgame entirely. Anyone with access to an LLM can spin up a competing product overnight. I believe defensibility needs to be built on <strong>five key pillars</strong>:</p><p>🔗 <strong>1.</strong> <strong>Workflow integration</strong> — The deeper an AI tool is embedded in a company’s workflow, the higher the switching costs. The most successful applications will not just automate tasks but will become an irreplaceable part of operations. When an AI tool is deeply ingrained into a process, replacing it is not just inconvenient, it is disruptive and costly.</p><p>📊 <strong>2. Proprietary data</strong> — AI is only as powerful as the data behind it. Startups sitting on unique, high-quality, domain-specific datasets (especially first-party or user-generated data) will hold a significant edge. Access to data = sustained advantage. Over time, models trained on proprietary data improve exponentially, creating a flywheel effect that locks out competitors. The best AI companies are not simply API consumers; they are innovators who refine and optimise models for hyper-specific use cases.</p><p>🛡 <strong>3.</strong> <strong>Brand and customer trust</strong> — In a world where AI models are interchangeable, customer relationships are gold. AI products that establish themselves as trusted providers in niche industries will gain lasting loyalty. Trust is not just about reliability; it is about demonstrating AI’s ability to make decisions in an ethical, transparent, and explainable way.</p><p>🎨 <strong>4.</strong> <strong>User experience and product design</strong> — AI should feel invisible. The best applications make the technology seamless, enhancing productivity without requiring users to fiddle with models or parameters. A delightful, intuitive experience creates a natural moat. Even the most powerful AI is useless if users find it difficult to integrate into their daily workflows.</p><p>👥 <strong>5.</strong> <strong>Team and talent</strong> — The strength of a startup’s team is a critical differentiator in AI defensibility. A highly skilled team with deep expertise in AI research, software engineering, and domain-specific knowledge can drive innovation, iterate rapidly, and create a product that outpaces the competition. Beyond technical ability, a strong team brings strategic vision, adaptability, and execution power, ensuring the company can pivot and refine its approach as the AI landscape evolves. Investors and customers alike will gravitate towards teams that demonstrate an ability to solve complex challenges and push technological boundaries.</p><h4>💡 <strong>In the big AI application play: where are the opportunities?</strong></h4><p>We are already seeing two dominant categories emerge in the AI application space:</p><p><strong>1.</strong> <strong>Vertical AI solutions</strong> — Think AI-built specifically for industries like healthcare, law, or finance. AI-powered contract review tools, AI-driven diagnostics for doctors, or risk assessment models for insurers. These leverage domain-specific expertise and datasets, making them harder to copy. Additionally, industry regulations and compliance create further barriers to entry, making domain expertise a competitive advantage.</p><p><strong>2. Workflow automation &amp; productivity tools</strong> — These are applications designed to slot seamlessly into day-to-day business operations. Think AI-enhanced writing assistants, automated customer support solutions, and AI-driven financial forecasting tools. Startups in this space are tackling efficiency at scale. The demand for AI-enhanced tools that boost productivity is soaring, as businesses seek ways to cut operational costs and streamline decision-making.</p><p><strong>Lessons from cloud &amp; SaaS: why AI startups need more than just AI</strong></p><p>Cloud computing giants like AWS and Azure built their dominance on high switching costs. AI applications, on the other hand, do not have that luxury. Instead, they need to continuously evolve, adapting as foundational models improve and as customer needs shift. The biggest SaaS winners over the last two decades have shown that defensibility does not come from the tech alone, it comes from embedding deeply into workflows, providing undeniable value, and owning customer relationships. This is why AI startups that do not merely focus on the AI component but instead double down on user experience and ecosystem integration stand to gain the most.</p><p>The harsh truth? Many AI startups today are just thin wrappers on top of foundational models. And that is not a sustainable position. To build lasting value (outside of the proprietary data play), AI companies need to:</p><p>✅ <strong>Continuously improve &amp; adapt</strong> — AI is evolving at breakneck speed. The startups that thrive will be the ones that do not just launch a product and call it a day, but keep iterating, improving, and expanding their moat. Staying ahead requires continuous retraining, model refinement, and a deep understanding of shifting market needs.</p><p>✅ <strong>Nail the Go-To-Market strategy</strong> — Distribution is half the battle. Companies that crack efficient customer acquisition, whether through partnerships, bottom-up adoption, or viral growth, will win out over those relying purely on the strength of their tech. The best AI product is useless without adoption, and go-to-market strategies must be tailored to each segment.</p><h4>🌍 <strong>The future of AI applications: what comes next?</strong></h4><p>While foundational models are consolidating into the hands of a few major players, the application layer is wide open. The startups that win will not just be the ones with the best AI, they will be the ones that solve real-world business problems<strong> </strong>seamlessly, efficiently, and in ways that are difficult to replicate. The future of AI is not just about intelligence; it is about utility. It is about how well an AI product integrates into the lives and workflows of users. And as AI continues to become more embedded in every aspect of work and life, companies that master this balance will be the ones that define the next era of technological disruption.</p><p>We are not just watching an evolution of AI. We are witnessing the next great shift in how technology creates value. The real question is: <strong>what will true defensibility look like at the AI application layer in 5 years? </strong>Only time will tell.</p><p>Thank you reading to the end! If you would like to discuss further feel free to reach out on <a href="https://www.linkedin.com/in/donald-harmitt-65199b170/">LinkedIn</a> and I would be happy to have to a chat.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=834dab144775" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Revolutionary Tech or Risky Bet? The Untold Truth About Quantum Computing’s Potential]]></title>
            <link>https://medium.com/included-vc/from-vision-to-value-why-quantum-computing-belongs-in-your-portfolio-today-bf46b5fa0795?source=rss-d85d57a8baaf------2</link>
            <guid isPermaLink="false">https://medium.com/p/bf46b5fa0795</guid>
            <category><![CDATA[deep-tech-investor]]></category>
            <category><![CDATA[deep-tech-startups]]></category>
            <category><![CDATA[quantum-computing]]></category>
            <category><![CDATA[quantum-computer]]></category>
            <category><![CDATA[deep-technology]]></category>
            <dc:creator><![CDATA[Donald Harmitt]]></dc:creator>
            <pubDate>Sat, 14 Dec 2024 21:17:32 GMT</pubDate>
            <atom:updated>2025-02-24T08:50:12.279Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NZTYN04q0WMLmYISWcN6fA.png" /><figcaption>Quantum computing market map — the list is non-exhaustive — feel free to reach out on <a href="https://www.linkedin.com/in/donald-harmitt-65199b170/">LinkedIn</a> to add your company</figcaption></figure><p><em>It all started innocently enough a while back — I was just browsing articles about emerging tech when I stumbled across quantum computing. I figured, “sure, I’ll give it a quick read.” Fast-forward a few hours, and I’d completely disappeared down the rabbit hole. Suddenly, my mind was buzzing with talk of qubits, entanglement, and algorithms that sounded straight out of one of Sheldon’s Big Bang Theory rants. What started as casual reading turned into late nights trying to wrap my head around quantum mechanics and here I am still fascinated! It’s not just the mind-bending science that drew me in though, it’s the potential impact. It has the potential to reshape industries, solve massive challenges, and even redefine what’s possible in fields like climate technology, pharmaceuticals, materials, and logistics. Quantum computing is still in its early days, but as I see it, this is the space investors should be paying attention to now, before the future catches them off guard.</em></p><h4>📖<strong><em>TL;DR</em></strong></h4><ul><li><strong>Quantum computing technology is advancing at an incredibly fast pace,</strong> with multiple milestones being achieved in error-correction (major challenge) in 2024.</li><li><strong>$450–850 billion annual value creation </strong>is expected between 2030–40 from quantum computing, driven by the 100s of potential use cases across industries such as finance, pharmaceuticals, AI, and climate.</li><li><strong>The investment landscape has been on the rise</strong> (both public and private), with Europe backing quantum computing to new highs year after year.</li><li><strong>Quantum-as-a-Service (QaaS) and vertical quantum computing are early market opportunities</strong> for investors who need returns today rather than tomorrow.</li></ul><p><strong>At a glance, here’s what the deep dive will be covering:</strong></p><ol><li>Quantum computing technology overview</li><li>Economic value (current and future), market size, and use cases</li><li>Investment landscape</li><li>Near-term opportunities for investors</li></ol><h3>🚀<strong>Let me introduce you to the silent revolution poised to transform industries and unlock trillions</strong></h3><p>Imagine a future where every new drug that reaches clinical trials is a hit, newly discovered materials negate the need for virgin resources and plastic waste, and a ‘crystal ball’ that can predict an impending financial crash long before it occurs. This is not science fiction — it’s the promise of quantum computing. Quantum computing has the potential to unlock significant economic value (billions to trillions of dollars), with far-reaching effects across a multitude of sectors, akin to what we are seeing with AI. With its potential to revolutionise industries it is almost mind-boggling as to how subdued the hype around the technology is.</p><p>On the other hand, this leaves ample opportunity on the table for those who can drown out the AI noise and focus their attention where the next frontier awaits. This is not a new phenomenon, the idea behind quantum computing has been around since it was proposed by Richard Feynman in the 1980s. The science behind quantum computers, quantum physics, is already leveraged in devices such as semiconductors, and MRI brain scanners.</p><p>Quantum technology goes beyond computing and includes quantum sensors and quantum communication. However, today I am giving quantum computing the limelight to impress the investors, i.e. you, with its potential and show you why the technology is ready for capital injection. Stay with us on this ride and enjoy the GIFs!</p><h3>🚨<strong>We’ve got a problem — AI is increasing computational demand well beyond supply</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/824/1*P7jc6MnJP0u5oIX_oOFlBw.png" /><figcaption>Source: <a href="https://docsend.com/view/pkmemb28nmvhug4e">The 2023 European Deep Tech Report</a></figcaption></figure><p>The dilemma is real. AI models are getting bigger and bigger, but computational speed progress is lagging. Data is rapidly increasing. With the AI avalanche and acceleration in GPU computation (thanks to NVIDIA), which has bestowed itself upon us, there is an even further increasing need for faster and more powerful computing infrastructures to deal with this data. Still skeptical?</p><blockquote>According to Dealroom, <strong>computing demand is doubling every 2–3 months while computational speed is doubling only every 2 years</strong>.</blockquote><p>The gap will continue to widen unless a technological innovation can increase computational speed back onto Moore’s law trajectory.</p><p>Since the 1960’s, the number of components on a microchip is observed to double every 2 years while it’s cost halves (Moore’s law). We, as humans, have enjoyed this luxury over the late 20th century and early 21st century. Our electronic devices, e.g. our phone, went from the size of a baguette to fitting in the palm of our hands while increasing in power. However, all good things come to an end. Chip manufacturers are wary that we may be approaching the physical limits of making electronic devices smaller, while increasing efficiency and power. But not to worry, our physicists and engineers always find a way. Quantum computing to the rescue?</p><h3>📈<strong>Quantum computing technology is maturing rapidly behind the scenes</strong></h3><p>Quantum computing is poised to solve problems above and beyond the capabilities of our classical computers, including high-performance computers. This is due to its differentiated physical processing capabilities and reduced computational timeframe, underpinned by quantum physics. In a nutshell, our classical computers use a binary system of silicon pieces called “bits” which can take up 2 different states (1 and 0). However, quantum computers use “quantum bits” (or “qubits” for short) which also can, in theory, be in a state of both 1 and 0 simultaneously, using a quantum effect called <strong><em>superposition, </em></strong>like a coin spinning on its edge.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/726/1*-tZAs1Gjcqfs5lNmhzk1fQ.png" /><figcaption>A classical bit vs a quantum bit | Source: <a href="https://docsend.com/view/pkmemb28nmvhug4e">The 2023 European Deep Tech Report</a></figcaption></figure><p>In quantum computing, superposition effectively means a qubit can exist in multiple states until it is measured and then collapses either into a state of 0 <strong>OR </strong>1. You can intentionally convert these qubits from their base state into superimposed states using quantum gates such as the Hadamard Gate, which are vital in the building of quantum algorithms.</p><p><strong><em>Quantum computing is not here to replace our everyday classical computers</em></strong></p><p>In an overly simplified way, a classical computer can only do a single calculation at once when dealing with multiple variables, but quantum computers can carry out multiple calculations simultaneously thanks to superposition, and another term called <strong><em>entanglement</em></strong> (qubits being able to interact with each other). The main takeaway from all this technical jargon is that <strong><em>quantum computing is ground breaking, but it is not here to replace our everyday classical computers</em></strong>. Not anytime soon anyway. Instead, they have very unique and interesting use cases for making sense of large amounts of complex data significantly faster and more efficiently in areas such as drug discovery, cleantech, finance and many more.</p><p>Let’s look at how each of these two computing categories would go about getting to the other side of a maze — this is a simplified schematic but provides a comprehensible analogy. We can begin to grasp the different capabilities of the classical vs. quantum computer and to what capacity quantum has an advantage, when we observe how they go about solving the same problem to arrive at the correct answer.</p><p>Let’s begin with the classical computer. At the entrance of the maze the computer would follow 1 path until it either reaches a dead end or gets to the exit on the first go. If it leads to a dead end, it would retrace its path back to the beginning and try different routes one by one until it gets to the exit.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4rTrvi-B7vhsPNpvZogoxQ.png" /></figure><p>Meanwhile, a quantum computer takes things to a whole new ball game. The quantum computer would use its superposition capabilities to walk down every path in the maze “simultaneously”, cancel out the paths which are most likely to be wrong, leaving the correct path remaining with the highest probability of being correct. This phenomenon is known as <strong><em>quantum parallelism</em></strong> (I know…so many technical terms). This phenomenon makes quantum computing extremely powerful for certain types of complex computations due to its ability to process data more efficiently in ways that classical computers can only dream of. Now, I’m sure you can already see why it is much faster than its classical counterpart, and why it has the potential to solve extremely complex tasks which our beloved classical computers are unable to solve today.</p><p>The superior processing power of quantum computers vs. conventional computers is also derived from <strong><em>quantum parallelism</em></strong>. <strong><em>Entanglement</em></strong> and <strong><em>superposition</em></strong> allow quantum computers to scale exponentially as the qubits effectively “interact” with each other. For instance, 2 qubits can store and process 4 bits of information while 4 qubits can store and process 16 bits of information. Very quickly we can see the exponential increase in computing power as you increase the number of qubits, compared to increasing bits in a classical computer which only increases linearly as you increase the number of transistors.</p><p><strong><em>Logical qubits are the name of the game in quantum computing</em></strong></p><p>Quantum computers are very error-prone today — this is no secret. We are in what is known as the noisy intermediate-scale quantum (NISQ) era, denoted by their error prone nature and low volume of qubits restricting computational power. Scaling quantum computers is not just about increasing the number of physical qubits. Instead, reliable “logical” qubits are just as important or even arguably more important for viable commercialisation. Now what are physical vs. logical qubits you may ask?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*F3KVqychd4TUqEdnO2Yx2g.png" /><figcaption>Comparison between a logical and physical qubit | Image credit: <a href="https://www.psiquantum.com/news-import/thinking-inside-the-blocks-fighting-faults-with-a-logical-instruction-set">PsiQuantum</a></figcaption></figure><p><a href="https://www.quera.com/glossary/logical-qubit">Logical qubits</a> are created by grouping physical qubits together and is used as an error-correction technique. Without these, reliable and large-scale error-tolerant quantum computing is not possible. Logical qubits essentially act as a corrector for the physical qubits whenever an error occurs, preventing errors from scaling exponentially with each subsequent operation as computational calculations occur in real-time.</p><blockquote><strong><em>The logical qubit works by “spreading” the information of a single physical qubit across several physical qubits</em></strong>, creating a redundancy.</blockquote><p>For example, if 3 out of 4 of the physical qubits in a logical qubit is in the “1” state and only 1 out of the four is in a “0” state then it can deduce that the “0” state is an error, whereby the physical qubit should be in a “1” state. Additionally, advancements can be made on the physical level by improving stability of physical qubits to increase how long they stay stable for.</p><p><strong><em>2024: a year which saw massive leaps in error-correction</em></strong></p><p>Two major breakthroughs in 2024 should be a wake-up call for quantum doubters and investors alike. Firstly, Microsoft and Quantinuum has made significant strides in 2024 by improving error-correction by a factor of 800 times using logical qubits. This means “the number of physical qubits required to create one logical qubit was reduced by a factor of 800”, improving the potential for scalability and error-correction in quantum systems. The takeaway from this is <strong><em>the sheer pace at which quantum technology is progressing</em></strong>. A year ago, these achievements were thought to be years away, yet they have come around much sooner than we thought.</p><p>Although there is a still a long way to go to create large-scale, fault tolerant quantum systems, <strong><em>the milestones achieved should be a wake-up call for investors regarding the speed of progress in quantum computing. </em></strong>We are on the brink of <a href="https://azure.microsoft.com/en-us/blog/quantum/2023/06/21/microsoft-achieves-first-milestone-towards-a-quantum-supercomputer/">level 2 out of 3 of Microsoft’s stages of quantum computing</a>, and the final level is to scale. Even though level 2 may be 1, 2, or 5 years away, once a certain volume of logical qubits is achieved to solve a game-changing use case reliably over classical computers, we will see investors flooding the market similar to what we saw with AI in 2023.</p><h3>💸 Quantum computing forecasted economic value creation is colossal and so are the use cases</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/498/0*zaTspV-FXG2ltJMh.gif" /></figure><p><strong><em>Even lower-range market size estimates for quantum computing are eye-watering</em></strong></p><p>Our world is quantum in nature. What does that mean? Naturally occurring processes, such as weather patterns and molecular behaviour, follow quantum mechanics, which lends itself readily to quantum computing techniques.</p><blockquote>In 2021, <strong><em>BCG forecasted a total annual value creation of $450–850 billion in operating income (EBIT) </em></strong>to end users by market maturity, taking into account both new revenue streams and cost reduction (roughly even split in value creation).</blockquote><p>This accounts for the forecasted annual value to providers which is estimated at $1–2 billion operating income within the current NISQ era prior to 2030, and $90–170 billion in operating income by maturity (by 2040). If McKinsey is more your cup of tea, they’ve estimated <strong><em>$900 billion to 2 trillion in economic value by 2030</em></strong> with a market size of <a href="https://www.mckinsey.com/featured-insights/the-rise-of-quantum-computing">$45–131 billion</a> by 2040. If neither of those float your boat, then Quantum Insider, the leading source of market intelligence in the quantum computing landscape, has forecasted between <a href="https://thequantuminsider.com/2024/09/13/the-quantum-insider-projects-1-trillion-in-economic-impact-from-quantum-computing-by-2035/"><strong><em>$200 billion and $1.1 trillion in economic value</em></strong></a><strong><em> between 2025–2030 </em></strong>with the majority derived from new revenue.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/968/1*9hJWbb3AeJLQfMam4lUKbQ.png" /><figcaption>Quantum computing forecasted value creation | Source: <a href="https://www.bcg.com/publications/2024/long-term-forecast-for-quantum-computing-still-looks-bright">BCG</a></figcaption></figure><p>When you think about the billion-to-trillion-dollar industries quantum is poised to disrupt and revolutionise, the colossal market estimates start to seem reasonable and even somewhat conservative, especially once universal enterprise-grade fault-tolerant systems are available. <strong><em>As of 2024, the quantum computing market is forecasted to surpass a value of $1 billion</em></strong> with a very impressive compound annual growth rate (CAGR) of 36%, <a href="https://thequantuminsider.com/2024/09/13/the-quantum-insider-projects-1-trillion-in-economic-impact-from-quantum-computing-by-2035/">with full stack and hardware making up 70% of the market</a>.</p><blockquote>BCG believes quantum computing is a “winner-take-most” technology, with <a href="https://www.bcg.com/publications/2023/enterprise-grade-quantum-computing-almost-ready#:~:text=Quantum%20computing%20in%20most%20industries,to%20find%20themselves%20locked%20out.">up to 90% of the value</a> created by quantum, going to early adopters.</blockquote><p>The forecasted revenue to providers is a key metric, as a proxy for market size to VCs and other investors who are keen to back companies who will capture a respectable share of this value. Assuming a conservative market share, these forecasted figures present eye-watering valuations and a market large enough to <strong><em>generate generous multiples of fund size returns</em></strong> and beyond, for VCs and other fund managers. Software companies are likely to have more attractive EV/EBITDA multiples in the medium- to long-term due to the nature of the subscription business model, nevertheless, hardware players are likely to capture significant value in the near term and beyond through cloud offerings (quantum-as-a-service) and through hardware sales to larger corporations. It is key to note that we are unlikely to see a smooth growth with quantum computing. Instead, the growth trajectory is likely to follow that of AI and reach an inflection point, leading to exponential market growth.</p><h3>🗝️<strong>Quantum computing can solve all our worries. Or does it?</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6cF0zsVWLbpTJHPr2oGS6w.png" /></figure><p>Quantum computing has the potential to carry out a plethora of use cases and solve a vast array of problems. It is easy to get lost in what it is truly capable of. Well, there are 4 computational archetypes that are accepted across the industry: <strong>simulation,</strong> <strong>optimisation, machine learning, and cryptography</strong>.</p><ol><li><strong>Simulation — </strong>simulating the dynamics of complex quantum-mechanical systems or processes.</li><li><strong>Optimisation — </strong>optimisation is all about finding the best and most efficient solution in the fastest amount of time.</li><li><strong>Machine learning — </strong>processing data sets to simplify computations exponentially faster — think AI.</li><li><strong>Cryptography — </strong>quantum factorisation (finding the product of two prime numbers) to decrypt RSA encrypted systems.</li></ol><p><em>Note: RSA (Rivest-Shamir-Adleman) is one of the most widely used public-key encryption algorithms, used for secure data transmissions.</em></p><p><strong>Simulation use case:</strong></p><ul><li>😷<strong><em>Pharma drug discovery</em></strong> — The drug discovery value chain is open to quantum technology disruption across its entirety from target identification to drug production and distribution. Conventional computers are unable to simulate complex molecules, which also happen to be quantum in nature. Quantum computing has the potential to shorten the pre-clinical drug discovery timeline which can take an average of 6 years, including a multitude of iterations to find a hit target, designing the right molecules for the drug, and in testing drug toxicity. The simulation potential of complex molecules could accurately predict protein structures and protein-to-drug interactions, reducing the need for relying on trial and error, and massively increase the time to market for effective drugs. Even though AI and computer-aided drug design has made headway in this field, there are limitations and errors which quantum computing can solve.</li></ul><p><strong>Optimisation use case:</strong></p><ul><li>💸<strong><em>Finance portfolio optimisation</em> </strong>— Monte Carlo simulations, a quantitative finance tool for derivative pricing, is an area that can be sped up and optimised to enhance risk management and increase decision-making in trades. After all, time is money. Additionally, quantum computing could identify different buying and selling scenarios for a wider combination of assets to further optimise portfolio gains and suggest more profitable trading-strategies above and beyond classical computing capabilities. High-performance computers are useful at carrying out these tasks, however, quantum computing can significantly increase speed and scenario analysis by analysing significantly more scenarios in a given time. The extent of the incremental gain and whether the rate of return will warrant the investment is up for debate. However, <strong>even a 1–2 % gain on trillions of algorithmically tradable assets is a significant ramp up in additional revenue and economic value</strong>. Firms like Goldman Sachs are not just dipping their toe in the water but instead going for a swim. This was shown in 2021, when they <a href="https://www.businesswire.com/news/home/20210921005441/en/Goldman-Sachs-QC-Ware-and-IonQ-Demonstrate-Quantum-Algorithms-Proof-of-Concept-That-Could-Revolutionize-Financial-Services-Other-Industries">developed an algorithm in collaboration with startups IonQ and QC Ware</a>, that has the potential to significantly speed up Monte Carlo simulations. The financial sector is gearing up to receive incremental gains from what quantum has to offer, with the sector being highlighted as one forecasted to receive the most value from quantum technology.</li></ul><p><strong>Machine learning use case:</strong></p><ul><li>🚃<strong><em>Automotive: autonomous-vehicle AI algorithm design</em></strong> — There is no way we could talk about quantum without AI having some part to play. AI and quantum computing is thought to be a match made in heaven when we begin to think about the synergistic advancements each technology can bring to the other. Quantum linear algebra is an array of quantum techniques which is primarily applied to AI and machine learning to massively shorten training times within machine learning pipelines. For example, in autonomous driving, this can drive down the cost of synthetic data generation and improve vehicle performance in unfamiliar scenarios, effectively reducing R&amp;D timelines.</li></ul><p><strong>Cryptography use case:</strong></p><ul><li>🕵<strong><em>Cryptography</em></strong> — our encryption systems used to protect our most private data is underpinned by using extremely large prime numbers that would take even high-performance computers millions of years to factorise using brute force to break the encryption. However, Shor’s algorithm has proven that it is capable of breaking today’s RSA systems, as soon as an error-correcting quantum computer powerful enough to run the algorithm is developed. Governmental bodies and businesses alike are preparing themselves to adopt quantum-resistant cryptographic algorithms which are being developed and integrated. It will take some time, firstly, for standardisation of the algorithms, and secondly, for companies to incorporate the algorithms which is more than a plug-and-play operation thanks to quantum computing’s unique properties.</li></ul><h3>🔥The investment landscape is heating up, especially in Europe</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/0*EwsN3-olu6nS5TCR.gif" /></figure><p><strong><em>Some would even dare to call it the modern-day Manhattan Project</em></strong></p><p>Quantum computing has come a long way due to billions of investments being poured into research &amp; development and is already forecasted to be <a href="https://thequantuminsider.com/2024/09/13/the-quantum-insider-projects-1-trillion-in-economic-impact-from-quantum-computing-by-2035/"><strong><em>valued at over $1 billion in 2024</em></strong></a>. Governments across the globe view advancements in quantum computing as a must to secure national sovereignty and win the quantum arms race. Consequently, they have been pouring generous amounts of public funding into the sector.</p><p>The RSA encryption dilemma mentioned earlier is at the forefront of quantum strategies of countries worldwide as they compete to achieve quantum supremacy. If global powers do not act in time, the first to build a powerful, stable, and fault-tolerant quantum computer has the potential to destabilise their rivals’ encryption methods. The adversaries will likely unlock a plethora of encrypted information if systems are not transferred to post-quantum encryption in time (which currently still requires standardisation). On a small scale, this means accessing the average Joe’s banking information and on a larger scale retrieving damaging geopolitical secrets.</p><p>We have seen public investment commitments significantly outweigh the private sector as the incentive to be the first to cross the finish line is at the forefront of government leaders across the globe. However, committed government investment to quantum is over a lengthy time horizon, hence, annually deployed private capital significantly outweighs public investment.</p><p><strong><em>Europe is investing in quantum like there is no tomorrow</em></strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tTTEYvuNRkNYAZqQTRErNw.png" /><figcaption>Image credit: <a href="https://sifted.eu/pro/briefings/quantum-computing-2024">Sifted</a></figcaption></figure><p>Public investment commitments have grown considerably over the past few years amassing to between <a href="https://www.meetiqm.com/newsroom/press-releases/state-of-quantum-report-2024">$40–50 billion</a> with key regions such as the US allocating investment (announced) upwards of $3B across 2020–2023 into quantum R&amp;D. Across the pond, the UK, who has hopes of a quantum-enabled economy by 2033, has been pumping considerable capital of their own including a recent <a href="https://www.theguardian.com/science/article/2024/jul/26/quantum-leap-for-uk-as-tech-industry-receives-100m-boost">£106 million government funding</a> in July 2024 towards hubs focused on primarily national security, clean energy, and healthcare. Other European regions such as Netherland and Switzerland, have built quantum hubs in a bid to position themselves as global leaders in the space.</p><p>However, <strong><em>public investment alone is not enough</em></strong>. In fact, it is the catalyst needed to garner investment from the private sector by bringing quantum computing closer to achieving commercial use cases. Private investment into quantum computing did not bode very well in 2023 and was initially thought to be a result of investor sentiment towards the maturity of the technology. The somewhat positive news for the quantum space is that this is broadly in line with the wider macroeconomic environment which saw a sharp decline in annual investment in the venture capital ecosystem.</p><blockquote>From 2019 to 2023 private investment into quantum technologies (including sensing and communications) grew more than 4x from $400 million to $1.8 billion, peaking at $2.6 billion in 2022.</blockquote><p>The decline since 2022 has raised suspicions of a “quantum winter” and investors shying away from the technology until stronger commercial applications become evident. However, the 50% decline in private investment between 2023–24 matched the sentiment in the wider venture capital industry which saw a <a href="https://dealroom.co/uploaded/2023/12/Global-end-of-2023-Venture-Unwrapped-Dealroom.pdf?x30228">38% decrease in investment in 2023 compared to the year prior</a>. Private investors and founders, particularly in Europe, can seek solace in the 2024 investment-to-date in quantum computing, which has already risen above 2023 levels and pre-2021 levels, suggesting a rebound in investor sentiment alongside the <a href="https://www.mckinsey.com/~/media/mckinsey/business%20functions/mckinsey%20digital/our%20insights/steady%20progress%20in%20approaching%20the%20quantum%20advantage/quantum-technology-monitor-april-2024.pdf?shouldIndex=false">increase in deployed public funding in 2023</a>.</p><p>While the global VC quantum landscape peaked in 2021–22 then declined, Europe’s quantum computing funding has held strong and has been on the rise year-on-year since 2018. With $572 million already raised in 2024 (as of September 2024), Europe can continue this annual trend with a quarter still remaining. The global decline in quantum funding has been largely due to the Americas which saw an 80% decline in investment towards the sector in 2023. This shows a stark contrast in investor sentiment in the US versus Europe, as private investors in Europe seem to be showing more optimism towards quantum computing and the talent in the continent.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Vux8Fmt0pjrMM73JKzb4jA.png" /><figcaption>Major quantum computing startups funding round deal sizes (2023–Q3 2024)</figcaption></figure><p>For founders, <strong><em>Europe should be on top of the priority list to start a quantum technology company </em></strong>which is atypical to what we’ve seen with many European founders choosing to set up shop in the US due to larger market and cheque sizes. Another observation with Europe fighting their stereotype is the shift towards deploying capital in later stages as we are seeing total Series B and C annual deal values increases year-on-year. This should give European founders and VCs confidence that the quantum computing has arrived and is here to stay. Some notable later stage fundings in Europe in 2023–24 include Oxford Quantum Circuits ($100m Series B), Pasqal (€100m Series B) and Riverlane ($75m Series C).</p><h3>🚧<strong><em>Decline in newly founded quantum computing startups has not deterred investment into the sector</em></strong></h3><p>Interestingly, the number of newly founded quantum technology startups has been on the decline from 2019 to 2023 even though investment has increased significantly in the same timeframe. There are several angles that can potentially explain the reason for this trend.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8QyJ9qt5rk0yr_R7Ygy4IA.png" /><figcaption>Source: <a href="https://www.mckinsey.com/~/media/mckinsey/business%20functions/mckinsey%20digital/our%20insights/steady%20progress%20in%20approaching%20the%20quantum%20advantage/quantum-technology-monitor-april-2024.pdf?shouldIndex=false">McKinsey</a></figcaption></figure><ol><li><strong>Despite a growing interest in software players, hardware is the main bottleneck,</strong> but has a high barrier of entry due to the substantial capital required and technical expertise which is not very widespread currently. Hence, combining this with the unfavourable macroeconomic and venture capital landscape, this could be a factor as to why less startups are being founded in this sector.</li><li><strong>Investors are likely to be more selective towards more established startups</strong> displaying progress rather than a riskier bet on new companies in a nascent field, as they assess timelines for quantum computing commercialisation. This indicates a shift towards scaling and maturing existing startups which we have seen with majority of <a href="https://www.mckinsey.com/~/media/mckinsey/business%20functions/mckinsey%20digital/our%20insights/steady%20progress%20in%20approaching%20the%20quantum%20advantage/quantum-technology-monitor-april-2024.pdf?shouldIndex=false">investments concentrated towards Series A and B</a> compared to seed rounds.</li><li><strong>Talent is likely to be another bottleneck for growth</strong>. Quantum computing requires niche domain expertise which is limited and with the majority of new startups being founded 3+ years ago, majority of quantum computing talents may be working at existing startups.</li><li><strong>Generative AI has been the focus in 2023</strong><em> </em>demonstrated by a 2.5x increase in investment into the sector compared to the previous year. With favourable use cases and near-term commercial potential, quantum technology waned in comparison from an investment standpoint. Quantum computing is largely esoteric at the moment and needs to prove its reliable application in commercial use cases before widespread adoption is seen, akin to Gen AI.</li></ol><p><strong><em>US, China, Japan, Europe, and Canada: where should VCs and private investors put their money?</em></strong></p><blockquote>In a deeptech sector like quantum computing which is very reliant on the underlying technology and intellectual property, grants filed and granted are a good indicator of startup activity especially as many startups in the sector are spun out from academic labs.</blockquote><p>Unsurprisingly, the US and China dominates when it comes to patents filed, with US and Japan leading on volume of patents granted, demonstrating the investment and push towards innovation within quantum computing in these regions and also due to the sheer population size of these countries. VCs may be more drawn to the regions with higher granted patents, e.g. US, Japan, and France, as it suggests a higher likelihood of translating innovation into viable products, which is more important in an industry which is still proving its commercial prowess.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ArMnSIZSTi4IetYK9CyTxA.png" /><figcaption>Source: <a href="https://www.mckinsey.com/~/media/mckinsey/business%20functions/mckinsey%20digital/our%20insights/steady%20progress%20in%20approaching%20the%20quantum%20advantage/quantum-technology-monitor-april-2024.pdf?shouldIndex=false">McKinsey</a></figcaption></figure><p>In Europe, countries, including France, Germany, and Great Britain, are demonstrating leading quantum computing innovation activity when using patents filed and granted as a proxy. This comes as no surprise with each of the 3 countries committing government funding upwards of, $1 billion, $2 billion, and $3 billion to the sector, respectively, driving research and development efforts to become a global leader.</p><p>VCs and other private investors should keep a close eye on these regions to scope out disruptive startups with strong commercial abilities. Other regions such as Switzerland which have a Centre of Competence for Quantum and AI called Quantum Basel are strong contenders in the race and have granted patent numbers above the likes of Great Britain. Similarly to the wider deeptech landscape, universities and quantum hubs will be strong hotspots for sourcing deals, as scientists transition their research projects into commercialisation.</p><h3>☁️<strong><em>Quantum-as-a-Service (QaaS) cloud offering is the bridge between quantum and end user adoption today</em></strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/0*5y0a4x3FZBxyE1kr.gif" /></figure><p>For investors concerned about time to wide-scale commercialisation, <strong><em>hybrid classical computing is the near-term answer which offers value to end-users</em></strong> and drive commercialisation primarily through cloud services e.g. Amazon Braket. This allows businesses to access quantum-enhanced processing without the high capital expenditure and enables the ability to develop personalised use cases in line with the progression of quantum technology hardware and software.</p><p>Quantum computing-as-a-Service (QaaS) negates the need for high capital expenditure on quantum hardware at a time when the market is nascent, and the cost is expected to decline dramatically as technology maturity grows. Additionally, this lower barrier to entry allows early market adoption not just for large corporates, but also SMEs who can begin building personalised use cases while leveraging the incremental improvements in the technology over time to solve these use cases. Not to mention, the spare cash from adopting the as-a-service model will be vital in hiring talented expertise with domain knowledge to identify the most value-add use cases for enterprises, and to help companies capitalise on new revenue streams and cost efficiencies.</p><p>The segment of the market expected to witness the most growth is QaaS, driven by the growth we are witnessing with cloud-based services offered by AWS, Azure, and Google, alongside the increasing in uptake by enterprises through this channel. QaaS currently compromises roughly <a href="https://thequantuminsider.com/2024/09/13/the-quantum-insider-projects-1-trillion-in-economic-impact-from-quantum-computing-by-2035/">13% of the quantum computing market</a>, but is expected to make up over 40% of the quantum computing market by 2030, demonstrating the immense growth expected in this segment as the market matures and overtakes professional services and steals market share from hardware.</p><p>The hybrid quantum computing approach is the dominant operating model approach largely owing to the nascency of the quantum computing market and the lack of commercial maturity of standalone quantum computing technology. Today, the majority of commercial use cases involve high-performance computing (HPCs aka supercomputers) which are supplemented by quantum algorithms. However, <strong><em>as the market matures, we will see turnkey services using standalone quantum computers</em></strong>, rather than customised solutions to fit niche business cases which we see today.</p><p>Players like AWS have already developed quantum computing cloud services (Amazon Braket) which allows developers to access quantum computing on demand (per shot and per task), or per hour on Braket Direct if you want to reserve dedicated device access, to develop algorithms leveraging quantum hardware companies (Ion Q, Rigetti, IQM, and QuEra). A user gets the option to try multiple hardware technology types through this service, via a pay-per use model. However, affordability depends on the number of task and number of shots within each task.</p><p>With the wider uncertainty of quantum technology, <strong><em>increasing access through cloud is likely to be a significant driver for quantum computing uptake</em></strong> as businesses can “try before they buy” without committing to extremely high capital expenditure and drowning their balance sheet. Amazon and Microsoft dominate this space and it is difficult to envision startups challenging these incumbents who have a firm grip on the cloud market. However, as mentioned, VCs looking to invest in hardware, cloud computing is the bridge that will drive early revenue generation in commercial use cases.</p><p><strong><em>Vertical quantum solutions is a potential play for VCs ready to dive in now</em></strong></p><p>The goal is to achieve universal fault-tolerant quantum computing, yet quantum computers today are highly specialised to solve specific problems. This is not a bad thing. Quantum computing is already being put in the hands of end users through application-specific quantum solutions. As discussed, there are industries which lends itself readily to quantum computing such as drug discovery, logistics optimisation, financial portfolio optimisation etc., based on problems in these use cases having more quantum mechanics-like characteristics. Startups have begun providing tailored quantum solutions for clients in these industries, building enterprise solutions that run on quantum hardware.</p><p>For example, <a href="https://www.qcware.com/">QC Ware</a> is a quantum computing software startup building quantum enterprise solutions for pharmaceuticals and material science. QC Ware’s Promethium platform combines quantum molecular discovery applications with NVIDIA’s AI and AWS cloud services to offer an advanced molecular discovery platform. Enterprises can access Promethium on AWS, another example of QaaS business model being leveraged, this time for application-specific quantum solutions. VCs and private investors who would like to bet on already commercially viable use cases, can target these application-specific quantum solutions whose software will mature in-line with both the quantum computing industry and the specific industry vertical to which the technology is being applied.</p><p>VCs should look for quantum software startups who are solving real-world enterprise pain points, defined by “demand-pull” use cases rather than “technology push”. In the example of QC Ware, their platform promises significant time improvement with high orders of magnitude and substantial costs savings, which are key issues in the pharmaceutical industry.</p><h3>⌛Concluding notes</h3><p>The quantum market is still in its infancy and investors’ apprehensions are somewhat warranted. However, the technology is advancing rapidly, the market size is enormous, public and private investors are assembling, and there are many opportunities for startups to solve key challenges in the space. If that is not a market ripe for investment, then what is? With that being said, investors need to do their due diligence on startup technologies to understand the true potential in a market that is still looking for its winners. The quantum computing market has not had its ChatGPT moment yet, but with the potential AI and quantum computing have together, it will only be a matter of time. Stay tuned for more content on quantum technology. Watch this space!</p><p><em>You made it to the end! Feel free to reach out to me and connect on </em><a href="https://www.linkedin.com/in/donald-harmitt-65199b170/"><em>LinkedIn</em></a>,<em> if you have any questions or thoughts you would like to discuss. Look out for the next deep dive which will delve further into the technical side of quantum computing technology (hardware and software focus).</em></p><p><em>I would like to say a massive thank you to Nikita Thakrar, the Included VC team, and my cohort members for their inspiration and guidance in crafting this deep dive!</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bf46b5fa0795" width="1" height="1" alt=""><hr><p><a href="https://medium.com/included-vc/from-vision-to-value-why-quantum-computing-belongs-in-your-portfolio-today-bf46b5fa0795">Revolutionary Tech or Risky Bet? The Untold Truth About Quantum Computing’s Potential</a> was originally published in <a href="https://medium.com/included-vc">Included VC</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>