Infinity
I have always had a slightly dangerous relationship with optimism.
I once ran a Substack about one-person businesses. I genuinely believed we could one day have billions of businesses run by individuals, changing the modern idea of a “job” forever. I saw it as a more interesting form of decentralized democracy, where economic participation was no longer a popularity contest controlled by a handful of employers.
My worldview is fairly simple and has remained consistent over the years. If a possibility crosses your mind, it is worth exploring. If something could lead to a better outcome for society, we should at least figure it out.
So I have been called optimistic. Fearless. An infinite thinker. Some people also call it reckless behavior. And, I don’t really care which label is correct.
What I have noticed, though, is that the older I get, the harder it is to stay optimistic without also developing a healthy respect for what can go wrong.
There is a line often attributed to Einstein: “Two things are infinite: the universe and human stupidity; and I’m not sure about the universe.” He probably never said it, but I understand the sentiment more now than I used to.
Before COVID, when I was working on blockchain, I used to say that one day humans would kill each other for survival. Then COVID arrived and showed how quickly a shock in one place can cascade through tightly connected systems everywhere.
For the past few years, I have followed developments in post-quantum encryption, cybersecurity, and critical infrastructure, and watched almost everything we do become dependent on digital systems. I have grown uncomfortable with how little most of us understand about the infrastructure underneath modern life.
I don’t assume our banks, payment systems, cloud platforms or communications networks will always be available simply because they are available today. That is not paranoia. It is a question of system design.
The next major conflict may not begin at a land, sea or sky border. It could begin with banks, hospitals, telecom networks and power grids failing at roughly the same time, while governments are still trying to figure who caused it. The uncomfortable part is that almost every piece of that scenario has already happened somewhere. Just not all at once.
History keeps repeating the same patterns
Wars have always been mostly about controlling information. The weapons change, the infrastructure changes, and the speed changes, but the underlying principle stays remarkably consistent.
World War I. The day after Britain declared war, a British cable ship cut Germany’s undersea telegraph cables. Germany was forced onto radio and routes the British could intercept. Communications became a strategic weapon: information could be intercepted, redirected, and controlled.
World War II. Codebreaking became industrialised. Bletchley Park grew from about 200 people in 1939 to nearly 9,000 by late 1944. Cryptography mattered so much that breaking codes required building entirely new kinds of machines.
The Cold War. In 1983, a Soviet early-warning system reported that five American nuclear missiles were heading toward the Soviet Union. Stanislav Petrov was the duty officer. The system said attack. He decided it was probably wrong, and he was right: the satellites had mistaken sunlight reflecting off clouds for missile launches.
That story matters because Petrov did not save the world with better technology. He saved it by distrusting the technology.
Ukraine. On the day Russia invaded in 2022, a cyberattack disrupted Viasat’s KA-SAT satellite network. It hit Ukrainian military communications, but the damage did not stop at the battlefield. About 5,800 wind turbines in Germany lost remote monitoring and control. The physical war had barely begun, and the digital war was already crossing borders.
The pattern is hard to ignore:
- Networks are attacked before, or alongside, physical infrastructure.
- Information can be more strategically valuable than going nuclear.
- Automated systems can turn uncertainty into action.
- Digital attacks do not respect national borders.
- Very small weaknesses can have enormous consequences.
A misconfigured system. A compromised password. A vulnerable supplier. A software update. One decision made at machine speed. That is the part that worries me most.
For most of modern history, geopolitical power came from physical resources: oil, steel, shipping lanes, factories, and military bases.
Today, three more resources increasingly decide how resilient a economy is: cloud, compute, and cyber. Data connects all three, and AI is increasingly orchestrating them.
Before the Russian invasion, Ukrainian law required government data to stay on servers inside the country. As the threat grew, parliament changed the law a week before the invasion so data could move to the cloud. Microsoft then helped move data from 16 of Ukraine’s 17 ministries out of the country. When missiles hit a government data facility, the records had already been copied to the cloud.
This is one of cloud infrastructure’s most interesting traits: a country can lose a building without losing its institutional memory. But there is a second side. The more a government depend on foreign cloud providers, the more its digital sovereignty sits somewhere else. The cloud solves one problem while creating another.
The AI economy runs on advanced semiconductors, and their manufacture is concentrated in very few places. TSMC in Taiwan makes more than 90% of the world’s most advanced chips. Those chips power AI systems, smartphones, data centres, and increasingly sophisticated defence systems.
That creates a strange geopolitical situation: one small region has become one of the most important pieces of infrastructure in the global economy. If advanced chip production in Taiwan were seriously disrupted, the consequences would reach far beyond Taiwan.
This is why chips have become geopolitical instruments. Since October 2022, the United States has restricted China’s access to advanced AI chips and chipmaking equipment. China, in turn, is investing heavily in domestic alternatives.
The chip is no longer just a component. It is leverage.
Quantum changes the meaning of encryption
The quantum problem is stranger, because the threat does not need the technology to exist yet. Much of today’s digital security relies on maths problems that ordinary computers find extremely hard to solve. A powerful enough quantum computer and annealing could change that. Nobody knows exactly when such machines will arrive, but the security community has already named a related problem: harvest now, decrypt later.
An attacker can steal encrypted data today and store it. If quantum computing later makes that encryption breakable, the data could become readable years from now.
That changes the timeline of cybersecurity. The threat does not begin when the quantum computer arrives. It begins when valuable information is collected and stored in anticipation of that day. It is why NIST published its first quantum-resistant encryption standards in August 2024. Post-quantum cryptography is not really a futuristic technology problem. It is an infrastructure migration problem.
Cyber is the ability to plant landmines
In 2024, US and allied agencies, including Canada’s Centre for Cyber Security, warned about Volt Typhoon, a Chinese state-sponsored hacking group that had compromised communications, energy, transportation, and water systems.
The concern was not simply espionage. It was positioning for future disruption. Get inside the network. Understand the infrastructure. Stay quiet. Wait.
That is a very different model of warfare. You don’t necessarily attack when you get in. You get in because you may want the ability to attack later. And once critical infrastructure is digitally connected, the line between civilian and military targets becomes very hard to draw.
AI connects the whole system
This is where AI changes the equation. AI needs compute. Compute needs chips. Chips depend on highly concentrated supply chains and energy. AI increasingly runs in the cloud, and the cloud connects governments, companies, and critical services.
Attackers can use AI to find vulnerabilities, write malicious code, automate reconnaissance, and scale attacks. Defenders can use AI to do the same from the other side. Eventually, increasingly autonomous systems may make decisions faster than humans can understand what is happening.
Get Vartika Manasvi’s stories in your inbox
Join Medium for free to get updates from this writer.
That problem is not mainly about intelligence. It is about speed. Humans are slow; computers are not.
If two automated systems are responding to each other, the escalation loop can move faster than the people responsible can step in. This is where the Petrov story becomes relevant again. He had time to think. The question is whether we will preserve that time in an AI-enabled conflict.
A digital weapon does not need to cross a border
We already have examples of how much damage can be done without conventional weapons.
Stuxnet showed that software could physically destroy industrial machinery. The malware, widely attributed to the United States and Israel, targeted Iran’s Natanz nuclear facility and destroyed around 1,000 centrifuges.
NotPetya showed something else. A 2017 cyberattack aimed at Ukraine spread worldwide and caused more than $10 billion in damage, according to a White House estimate. Companies such as shipping giant Maersk were severely disrupted.
Viasat added another dimension. A cyberattack at the opening of the war in Ukraine disrupted satellite communications and spilled over into European infrastructure.
So the important question is no longer whether cyberwarfare can cause physical consequences. We know it can. The question is what happens when the systems that detect and respond to these attacks also become increasingly autonomous.
Imagine the sequence
The following scenario doesn’t require science fiction. It only requires technologies and behaviours that already exist to start interacting with each other.
First, countries and companies keep giving AI systems more autonomy, because autonomy creates speed. Nobody wants to be slower than their competitors.
Then, cyber teams quietly establish access to critical infrastructure: power, water, telecommunications, transportation, financial systems. Nothing happens. The access simply exists.
Then geopolitical tension rises around a strategically important region. Cybersecurity systems go on high alert. AI systems are authorised to detect and respond to threats on their own, because humans cannot keep up with the volume and speed of attacks.
Then something unusual happens. An AI system detects activity that looks like the start of an attack. Its instructions tell it to neutralise the threat, so it responds. The other side’s system sees that response as an attack, and responds again. The cycle accelerates.
The humans arrive late. Not because they are incompetent, but because the machines were faster.
Then the attack spills beyond its original target. A hospital loses access to critical systems. A port stops operating. A payment network goes offline. A telecom provider goes dark. A power system drops into degraded mode. A cloud provider is pressured to disconnect customers.
Nobody can immediately establish who started the escalation. And because nobody knows who started it, nobody wants to be the first to stop responding. At that point, the conflict is no longer just military. It is economic, financial, and social.
The most dangerous part is that nobody necessarily wanted this outcome. Every individual decision made sense: move faster, defend yourself, respond to the intrusion, protect your infrastructure, don’t fall behind.
That is exactly what makes systemic risk so difficult. You don’t need evil actors. You only need rational decision makers operating inside a system that rewards escalation.
The internet probably won’t disappear
I don’t think the internet will simply switch off one morning. There is no single switch, unfortunately. The internet is a collection of interconnected networks designed, in part, to route around failures.
We have already lived through surprisingly large disruptions without the whole network collapsing.
In July 2024, a faulty CrowdStrike software update crashed about 8.5 million Windows computers worldwide. It wasn’t a cyberattack, yet airlines, hospitals, and businesses were disrupted. It was a useful reminder that you don’t need an adversary to create systemic failure. Complexity creates its own failure modes.
Then there are the physical networks underneath the digital ones. Undersea cables carry about 99% of international data traffic, and when they are damaged, whole regions can lose connectivity. In early 2024, damage to Red Sea cables disrupted up to 70% of Europe–Asia data flow, according to one operator. Later that year, two Baltic Sea cables were cut on the same day, and investigators suspected sabotage.
Governments themselves also switch the internet off. In 2024, there were 296 shutdowns in 54 countries, a record, many of them during conflicts.
So I don’t think the most likely future is a global blackout. It may be something more subtle: fragmentation.
A world where the internet still works, but increasingly as separate national or geopolitical ecosystems. Different clouds. Different standards. Different chips. Different AI models. Different encryption. Different rules about data. Different definitions of digital identity.
A splinternet, rather than an internet that has disappeared. The lights stay on, but the world becomes less connected, once again.
What should we actually do about it?
The obvious answer is build: more cyber defence, compute, chips, encryption, more AI perhaps? We need all of it. But I don’t think technology alone solves this problem. We also need institutional design.
The lesson of the Cold War is not simply that we needed better weapons. It is that we needed ways to communicate when everyone was operating under extreme uncertainty and fear. The Washington–Moscow hotline was agreed in June 1963, a few months after the Cuban Missile Crisis brought the two sides to the brink. When machines, military systems and people are operating under pressure, communication can itself be a form of defence.
We need the digital equivalent:
- Rapid communication channels between governments during major cyber incidents.
- Redundancy in critical infrastructure, so one failure doesn’t become everyone’s failure.
- Resilient access to compute and semiconductor supply chains.
- An early move to post-quantum world, before it becomes urgent.
- Clear international norms against attacking hospitals, water systems, energy infrastructure, and other civilian systems.
Most importantly, we need to think carefully about how humans stay in the loop. I am not saying that every cybersecurity decision needs manual human judgement. That would be impossible. But there is a difference between automated defence and automated retaliation. One protects a system. The other can escalate a conflict. That distinction matters.
The uncomfortable conclusion
I started thinking about technology because I was fascinated by what humans could build. Blockchain could change financial systems. AI could change how we work. One-person businesses could change employment. Cloud computing could democratise access to infrastructure. Quantum computing could change what is computationally possible.
I still believe all of that. I am still an optimist. But optimism without systems thinking is just wishful thinking.
The technologies we are building are not developing independently. Cloud, compute, quantum, cyber and AI are converging and becoming one interconnected infrastructure. And the more interconnected the system becomes, the further a small failure can travel through it.
That is what history keeps teaching us. The wires matter. The code matters. The supply chain matters. The person sitting in front of the warning system matters.
Sometimes the most consequential decision is not the decision to act. It is the decision to pause. And only humans have that ability to not react.
The next world war may not begin with someone deciding to start one. It may begin with everyone deciding they cannot afford to be the one who waits. That is the part worth thinking about now. Not when the systems fail. Before they do.
