<?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 Eyup Korkmaz on Medium]]></title>
        <description><![CDATA[Stories by Eyup Korkmaz on Medium]]></description>
        <link>https://medium.com/@eyupkorkmaz?source=rss-63716835eb64------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*K77kcGKSAiV2814R.</url>
            <title>Stories by Eyup Korkmaz on Medium</title>
            <link>https://medium.com/@eyupkorkmaz?source=rss-63716835eb64------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 18:59:57 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@eyupkorkmaz/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[Batteries are not a component anymore. They are part of the product design ]]></title>
            <link>https://medium.com/@eyupkorkmaz/batteries-are-not-a-component-anymore-they-are-part-of-the-product-design-5d36b91f7326?source=rss-63716835eb64------2</link>
            <guid isPermaLink="false">https://medium.com/p/5d36b91f7326</guid>
            <category><![CDATA[lithium-battery]]></category>
            <category><![CDATA[battery]]></category>
            <category><![CDATA[electrical-vehicles]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[smartphones]]></category>
            <dc:creator><![CDATA[Eyup Korkmaz]]></dc:creator>
            <pubDate>Sat, 24 Oct 2026 15:58:46 GMT</pubDate>
            <atom:updated>2026-10-24T15:58:46.351Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rNrkj5kZeUBJhSsb0hfzSQ.jpeg" /><figcaption>credit to freepik</figcaption></figure><p><strong>The AI era has made every device hungrier ⚡</strong></p><p>A few years ago, a phone mostly waited for you to tap the screen. Today it runs AI assistants, translates calls live and edits photos on the device itself. That means long bursts of high power and a lot of heat, in a body that still has to be slim.</p><p>The same is true on a much larger scale in a data centre. Thousands of GPUs run all day, and the power has to be there every minute. Different size, same question: where does the energy come from, and how do we use it well?</p><p>Looking across the entire tech ecosystem, battery innovation is moving along two distinct vectors: massive macro-infrastructure deployment and hyper-dense micro-system packaging.</p><p><strong>1. Macro Scale: EVs, Grid Storage, and AI Data Centers</strong></p><p><strong>1.1. Electric vehicles still set the pace 🚗</strong></p><p>EVs remain the biggest driver of battery demand. The IEA says EV battery deployment reached around 1.2 TWh in 2025, with growth close to 30%. Lower-cost LFP batteries keep winning share, while sodium-ion and solid-state are getting serious investment.</p><p>What matters is that EV volume brings down cost for everyone. Progress in one area often helps another.</p><p><strong>1.2. Data centres and long-duration storage ⚡</strong></p><p>Short-term lithium-ion storage is useful, but renewables, factories and AI data centres may need power for hours or even days. That is why flow batteries, iron-air batteries, hydrogen and reversible solid oxide cells are worth watching.</p><p>Many of these are still working on durability, cost and manufacturing scale. I expect future data centres to mix several power and storage technologies, not rely on one.</p><p><strong>1.3. Long-Duration Energy Storage (LDES):</strong> Beyond traditional short-duration lithium-ion, grid-scale systems require multi-hour resilience. Emerging technologies like flow batteries, iron-air systems, and reversible solid oxide cells (rSOC) — which generate hydrogen from excess electricity and convert it back on demand — are entering the commercial pilot phase.</p><p><strong>2. Micro Scale: On-Device AI, Wearables, and Spatial Hardware</strong></p><p><strong>2.1 Phones: more capacity without more size 📱</strong></p><p>The most practical change right now is the silicon-carbon anode. Adding silicon lets a battery hold more energy in the same space. This is why some new flagship phones carry 5,000 mAh or more and still look slim.</p><p>Fast charging helps, but it also wears the battery. So the smart part is not only the cell. It is the power management chip and how the phone shares the load.</p><p><strong>2.2. Wearables: every bit of space counts ⌚</strong></p><p>A smart ring is a good example. There is almost no room inside. A normal wound battery wastes space in a round shape, so makers now use stacked layers and custom-shaped cells.</p><p>Safety matters even more here, because the device sits on your skin all day. That is why small solid-state and semi-solid cells are getting attention.</p><p><strong>2.3. Glasses and headsets: weight is the enemy 🕶️</strong></p><p>If smart glasses feel heavy, people take them off. So designers put small cells in the arms of the glasses or in the strap of a headset to balance the weight. Then they have to carry the heat away from your face. It is a battery problem, a weight problem and a heat problem all at once.</p><p><strong>2.4 Edge IoT &amp; Medical Devices 🏥</strong></p><p>Industrial sensors and implantable medical patches prioritise low self-discharge chemistries (&lt;1% annual leakage) and hybrid micro-energy harvesting to ensure 5-to-10-year zero-maintenance lifespans.</p><p><strong>What I take from all this 💡</strong></p><p>The old way of thinking was simple: find a better battery chemistry and the product gets better.</p><p>That is no longer enough. The best products will win by designing everything together:</p><ul><li>the chip and how much power it draws,</li><li>the software that decides when to work hard and when to rest,</li><li>the battery management system,</li><li>the heat design and the shape of the device,</li><li>and the battery cell itself.</li></ul><p>Two phones with the same battery size can last very different amounts of time. The difference is the system around the battery.</p><p>In my view, this will be the new line between good and great hardware. Performance per watt will matter as much as raw performance. And I do not believe one battery type will ever fit every use. A smart ring, a hospital backup system and an AI data centre need very different things.</p><p><strong>My takeaway</strong></p><p>We talk a lot about faster chips and smarter AI models. But energy is what lets all of that run in the real world. The companies that treat the battery as part of the whole design, not a part added at the end, will have the advantage.</p><p>What do you think? Which will matter most over the next five years: a new battery chemistry, or smarter co-design of chip, software and battery?</p><p>#BatteryTechnology #AI #Semiconductors #EdgeAI #Wearables #ElectricVehicles #DataCenters #EnergyStorage #Innovation</p><p><strong><em>Disclaimer: These are my personal views and do not represent the views or strategies of any current, previous or future employer.</em></strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5d36b91f7326" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Will Not Stop: The Next Industrial Revolution Is Already Here]]></title>
            <link>https://medium.com/@eyupkorkmaz/ai-will-not-stop-the-next-industrial-revolution-is-already-here-c8b9b2c12159?source=rss-63716835eb64------2</link>
            <guid isPermaLink="false">https://medium.com/p/c8b9b2c12159</guid>
            <category><![CDATA[history]]></category>
            <category><![CDATA[humanity]]></category>
            <category><![CDATA[life]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[future]]></category>
            <dc:creator><![CDATA[Eyup Korkmaz]]></dc:creator>
            <pubDate>Fri, 18 Sep 2026 16:23:56 GMT</pubDate>
            <atom:updated>2026-09-18T16:23:56.866Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1010/1*bD4WLCpumZkqdcSu0G7VGA.png" /><figcaption>credit to freepik</figcaption></figure><p>There is a lot of discussion today about slowing down Artificial Intelligence.</p><p>Some technology leaders are warning about the risks of AI. Governments are discussing new rules. Companies are talking about responsible AI, safety and the possible impact on jobs.</p><p>These discussions are important.</p><p>But I believe we should also look at what these companies and countries are actually doing, not only what they are saying.</p><p>And what they are doing tells us something very different:</p><p><strong>AI will not stop.</strong></p><h3>We Have Seen This Before</h3><p>Before the industrial revolution, horses were one of the most important sources of power for human activity.</p><p>They transported people. They moved goods. They worked on farms. They were used in construction and in wars. Entire industries were built around horses.</p><p>Then came the automobile.</p><p>Henry Ford did not invent the car, but mass production made cars affordable and accelerated their adoption.</p><p>Gradually, cars and trucks replaced horses for most transportation and physical work.</p><p>Did horses disappear?</p><p>No.</p><p>But their economic role changed dramatically.</p><p>I believe we are entering a similar transition with Artificial Intelligence.</p><p>I am obviously not comparing humans with horses. The point is about the economic role of a technology.</p><p>For decades, companies needed people to perform repetitive administrative work, analyse basic information, prepare reports, answer standard questions, process transactions and coordinate simple activities.</p><p>AI can increasingly perform many of these tasks.</p><p>Humans will still be here. Human creativity, leadership, judgement, relationships and responsibility will remain extremely important.</p><p>But many routine activities that require human time today will increasingly be done by machines.</p><h3>Listen to the Investments, Not Only the Speeches</h3><p>This is where I see an interesting contradiction.</p><p>Technology executives increasingly talk about AI risks and sometimes even about slowing development.</p><p>At the same time, the same industry is making enormous investments in AI infrastructure.</p><p>OpenAI, SoftBank, Oracle and MGX announced Stargate with an intention to invest up to <strong>$500 billion over four years</strong> in US AI infrastructure.</p><p>Elon Musk’s AI infrastructure around Colossus in Memphis is moving in the same direction. The company says its plan is to equip the facility with up to <strong>one million GPUs</strong>.</p><p>Meanwhile, global corporate AI investment reached about <strong>$582 billion in 2025</strong>, according to Stanford’s 2026 AI Index.</p><p>So we should ask a simple question.</p><p>If the industry really expects AI to slow down, why is everyone building so much infrastructure?</p><p>To me, some of the discussion about slowing AI looks less like the end of a war and more like a <strong>ceasefire during a war</strong>.</p><p>You stop for a moment. You reorganise. You build capacity. You prepare for the next move.</p><p>Your competitors are doing exactly the same thing.</p><h3>AI Is Also a Geopolitical Race</h3><p>Another reason AI will not stop is competition between countries.</p><p>The United States and China clearly see AI as a strategic technology.</p><p>The US still dominates private AI investment. Stanford estimates US private AI investment reached $285.9 billion in 2025, more than 23 times China’s reported private investment, although that comparison does not fully capture China’s state-backed investment. At the same time, Chinese AI models have rapidly closed much of the performance gap with leading US models.</p><p>This makes a coordinated global slowdown extremely difficult.</p><p>If the US slows down, it will worry that China will move faster.</p><p>If China slows down, it will have exactly the same concern about the US.</p><p>Neither side wants to become dependent on the other for such a strategic technology.</p><h3>And What About Europe?</h3><p>Europe is in a difficult position.</p><p>It has taken an important global role in AI regulation through the EU AI Act. Some requirements for general-purpose AI models already apply, with further enforcement now coming into effect.</p><p>Regulation is necessary.</p><p>But regulation cannot replace innovation.</p><p>Europe risks becoming better known for regulating AI than for creating the companies, models, chips and infrastructure behind it.</p><p>The gap is becoming difficult to ignore. Even European leaders are increasingly discussing the danger of depending on American and Chinese AI technology.</p><p>Europe needs both: <strong>good regulation and serious technology investment.</strong></p><h3>Automation Is Already Moving Into the Physical World</h3><p>AI is also moving beyond software.</p><p>China’s Pony.ai, for example, is already operating hundreds of Level 4 autonomous trucks, with some operating without human drivers.</p><p>Japan is developing something even more interesting.</p><p>Its <strong>Automated Logistics Road</strong> programme aims to create dedicated infrastructure where goods can move automatically without drivers. The long-term concept includes a Tokyo–Osaka freight corridor, while Japan is already running experiments and supporting autonomous-truck deployment to address its serious shortage of drivers.</p><p>Imagine what this means.</p><p>A warehouse receives an order automatically.</p><p>AI plans the shipment.</p><p>Robots prepare the goods.</p><p>Autonomous systems move the cargo hundreds of kilometres.</p><p>Another automated facility receives it.</p><p>Very little human intervention may eventually be required.</p><p>This is not science fiction anymore.</p><h3>AI Will Not Stop</h3><p>Every major technology transition creates fear.</p><p>The steam engine changed physical work.</p><p>Electricity changed factories.</p><p>Cars changed transportation.</p><p>Computers changed offices.</p><p>The internet changed communication and commerce.</p><p>AI will change knowledge work.</p><p>There will be problems. There will be regulation. There will be mistakes. There will certainly be jobs that disappear and new jobs that emerge.</p><p>But I do not believe AI development itself will stop.</p><p>The economic incentives are too large.</p><p>The investments are too large.</p><p>And the geopolitical competition is too important.</p><p>The more useful question is therefore not:</p><p><strong>“How do we stop AI?”</strong></p><p>It is:</p><p><strong>“How do we prepare ourselves, our companies and our societies for a world where AI is everywhere?”</strong></p><p>That is the conversation we should be having.</p><p>#ArtificialIntelligence #AI #FutureOfWork #TechnologyTrends #Automation #DigitalTransformation #Innovation #TechStrategy</p><p><em>Disclaimer: These are my personal views and do not represent the views or strategies of any current, previous or future employer.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c8b9b2c12159" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How AI Is Reshaping Banking Industry from the Inside Out!]]></title>
            <link>https://medium.com/@eyupkorkmaz/how-ai-is-reshaping-banking-industry-from-the-inside-out-c0eb11e67fff?source=rss-63716835eb64------2</link>
            <guid isPermaLink="false">https://medium.com/p/c0eb11e67fff</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[banking-transformation]]></category>
            <category><![CDATA[banking]]></category>
            <dc:creator><![CDATA[Eyup Korkmaz]]></dc:creator>
            <pubDate>Sun, 06 Sep 2026 22:36:37 GMT</pubDate>
            <atom:updated>2026-09-06T22:36:37.013Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rSHNIDFhDwrK6zZ3NyV3Yg.jpeg" /><figcaption>credit to freepik</figcaption></figure><p>Banking is changing faster than ever. For millions of people, a mobile phone is now the primary bank branch. We rarely step into physical buildings to transfer money, apply for loans, or check balances.</p><p>At the same time, keeping customer loyalty has become a real challenge. Fintechs, payment apps, embedded finance platforms, and independent AI shopping assistants now sit directly between banks and their users. When third-party apps handle day-to-day payments or compare rates automatically, traditional banks risk becoming background infrastructure.</p><p>Artificial intelligence is quickly becoming part of everyday banking. A few years ago, most banks used AI for limited tasks such as chatbots, fraud alerts, or basic data analysis. Today, the role of AI is much bigger.</p><p>Banks are now using AI both internally and in the way they engage with customers. The next stage is not only about making existing processes faster. It is about changing how banking work is done.</p><p>To stay relevant, banks are moving beyond digital-first services toward hyper-personalized banking. The engine driving this shift is artificial intelligence.AI Is Becoming Part of Internal Banking Operations</p><p>Banks handle large amounts of data, documents, transactions, and regulatory requirements every day. Many of these activities involve repetitive work and manual checks. AI is helping banks automate more of these processes.</p><p>For example, AI can extract information from documents, check data, and help process customer applications. In lending, AI can support the initial review of a loan application before it reaches a human decision-maker.</p><p>AI is also being used in areas such as transaction monitoring, fraud detection, compliance, customer onboarding, and internal reporting.</p><p>The biggest change is the move towards agentic AI.</p><p>Traditional AI tools mainly provide answers or recommendations. Agentic AI can go further. It can complete several connected tasks with limited human involvement.</p><p>Imagine a digital assistant that receives a customer request, collects the required information, checks documents, updates internal systems, performs initial risk checks, and then sends the case to the right employee for approval.</p><p>This could significantly reduce manual effort and improve processing times.</p><p>However, this does not mean people will disappear from banking operations. Human oversight will remain important, especially for decisions involving credit, risk, compliance, and customer impact.</p><h3>AI Is Also Changing Customer Engagement</h3><p>The second major area of change is customer experience.</p><p>Mobile banking has already become the main way many customers interact with their bank. The next step is making that experience much more personal.</p><p>Historically, banks grouped customers into broad categories and offered similar products to each group. AI allows banks to understand customers at a much more individual level.</p><p>It can analyse spending patterns, income, savings, borrowing, investments, and previous interactions. Based on this information, banks can provide more relevant suggestions.</p><p>For example, a banking app may notice that a customer is spending more than usual and warn them before they face a cash-flow problem. It could suggest a better savings plan, recommend a suitable loan, or provide investment ideas based on the customer’s financial situation.</p><p>Instead of customers searching for banking products, banking services could increasingly come to customers at the right moment.</p><p>This is where hyper-personalization becomes important.</p><p>Customers will expect banks to understand their needs and provide useful advice in real time. A generic message sent to millions of customers will become less effective.</p><h3>The Customer Relationship Is Becoming More Competitive</h3><p>There is another challenge for traditional banks.</p><p>Fintech companies, payment platforms, embedded-finance providers, aggregators, and AI assistants are increasingly becoming the interface between banks and customers.</p><p>A customer may use banking products without regularly opening the bank’s own application. In the future, they may even ask an AI assistant to compare accounts, move money, find a loan, or manage investments on their behalf.</p><p>This creates a new risk for banks. They may continue to provide the financial product while losing the direct relationship with the customer.</p><p>For this reason, adopting AI is not only about reducing costs. It is also about staying relevant.</p><h3>The Future Will Be About Trust as Much as Technology</h3><p>AI offers banks a major opportunity to become faster, smarter, and more personal. But banks also need to use it carefully.</p><p>Customer data must be protected. AI decisions need to be explainable. Banks must also manage risks such as bias, incorrect recommendations, fraud, and over-automation.</p><p>The banks that succeed will not simply be the ones that use the most AI.</p><p>They will be the ones that combine technology with trust, strong controls, and good human judgement.</p><p>AI is changing banking from the inside out. The transformation has already started, and the next few years are likely to change both how banks operate and how customers experience banking.</p><p>#Banking #ArtificialIntelligence #Banking #DigitalBanking #FinTech #AgenticAI #BankingTransformation</p><p><em>Disclaimer: All views expressed in these posts are solely my own and do not represent the official positions, strategies, or opinions of any current, former, or future employer.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c0eb11e67fff" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Humanoid Robots: From AI Software to a New Physical Economy]]></title>
            <link>https://medium.com/@eyupkorkmaz/humanoid-robots-from-ai-software-to-a-new-physical-economy-e3c98bc81cb3?source=rss-63716835eb64------2</link>
            <guid isPermaLink="false">https://medium.com/p/e3c98bc81cb3</guid>
            <category><![CDATA[robotics-automation]]></category>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[humanoid-robot]]></category>
            <category><![CDATA[robots]]></category>
            <dc:creator><![CDATA[Eyup Korkmaz]]></dc:creator>
            <pubDate>Sun, 30 Aug 2026 21:17:12 GMT</pubDate>
            <atom:updated>2026-08-30T21:17:12.175Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xnx6d9SA2v4DJFthF8XJLw.jpeg" /><figcaption>Credit to Getty Images</figcaption></figure><p>For decades, robots were mainly fixed machines working behind safety fences. Humanoid robots change that model. They are designed to work in environments built for people: factories, warehouses, offices and, eventually, our homes.</p><p>Elon Musk has offered one of the boldest forecasts: at least <strong>10 billion humanoid robots by 2040</strong>, costing around <strong>$20,000–$25,000 each</strong>. That would mean more humanoid robots than people on Earth. I see this as a long-term vision rather than a base-case forecast. Goldman Sachs estimates a <strong>$38 billion market by 2035</strong>, while Morgan Stanley sees a much larger ecosystem of around <strong>$5 trillion by 2050</strong>, with more than 1 billion humanoids in use.</p><h3>Where will humanoids start?</h3><p>The first large market will not be the home. It will be <strong>manufacturing and logistics</strong>, where the environment is structured and the return on investment is easier to measure.</p><p>Initial jobs will include moving materials, loading and unloading, parts sequencing, machine tending, sorting, inspection and repetitive assembly. Automotive, electronics, semiconductor manufacturing and warehouses are natural early segments. Dangerous and physically demanding work will also be attractive. Morgan Stanley expects industrial and commercial applications to represent around 90% of humanoids even by 2050.</p><p>I expect two robot designs to coexist. <strong>Bipedal humanoids</strong> will matter where robots must use stairs, human workstations and existing tools. <strong>Wheeled humanoids</strong> may win many flat-floor jobs because they can be simpler, more stable and more energy-efficient. Home robots will follow, but they need much higher safety, reliability and autonomy. 1X is already testing this direction with NEO, while Sanctuary AI has also explored wheeled designs based on industrial customer feedback.</p><h3>Semiconductors: the brain and muscle control</h3><p>Humanoid robots are physical AI systems, and semiconductors sit at the center of almost every function.</p><p>AI processors handle vision, language, reasoning and planning. Microcontrollers and real-time processors control the joints. Power semiconductors convert battery power into precise motor movement. Other chips manage cameras, touch and position sensors, batteries, memory, connectivity and safety.</p><p>NVIDIA is building a strong position with <strong>Jetson Thor and Isaac GR00T</strong>. At the same time, <strong>Infineon, Texas Instruments and STMicroelectronics</strong> are developing motor-control, sensing and power solutions for humanoids. As robots add more joints, cameras and intelligence, semiconductor content will become a major part of the value chain.</p><h3>Cables: the nervous system people rarely discuss</h3><p>Cables may look like a small part of a humanoid, but they are critical. They run through shoulders, arms, wrists, hands, hips, knees and the torso, carrying power and high-speed data while constantly bending and twisting.</p><p>These cables need to be <strong>thin, light, flexible, resistant to millions of motion cycles and able to keep signal quality in very tight spaces</strong>.</p><p>This creates an interesting opportunity for specialist suppliers. <strong>igus</strong> has its chainflex robot cable family for high-torsion motion. <strong>Proterial</strong> offers robot cables for bending, torsion and small spaces. <strong>OKI</strong> is developing slim high-flex cables aimed at AI robots. <strong>BizLink</strong> already highlights custom cables for humanoids and very high flex-life applications. <strong>TOTOKU</strong> brings ultra-thin, high-strength and flexible wiring. <strong>Gore</strong> focuses on lightweight high-flex automation cables, while <strong>Junkosha</strong> offers thin, lightweight fluoropolymer robotics cables. <strong>LAPP and SAB Bröckskes</strong> are also established suppliers of torsion-rated robotic cables.</p><p>Not every supplier has announced a humanoid design win, but the technical requirements are moving directly toward their areas of expertise.</p><h3>Who is leading today?</h3><p><strong>Tesla</strong> is pushing Optimus toward production. <strong>Figure</strong> is deploying Figure 03 at BMW and scaling its BotQ manufacturing system. <strong>Agility Robotics’ Digit</strong> has moved more than 100,000 totes in a commercial GXO operation. <strong>Boston Dynamics</strong> has moved Atlas into a product phase, with deployments planned at Hyundai and Google DeepMind. <strong>Apptronik</strong> is expanding Apollo 2 and its Robot Park training network with Google DeepMind. In China, <strong>UBTECH</strong> is scaling Walker S2 for industrial use, while <strong>Unitree</strong> is pushing lower-cost humanoid platforms. <strong>1X</strong> is taking a different route with NEO, a home robot offered at a $20,000 early-access price.</p><h3>My view</h3><p>Humanoids will not arrive through one “magic robot.” They will enter the economy <strong>task by task and industry by industry</strong>.</p><p>The opportunity is not only in robot brands. It is also in the ecosystem behind them: <strong>semiconductors, actuators, motors, gears, batteries, sensors, connectors and cables</strong>.</p><p>The numbers to watch are not dancing speed or demonstration videos. They are <strong>productive hours, reliability, cost per hour and real customer deployments</strong>. Recent experience in China also shows that impressive hardware demonstrations do not automatically translate into useful factory work.</p><p>Physical AI is moving from the lab into the real economy. The next decade will show how fast it can scale.</p><p>#HumanoidRobots #Robotics #PhysicalAI #Semiconductors #AI #Automation #Manufacturing #TechnologyTrends</p><p><strong><em>Disclaimer: The views expressed here are my personal views and do not represent the positions of any current, former or future employer.</em></strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e3c98bc81cb3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Sovereign AI Is Not Just a Data Residency Checkbox. It is an Architecture Decision]]></title>
            <link>https://medium.com/@eyupkorkmaz/sovereign-ai-is-not-just-a-data-residency-checkbox-it-is-an-architecture-decision-cb208701d620?source=rss-63716835eb64------2</link>
            <guid isPermaLink="false">https://medium.com/p/cb208701d620</guid>
            <category><![CDATA[enterprise-architecture]]></category>
            <category><![CDATA[genai]]></category>
            <category><![CDATA[sovereign-ai-architecture]]></category>
            <dc:creator><![CDATA[Eyup Korkmaz]]></dc:creator>
            <pubDate>Sun, 16 Aug 2026 14:25:25 GMT</pubDate>
            <atom:updated>2026-08-16T14:25:25.832Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_BOU0tQbYvIwjNbPd5pCAg.jpeg" /></figure><p>A few years ago, when enterprises talked about AI, the conversation was simple: pick an API, send your data, get an answer back. Fast, cheap to start, easy to demo to the board.</p><p>That era is ending.</p><p>Walk into any large enterprise today, a bank, a hospital group, a government agency, a telecom operator, and you’ll hear a different set of questions. Where does our data actually go when we call this model? Who can see it? Can we move this workload if a vendor changes their terms, or if a region tightens its rules? What happens to our costs when ten thousand employees start using this every day instead of ten?</p><p>These aren’t compliance footnotes anymore. They’re becoming the main event. And that’s why “sovereign AI” is showing up in almost every serious enterprise conversation I’m having right now, not as a regulatory buzzword, but as a real architecture decision that sits right next to your budget, your security posture, and your competitive strategy.</p><p><strong>From experimentation to infrastructure</strong></p><p>Most companies started their AI journey the easy way: call a cloud API, test a use case, show some quick wins. That’s a fine way to learn. It’s a risky way to run a business.</p><p>Once AI moves from a pilot into something your teams actually depend on every day, three problems show up that never mattered much during the demo phase.</p><p>First, cost. Paying per token feels cheap when you’re testing with a handful of users. It stops feeling cheap when usage scales into the millions of requests a month. Suddenly finance wants to know why the AI bill looks like a second cloud bill nobody budgeted for.</p><p>Second, latency. Some use cases: a factory floor, a trading desk, a piece of medical equipment, simply can’t wait for a round trip to a data center on another continent. The answer needs to come from somewhere close.</p><p>Third, and most important: control. If your customer data, your product designs, or your patient records are flowing through someone else’s infrastructure, in someone else’s country, under someone else’s terms — how much do you really control your own business?</p><p>This is where sovereignty stops being a legal checkbox and becomes a design question: where should the model actually run?</p><p><strong>Cloud, edge, or on-premises isn’t a competition: It is a mix</strong></p><p>Here’s the mistake I see leaders make: treating this as a single choice, like picking a phone plan. Cloud versus edge versus on-premises. Pick one, move on.</p><p>In practice, the companies getting this right are building a mix, and matching the workload to the place that makes sense for it. A customer chatbot answering general questions can comfortably live in the public cloud. A model reading sensitive contracts or medical scans probably needs to run inside your own walls, or in a sovereign cloud in your own country. A model helping a robot arm make a split-second decision on a factory floor needs to run right there, on the device or the local edge server, with no network round trip at all.</p><p>This is really a hybrid problem. You need one platform that can move workloads between these environments without your teams having to rebuild everything each time. Patterns like Red Hat OpenShift and hybrid cloud orchestration exist for exactly this reason — not to force everything into one place, but to give you a consistent way to run AI wherever it needs to run, and move it later if your needs change.</p><p><strong>Cloud, Edge, or On-Premises: It Is Not a Competition</strong></p><p>The biggest mistake leaders make is treating this as an all-or-nothing choice. Cloud versus edge versus on-premises.</p><p>In practice, the organisations getting this right build a hybrid mix, matching the specific job to the right environment:</p><ul><li><strong>Public Cloud:</strong> Great for general customer chatbots or heavy, non-sensitive model training.</li><li><strong>On-Premises or Sovereign Cloud:</strong> Necessary for models processing sensitive customer data, financial records, or IP inside national borders.</li><li><strong>The Edge:</strong> Essential for local sensors, devices, or plant hardware that need split-second decisions with zero network delay.</li></ul><p>This is a classic hybrid problem. You need a single, open platform, using enterprise patterns like Red Hat OpenShift, to manage workloads across these environments seamlessly. That way, your teams build a solution once and deploy it anywhere, without rewriting code every time business requirements shift.</p><p><strong>Sovereignty as strategy, not just as a rule</strong></p><p>Regulation is part of the story, the EU AI Act and similar rules are pushing companies to know exactly where their data and models sit. But the smarter executives I talk to aren’t asking “<strong>how do we stay compliant</strong>.” They’re asking a bigger question: what happens to our business if we become fully dependent on one vendor, one model, or one region?</p><p>That’s a strategic risk, not just a legal one. If your entire AI capability lives inside one hyperscaler’s infrastructure, you’ve quietly handed them enormous leverage over your pricing, your roadmap, and your data. Recent research backs this up, a large share of executives now say AI sovereignty has to be part of their core business strategy, not an IT afterthought.</p><p><strong>The practical takeaway</strong></p><p>If you’re leading technology decisions right now, the question worth asking isn’t “cloud or edge.” It’s this: for each part of our business, where should intelligence actually run, and do we have the freedom to change our mind later without starting from scratch?</p><p>Get that architecture right, and sovereignty stops being a constraint. It becomes the thing that gives you options everyone else has already given away.</p><p>#GenAI #EnterpriseArchitecture ##SovereignAI #HybridCloud #EnterpriseAI #CloudArchitecture #EdgeComputing</p><p><strong><em>Disclaimer:</em></strong><em> All views expressed in these posts are solely my own and do not represent the official positions, strategies, or opinions of any current, former, or future employer.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cb208701d620" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cooling Is Becoming the Next Strategic Layer of AI Innovation]]></title>
            <link>https://medium.com/@eyupkorkmaz/cooling-is-becoming-the-next-strategic-layer-of-ai-innovation-ef485791032e?source=rss-63716835eb64------2</link>
            <guid isPermaLink="false">https://medium.com/p/ef485791032e</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[cooling-system]]></category>
            <category><![CDATA[electrical-vehicles]]></category>
            <category><![CDATA[hbm4]]></category>
            <category><![CDATA[data-center]]></category>
            <dc:creator><![CDATA[Eyup Korkmaz]]></dc:creator>
            <pubDate>Sat, 27 Jun 2026 13:27:03 GMT</pubDate>
            <atom:updated>2026-06-27T13:27:03.791Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RNd9eLMna4Q9axlRHJVR_g.jpeg" /><figcaption>credit to freepik</figcaption></figure><p><strong>From Room Cooling to Chip-Level Cooling</strong></p><p>For decades, data center cooling was mostly a facilities topic. The main challenge was to keep the room at the right temperature. Air conditioning, airflow design, hot aisles and cold aisles were enough to protect servers and maintain uptime.</p><p>That world is changing very quickly.</p><p>With AI, advanced packaging, high-bandwidth memory and heterogeneous chip architectures, the thermal problem is no longer only at the room level. It is moving closer to the rack, then to the server, then to the package, and now increasingly to the chip itself. Cooling is no longer a back-end infrastructure issue. It is becoming a front-end design constraint for performance, reliability and total cost of ownership.</p><p><strong>Why Semiconductor Companies Are Investing in Cooling</strong></p><p>This is why the recent interest of semiconductor and electronics companies in cooling and immersion-cooling companies is so important. Cooling may not be their traditional core business, but it is becoming directly connected to their core products. A faster AI accelerator, a denser HBM stack, or a more integrated chiplet system cannot deliver its full value if heat is not managed properly.</p><p>In the AI data center, performance per watt is now one of the most important metrics. AI chips are consuming more power, racks are becoming denser, and hyperscalers are pushing infrastructure to limits that traditional air cooling cannot easily support. Direct-to-chip liquid cooling and immersion cooling are therefore moving from experimental technologies to strategic infrastructure options.</p><p><strong>Cooling Is Now Part of the Performance Stack</strong></p><p>This is not only about avoiding overheating. It is about unlocking performance. When chips run too hot, they throttle. When memory gets too hot, bandwidth and reliability can be affected. When thermal headroom is limited, system designers must compromise. In other words, cooling is becoming part of the performance stack.</p><p>High-bandwidth memory is a good example. HBM has become critical for AI because large models require massive memory bandwidth. As the industry moves from HBM3E to HBM4 and beyond, the thermal challenge becomes more complex. HBM is vertically stacked, tightly integrated with logic, and placed close to high-power compute dies. This creates localized hot spots that cannot be solved only by cooling the surrounding air. The package itself needs better thermal paths, better materials and smarter system-level cooling.</p><p><strong>HBM4 and the Rise of Thermal Co-Design</strong></p><p>This is why the boundary between semiconductor design and thermal engineering is becoming thinner. In the past, chip companies could focus mainly on transistor density, process technology, architecture and packaging. Today, they also need to think about how heat moves through the package, the board, the server and the rack. Thermal design is becoming a co-optimization problem.</p><p><strong>Electrical Vehicles: A Different Cooling Challenge</strong></p><p>The same trend is visible in electric vehicles, but with a different set of requirements. EVs already include complex thermal systems for batteries, power electronics and motors. Now, with advanced driver-assistance systems, autonomous driving features, central compute architectures and chiplet-based processors, vehicle electronics are becoming more like mobile data centers.</p><p>However, EV cooling is not the same as data center cooling. A data center is a controlled environment with relatively stable operating conditions. A vehicle operates in the real world: summer heat, winter cold, vibration, dust, humidity, limited space and strict safety requirements. Automotive cooling must be compact, reliable, cost-efficient and serviceable over many years. As chiplets enter automotive compute platforms, thermal management will need to address not only total heat, but also uneven heat distribution across multiple dies and packages.</p><p><strong>Edge AI Brings Cooling to Consumer Devices</strong></p><p>Consumer electronics are another important frontier. Smartphones, laptops, AR glasses and edge AI devices are gaining more AI capability. But these products have very different cooling constraints. There is no large chiller, no server rack and no continuous liquid loop. The device must fit in a pocket, remain comfortable to touch, operate silently and preserve battery life.</p><p>This creates a new challenge: how do we bring more AI performance to the edge without creating unacceptable heat? The answer will likely include better materials, vapor chambers, graphite layers, package-level thermal innovation, power-aware AI models and hardware-software co-design. Edge AI will not scale only through better chips. It will also require better thermal systems.</p><p><strong>Cooling as a New Investment Theme</strong></p><p>The broader message is clear: cooling is becoming a strategic investment theme across the semiconductor value chain.</p><p>For investors, this creates opportunities beyond traditional chip companies. Cooling-fluid suppliers, cold-plate designers, immersion-cooling specialists, thermal-interface-material companies, advanced packaging firms, sensors, simulation software and system integrators may all become more important. For semiconductor companies, investing in or partnering with cooling specialists is a way to protect future product roadmaps. For data center operators, it is a way to manage power density and operating cost. For automotive and consumer electronics companies, it is a way to deliver more intelligence in smaller and more constrained environments.</p><p><strong>The Next Bottleneck Is Heat</strong></p><p>The AI era is often described as a race for more compute. But compute is not only limited by silicon. It is limited by power, memory, packaging and heat.</p><p>Cooling used to be something that happened after the chip was designed. Now, it must be considered before the chip is finalised.</p><p>The winners in AI and semiconductors will not only be the companies that design the fastest processors. They will be the companies that understand the full system: silicon, memory, package, board, server, rack, vehicle and device.</p><p>In this new environment, thermal management is no longer an operational detail. It is becoming a competitive advantage.</p><p>#CloudComputing #TechStrategy #EnterpriseAI #GenerativeAI #memory #HBM #cooling</p><p><strong><em>Disclaimer:</em></strong><em> All views expressed in these posts are solely my own and do not represent the official positions, strategies, or opinions of any current, former, or future employer.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ef485791032e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Memory Compression Algorithms: The Next Major Battleground for AI Infrastructure]]></title>
            <link>https://medium.com/@eyupkorkmaz/memory-compression-algorithms-the-next-major-battleground-for-ai-infrastructure-11c15a766d04?source=rss-63716835eb64------2</link>
            <guid isPermaLink="false">https://medium.com/p/11c15a766d04</guid>
            <category><![CDATA[genai]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[hbm4]]></category>
            <category><![CDATA[memory-management]]></category>
            <dc:creator><![CDATA[Eyup Korkmaz]]></dc:creator>
            <pubDate>Sat, 23 May 2026 18:52:26 GMT</pubDate>
            <atom:updated>2026-05-23T18:52:26.997Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PjQIeNwvYRuTg2cmcneYdw.png" /><figcaption>credit to gemini</figcaption></figure><p>As Large Language Models (LLMs) rapidly scale toward trillion-parameter architectures and ultra-long context windows, memory bandwidth and HBM (High Bandwidth Memory) capacity are becoming the biggest bottlenecks in AI infrastructure. The industry is now shifting focus from only increasing compute performance to improving memory efficiency through advanced memory compression algorithms.</p><p>This trend is becoming strategically important for both AI training and inference workloads.</p><p>From a business perspective, the challenge is clear. AI infrastructure costs are increasing significantly due to GPU shortages, expensive HBM memory, rising power consumption, and the need for larger inference clusters. Enterprises want larger models, more concurrent users, and longer context windows, but these requirements directly increase memory demand and infrastructure spending.</p><p>This is where memory compression algorithms become commercially critical.</p><p>For inference workloads, several techniques are emerging as industry standards:</p><p>• Low-bit KV (Key-Value) quantisation using 2–4 bits</p><p>• Token eviction and learned eviction methods</p><p>• Grouped Query Attention (GQA)</p><p>• Hierarchical memory architectures</p><p>• Vector-based compression methods such as Google TurboQuant</p><p>These approaches directly reduce GPU memory bandwidth usage and optimise KV cache efficiency. Some vector-based techniques can already achieve 5–6x memory reduction by compressing KV cache entities down to roughly 3–3.5 bits per value without retraining models.</p><p>• The benefits are significant:</p><p>• Lower inference cost per token</p><p>• Longer context windows</p><p>• Higher concurrent user density</p><p>• Better GPU and HBM utilisation</p><p>• Improved scalability for enterprise AI deployment</p><p>However, there are trade-offs. Aggressive quantisation can reduce model accuracy and reasoning quality. Token eviction methods may negatively affect long-context consistency if not carefully tuned. Hierarchical memory architectures also introduce software complexity and latency management challenges.</p><p>For AI training workloads, a different set of techniques is becoming dominant:</p><p>• Low-rank methods such as LoRA and tensor decomposition<br> • Mixture of Experts (MoE) architectures<br> • Structured sparsity approaches</p><p>These methods optimise compute and memory efficiency by reducing active parameters while remaining largely compatible with existing GPU infrastructure. This is extremely important commercially because enterprises cannot constantly redesign their entire data centre architecture every 12–18 months.</p><p>One of the most interesting industry observations is that compression is unlikely to reduce total AI infrastructure demand.</p><p>Historically, every major efficiency gain in compute, networking, or memory has accelerated adoption rather than lowering aggregate infrastructure spending. The same pattern is now happening in AI. Compression technologies are making larger models and longer contexts economically feasible, which ultimately increases overall GPU, HBM, and DRAM consumption.</p><p>In other words, compression improves economic scalability rather than reducing demand.</p><p>Over the next five years, I expect several major trends:</p><p>1. KV cache compression will become a standard feature in production inference stacks.</p><p>2. Long-context optimisation will become a competitive differentiator for AI platforms.</p><p>3. Hybrid memory architectures combining HBM, DRAM, SSD, and intelligent caching layers will emerge.</p><p>4. AI accelerator vendors will increasingly co-design hardware with compression-aware software stacks.</p><p>5. Trillion-parameter models will become commercially deployable because compression reduces operational costs.</p><p>Technically, memory compression is becoming essential because compute scaling alone is no longer sufficient. Memory bandwidth limitations are now one of the primary constraints in AI system performance.</p><p>Commercially, memory compression enables enterprises to deploy larger AI systems with better economics, lower latency, and higher utilisation rates. This directly impacts cloud profitability, AI infrastructure ROI, and enterprise adoption speed.</p><p>In my opinion, memory compression algorithms will become one of the most strategically important layers of the AI infrastructure stack over the next decade.</p><p><strong><em>Disclaimer:</em></strong><em> All views expressed in these posts are solely my own and do not represent the official positions, strategies, or opinions of any current, former, or future employer.</em></p><p>#CloudComputing #TechStrategy #EnterpriseAI #GenerativeAI #memory #HBM</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=11c15a766d04" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The 6G-AI Convergence: Building the First Truly “AI-Native” Global Infrastructure]]></title>
            <link>https://medium.com/@eyupkorkmaz/the-6g-ai-convergence-building-the-first-truly-ai-native-global-infrastructure-cf1eeb7e27cb?source=rss-63716835eb64------2</link>
            <guid isPermaLink="false">https://medium.com/p/cf1eeb7e27cb</guid>
            <category><![CDATA[physical-ai]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[6g]]></category>
            <dc:creator><![CDATA[Eyup Korkmaz]]></dc:creator>
            <pubDate>Sun, 10 May 2026 18:27:25 GMT</pubDate>
            <atom:updated>2026-05-10T18:27:25.106Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Pzb5BGuDniO5F69W1dj5ug.jpeg" /><figcaption>credit to gemini</figcaption></figure><p>The digital landscape is currently witnessing a collision of two massive tectonic plates: the maturation of Generative AI and the architectural birth of 6G. While 5G was designed to connect things, 6G is being built to connect <em>intelligence</em>.</p><p>We are moving away from a world where AI lives in centralized data centers, separated from the network by latency. Instead, we are entering the era of <strong>AI-Native Telecommunications</strong>, where the network itself becomes the brain.</p><p><strong>Beyond Speed: The Structural Benefits of 6G</strong></p><p>To understand 6G, we must look beyond “faster downloads.” The true leap lies in three pillars: <strong>Sub-millisecond latency, Terahertz (THz) frequencies, and Native Intelligence.</strong></p><p>Unlike its predecessors, 6G is being designed with AI at the Protocol Layer. This means the network doesn’t just transport data; it predicts traffic patterns, heals its own signal interference, and dynamically allocates compute resources in real-time. For the first time, the “Network as a Sensor” becomes a reality, utilizing radio waves to “see” and map physical environments without the need for cameras.</p><p><strong>The Rise of Intelligent Private Networks</strong></p><p>One of the most transformative shifts will occur in the enterprise sector through <strong>6G Private Networks</strong>. Today’s factories and hospitals struggle with the “hand-off” between Wi-Fi and Cellular.</p><p>In a 6G environment, a smart factory becomes a single, autonomous organism. Private 6G networks will allow for “Micro-Clouds” within a single facility. This enables thousands of sensors and robotic actuators to sync with a local Generative AI model that can orchestrate complex logistics or predictive maintenance on the fly, keeping sensitive proprietary data entirely within the company’s physical walls.</p><p><strong>Physical AI</strong></p><p>In the transition from 5G to 6G, we are moving beyond digital connectivity into the realm of <strong>Physical AI</strong>, where the network acts as a distributed nervous system for the physical world. By integrating <strong>Edge Computing</strong> directly into the 6G architecture, we eliminate the latency of the centralized cloud, allowing AI to interact with physical objects in real-time. This synergy enables “Sensing-as-a-Service,” where 6G radio waves function like a high-resolution radar, allowing the network to perceive the movement, orientation, and even the breathing patterns of humans and machines without cameras. For industries like autonomous logistics and remote healthcare, this means the network isn’t just a pipe for data — it is an intelligent observer and participant that can predict physical collisions before they happen or optimize the trajectory of a robotic arm with microsecond precision. This fusion of the digital and physical layers marks the end of the “dumb pipe” era and the beginning of a world where the infrastructure itself is a living, breathing cognitive entity.</p><p><strong>The New Edge: Drones, Robots, and Smartphones</strong></p><p>The convergence of 6G and AI will fundamentally change the hardware we carry and the machines we deploy.</p><ul><li><strong>Drones &amp; Robots:</strong> Currently, high-resolution visual processing on drones is limited by onboard battery life and heat. 6G’s high-bandwidth, low-latency links allow the “vision” to be processed at the network edge. A drone or a mobile robot can now utilize a 100-billion parameter vision model to navigate complex environments in real-time, effectively outsourcing its “brain” to the nearest 6G cell tower.</li><li><strong>Smartphone Evolution:</strong> We are reaching the physical limits of what a pocket-sized device can process. 6G will usher in the “Thin Client” revolution for consumers. Your smartphone’s camera will no longer rely solely on its internal ISP (Image Signal Processor). Instead, 6G will allow “Distributed Computing,” where the phone captures raw data and uses the massive GPU power of the network edge to render professional-grade cinematic video or complex AR overlays instantly.</li></ul><p><strong>What 6G Brings to Consumers and Enterprises</strong></p><p>For the <strong>Consumer</strong>, 6G is the gateway to the “Internet of Senses.” We will move from 2D video calls to high-fidelity holographic communication and immersive Extended Reality (XR). Generative AI will act as a real-time intermediary — providing instant voice translation that matches the speaker’s tone or enhancing AR glasses that identify objects and provide context in the wearer’s field of vision.</p><p>For the <strong>Enterprise</strong>, 6G represents the ultimate tool for “Operational Sovereignty.” It removes the trade-off between mobility and power. Architects can walk through a digital twin of a building on a construction site with zero lag; surgeons can perform remote procedures with tactile haptic feedback; and logistics giants can manage “Dark Warehouses” where robots communicate with a level of precision that makes human intervention unnecessary.</p><p><strong>The Strategic Outlook</strong></p><p>As we look toward 2030, the transition to 6G will not be a simple hardware upgrade. It is a fundamental reimagining of how intelligence is distributed across the globe. For leaders and innovators, the goal is no longer just “getting connected” — it is about positioning your organization to tap into a ubiquitous, invisible layer of AI that is always on, always present, and integrated into the very fabric of the air around us.</p><p>The 6G era is not just about a faster internet; it is about the <strong>Internet of Intelligence.</strong></p><p><strong><em>Disclaimer:</em></strong><em> All views expressed in these posts are solely my own and do not represent the official positions, strategies, or opinions of any current, former, or future employer.</em></p><p><strong>#PhysicalAI #GenerativeAI #Agentic #AutonomousSystems </strong>#DigitalTransformation <strong>#EdgeComputing</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cf1eeb7e27cb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Sovereign AI: Why “Geopatriation” is the New Cloud Strategy]]></title>
            <link>https://medium.com/@eyupkorkmaz/sovereign-ai-why-geopatriation-is-the-new-cloud-strategy-c0daaa9ee297?source=rss-63716835eb64------2</link>
            <guid isPermaLink="false">https://medium.com/p/c0daaa9ee297</guid>
            <dc:creator><![CDATA[Eyup Korkmaz]]></dc:creator>
            <pubDate>Fri, 03 Apr 2026 22:12:49 GMT</pubDate>
            <atom:updated>2026-04-03T22:12:49.388Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9Cjnkgy5NwxN4jzYOtBB8g.png" /></figure><p>For over a decade, the cloud strategy was simple: go global, scale fast, and rely on hyperscalers.</p><p>That playbook is now being rewritten.</p><p>We are entering an era where <strong>AI is not just a capability, it’s national infrastructure</strong>. And with that shift comes a new architectural principle: <strong>Sovereign AI</strong> powered by <strong>geopatriation</strong>.</p><p><strong>What is Geopatriation and Why Now?</strong></p><p>Geopatriation is not just cloud repatriation. It’s a strategic move driven by <strong>geopolitics, regulation, and control</strong>.</p><p>At its core, it means:</p><p>Moving AI workloads, data, and systems from global clouds to <strong>local or sovereign environments</strong> to reduce geopolitical risk and maintain jurisdictional control.</p><p>This shift is accelerating fast. Gartner has already identified geopatriation as a <strong>top strategic technology trend</strong>, with enterprises rethinking where their systems run, not for performance, but for <strong>sovereignty and legal authority</strong>.</p><p><strong>Why Sovereign AI Matters in Today’s Geopolitical Environment</strong></p><p>AI is now at the center of global power dynamics.</p><p>We’re seeing:</p><ul><li><strong>US–China AI competition shaping global ecosystems</strong></li><li>Nations investing billions to avoid dependence on foreign AI platforms</li><li>Regulations fragmenting AI deployment across regions</li></ul><p>Countries like South Korea, the UK, and EU nations are actively building <strong>domestic AI infrastructure, models, and chips</strong> to reduce reliance on external powers.</p><p>This is not just about compliance anymore. It’s about:</p><ul><li><strong>National security</strong></li><li><strong>Economic competitiveness</strong></li><li><strong>Control over data, models, and decision systems</strong></li></ul><p>In fact, sovereign AI is increasingly viewed like <strong>energy or defense infrastructure</strong>, too critical to outsource.</p><p><strong>From Global Cloud to Multipolar AI Architecture</strong></p><p>The traditional model:</p><p>Centralized hyperscaler → global deployment</p><p>The emerging model:</p><p><strong>Multipolar, sovereignty-aware architecture</strong></p><p>We’re seeing the rise of:</p><ul><li>Regional cloud zones and sovereign clouds</li><li>Jurisdiction-aware AI pipelines</li><li>Hybrid architectures combining hyperscalers + local providers</li></ul><p>This is what I call the <strong>“Neo-Cloud” evolution 2.0</strong>, but with sovereignty as the primary design constraint.</p><p>Even hyperscalers are adapting. AWS and others are launching <strong>regionally isolated sovereign clouds</strong> to retain customers who demand jurisdictional control.</p><p><strong>Key Trends Driving This Shift</strong></p><ol><li><strong>Regulatory Explosion</strong><br> AI regulations are increasing rapidly across countries, forcing organizations to rethink data and model placement.</li><li><strong>Jurisdictional Risk</strong><br> Where your AI runs determines <strong>which laws apply</strong>, and that’s now a board-level concern.</li><li><strong>Data Gravity + AI</strong><br> AI amplifies data gravity. Moving models is easier than moving sensitive data across borders.</li><li><strong>Resilience &amp; Trust</strong><br> Organizations want infrastructure that remains operational even during geopolitical disruptions.</li><li><strong>Economic Nationalism (AI Nationalism)</strong><br> Countries are prioritizing <strong>local AI ecosystems</strong> to protect long-term competitiveness.</li></ol><p><strong>What This Means for Enterprise Architects</strong></p><p>This is where things get real.</p><p>Sovereign AI is not about abandoning hyperscalers. It’s about <strong>designing for optionality and control</strong>.</p><p>Here’s how I see forward-looking architectures evolving:</p><p><strong>1. Design for Jurisdiction Awareness</strong></p><p>Architect systems that understand:</p><ul><li>Where data resides</li><li>Where models are trained/inferred</li><li>Which laws govern them</li></ul><p><strong>2. Adopt a Federated AI Strategy</strong></p><ul><li>Distributed training/inference</li><li>Regional data pipelines</li><li>Policy-driven orchestration</li></ul><p><strong>3. Build a Sovereign AI Stack</strong></p><p>Think beyond cloud:</p><ul><li>Sovereign data layer</li><li>Sovereign model hosting (or fine-tuned local models)</li><li>Sovereign observability and governance</li></ul><p><strong>4. Embrace Hybrid + Multi-Cloud (Intentionally)</strong></p><p>Not for vendor diversification, but for:</p><ul><li><strong>Geopolitical risk mitigation</strong></li><li><strong>Regulatory compliance by design</strong></li></ul><p><strong>5. Partner with Regional Ecosystems</strong></p><p>Local cloud providers, telcos, and governments will play a bigger role in AI delivery.</p><p><strong>The Bigger Shift: From Efficiency to Control</strong></p><p>For years, cloud decisions were driven by:</p><p>Cost, scale, and speed.</p><p>Now, a new priority is emerging:</p><p><strong>Control, sovereignty, and resilience.</strong></p><p>Geopatriation signals a deeper transformation:</p><ul><li>From <strong>globalization → controlled decentralization</strong></li><li>From <strong>platform dependency → strategic autonomy</strong></li><li>From <strong>cloud-first → sovereignty-first</strong></li></ul><p><strong>Final Thought</strong></p><p>Sovereign AI is not a trend, it’s a structural shift.</p><p>The question is no longer:</p><p><em>“Which cloud should we use?”</em></p><p>But:</p><p><strong>“Where should our AI live and who ultimately controls it?”</strong></p><p>Architects who can answer that question, technically and strategically, will define the next decade of AI.</p><p><em>Disclaimer: All views expressed in these posts are solely my own and do not represent the official positions, strategies, or opinions of any current, former, or future employer.</em></p><p>#SovereignAI #Geopatriation #AIInfrastructure #DataSovereignty #CloudStrategy #AIArchitecture #DigitalSovereignty #MultiCloud #EnterpriseAI</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c0daaa9ee297" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Memory Prices Are Rising Rapidly? What It Means for the AI Era ?]]></title>
            <link>https://medium.com/@eyupkorkmaz/why-memory-prices-are-rising-rapidly-what-it-means-for-the-ai-era-b9c269d6d0aa?source=rss-63716835eb64------2</link>
            <guid isPermaLink="false">https://medium.com/p/b9c269d6d0aa</guid>
            <dc:creator><![CDATA[Eyup Korkmaz]]></dc:creator>
            <pubDate>Sun, 08 Mar 2026 01:19:08 GMT</pubDate>
            <atom:updated>2026-03-08T01:19:08.142Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*98huG42QNqO2AbT1iE2xgw.png" /><figcaption>credit to Gemini</figcaption></figure><p>Over the past few quarters, memory pricing has been rising at a much faster pace than many expected. For companies across semiconductors, cloud infrastructure, AI, and electronics manufacturing, this shift is becoming increasingly important to monitor. Memory has always been cyclical, but the current surge is being shaped by a combination of structural demand changes and supply-side constraints.</p><p>Here are a few key factors driving the rapid increase.</p><p><strong>1. AI Infrastructure Is Consuming Massive Amounts of Memory</strong></p><p>One of the most significant drivers is the explosive growth of AI infrastructure. Training and running large AI models requires enormous amounts of high-performance memory, particularly High Bandwidth Memory (HBM) and advanced DRAM.</p><p>Modern AI accelerators rely heavily on memory bandwidth to feed data to GPUs and specialized chips. As hyperscalers and cloud providers rapidly expand their AI data centers, the demand for these advanced memory technologies has surged. Unlike previous memory cycles driven mainly by PCs or smartphones, AI workloads require significantly higher memory density per system, amplifying demand.</p><p><strong>2. Supply Was Aggressively Cut During the Downturn</strong></p><p>The memory industry experienced a severe correction in 2023, when demand weakened and inventories built up across the ecosystem. In response, major manufacturers significantly reduced production, cut wafer starts, and scaled back capital expenditure.</p><p>Those supply reductions were necessary to stabilize the market at the time. However, when demand rebounded, especially from AI and data centers, the industry suddenly found itself with limited available capacity. This supply discipline is now contributing to the rapid price recovery.</p><p><strong>3. Manufacturing Complexity Is Increasing</strong></p><p>Memory technology transitions are becoming more complex. Moving to newer DRAM nodes or ramping advanced memory types such as DDR5, LPDDR5X, and HBM requires time to improve yields and scale production.</p><p>During these transitions, effective supply can remain tight even if overall capacity exists. Manufacturers must carefully allocate production between legacy memory products and next-generation technologies, which creates temporary shortages in certain segments.</p><p><strong>4. Capacity Is Being Redirected Toward Higher-Value Products</strong></p><p>Another structural change is the shift toward higher-margin memory products. Memory suppliers are prioritizing HBM and server-grade DRAM because they support the rapidly growing AI market and generate stronger returns.</p><p>As a result, less capacity is available for commodity segments such as PC DRAM or consumer devices. This reallocation tightens supply and can accelerate price increases across the broader market.</p><p><strong>Key Risks for Enterprises</strong></p><p>Rising memory prices create several challenges for companies across the technology ecosystem.</p><p>First, cost volatility becomes harder to manage. Memory can represent a significant portion of bill-of-materials costs for servers, AI hardware, and consumer electronics. Rapid price swings can compress margins and disrupt product pricing strategies.</p><p>Second, supply allocation risks may increase. When demand outpaces supply, memory vendors often prioritize strategic customers or higher-value applications, leaving some companies exposed to shortages.</p><p>Third, project timelines may be impacted, particularly for AI infrastructure deployment. If memory supply becomes constrained, scaling new data centers or AI clusters could take longer than planned.</p><p><strong>How Enterprises Can Mitigate the Risk</strong></p><p>While companies cannot control global memory cycles, there are several strategies that can help mitigate exposure.</p><p>Diversify suppliers and technologies. Relying on a single supplier or memory type increases risk. Building a broader supply base can improve resilience.</p><p>Establish longer-term agreements. Strategic procurement contracts can provide greater pricing visibility and supply assurance during volatile periods.</p><p>Improve demand forecasting. Better alignment between engineering teams, procurement, and suppliers can help anticipate memory needs earlier and avoid last-minute sourcing challenges.</p><p>Design for flexibility. Where possible, hardware architectures that support multiple memory configurations can reduce dependency on specific components.</p><p>Build strategic inventory buffers. While excessive stock carries cost, maintaining a modest buffer can help smooth supply disruptions during tight cycles.</p><p><strong>When Could Memory Prices Normalize?</strong></p><p>The short answer: Not anytime soon. While new mega factories are under construction in the U.S. and APAC, meaningful new capacity isnt expected to hit the market until late 2027 or 2028.</p><p>Predicting the exact timing of memory cycles is always difficult. However, several indicators suggest how the market may evolve.</p><p>If memory manufacturers begin expanding capacity again and yield improvements accelerate, supply could gradually catch up with demand. At the same time, some of the initial AI infrastructure buildout may stabilize after the current investment wave.</p><p>Many analysts expect memory pricing pressure to remain elevated in the near term before gradually stabilizing as new supply comes online. Historically, memory cycles tend to rebalance once supply expansion follows strong demand signals.</p><p><strong>The Impact on AI</strong></p><p>Ironically, while AI is a major driver of rising memory demand, it may also feel the impact of these price increases.</p><p>Higher memory costs can raise the total cost of AI hardware, increasing the capital required to deploy large training clusters. This could influence how quickly organizations scale their AI infrastructure or prioritize certain workloads.</p><p>At the same time, the industry may respond with architectural innovation, including more memory-efficient models, improved chip designs, and new memory technologies.</p><p>In that sense, the current memory cycle could accelerate innovation across the AI hardware ecosystem.</p><p>Memory has long been one of the most volatile segments of the semiconductor industry. But this cycle appears to be shaped not only by supply discipline but also by the structural rise of AI-driven computing.</p><p>For companies building the next generation of digital infrastructure, understanding memory market dynamics is becoming more important than ever.</p><p>Curious to hear how others in the semiconductor and AI ecosystem are navigating this shift.</p><p><strong><em>Disclaimer: </em></strong><em>All views expressed in these posts are solely my own and do not represent the official positions, strategies, or opinions of any current, former, or future employer.</em></p><p>#SupplyChain #TechnologyStrategy #DigitalInfrastructure #TechTrends #IndustryInsights #Semiconductors #MemoryMarket #AIInfrastructure #DRAM #NAND #HBM #HBM3E HBM4 #ArtificialIntelligence #AI</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b9c269d6d0aa" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>