I hold the diametrically opposite view to Burry. Having reviewed his position, I believe he fundamentally misreads the market dynamics that make AI so valuable and that misreading stems from a lack of technical understanding of what is actually being built. Whatever happens, AI compute will remain scarce for at least the next five years. And the reason is physical, not financial. ENERGY the real bottleneck: • Goldman Sachs estimates a 9.3 GW global power shortfall for compute in 2026, widening to 10 GW by 2028 and that assumes announced projects deliver on time. • GE Vernova's gas turbine backlog just hit 100 GW roughly FIVE YEARS of factory output already committed. Delivery slots are effectively sold out through 2030, with only ~10 GW left across 2029–2030 combined. • Turbine prices are heading toward $600/kW by end-2027 nearly triple 2019 levels. Only three manufacturers on Earth can build these machines, and all three are booked solid. • A data center can be built in 12–24 months. A grid connection in Northern Virginia now takes 5–7 years. You cannot print electrons. • That's why 60%+ of hyperscaler capex now goes into power, cooling, and construction not chips. SILICON the second wall: • On-demand GPU capacity effectively sold out. 36–52 week lead times. TSMC fully booked NVIDIA and Broadcom reportedly asked for more capacity and were turned down. • $650–725B in hyperscaler capex committed for 2026 they'd spend more if they could. Money has become the least scarce input in this equation. DEMAND and this is what Burry misses. Not AI labs buying from each other. The real economy, contractually committed: • Microsoft's commercial backlog: $678B, up 84% YoY signed contracts. Azure past $100B/yr at 43% growth, with demand still exceeding the capacity they can physically build. • Copilot: 30M+ paid enterprise seats; customers buying 50,000+ seats quadrupled in 12 months. • JPMorgan: $19.8B tech budget, 500+ AI use cases in production, an LLM platform serving 230,000+ employees. • Walmart: 72% of its $23B capital budget into AI and automation. 40%+ of new code AI-generated. • Enterprise genAI spend tripled in a year to $37B. 92% of the Fortune 500 are customers. So the equation is simple: demand from the entire real economy, colliding with a supply chain gated by physics electrons, turbines, transformers, wafers. Intelligence is the mother of all commodities, and everyone realized it at once. This is the steam engine and the internet again except the cost-effectiveness gains are orders of magnitude beyond anything before, and the constraint is no longer capital but energy itself. Burry is shorting the barrel while the fish are eating the ocean.

🚀 Grateful for this important milestone! 🚀 Vineland’s Planning Board has approved DataOne’s amended site plan, clearing the way for us to move forward with the completion of our AI Factory campus in New Jersey. We’re grateful to the Board for its careful review and especially to the Vineland community. To every resident who showed up, asked thoughtful and sometimes challenging questions, and helped us shape a stronger project: thank you for your trust and engagement. Being a good neighbor means listening, engaging openly, and continuing the conversation. Our responsibility to Vineland doesn’t end with an approval it continues as we complete construction and move toward full operations. I’m incredibly proud of the talented people who brought us to this milestone and grateful to everyone who has supported this journey. Big kudos to our customer Nebius and to their unwavering support. This is an important step forward for DataOne — and for City of Vineland . Looking forward to what comes next

A big shoutout to our customer Nebius, our incredible DataOne Vineland team and partners who are giving their very best to deliver this next gen AI Factories. Despite the harshest winter we’ve seen in more than two decades frozen ground, feet of snow, and relentless conditions they show up every single day and keep pushing forward, no matter what. It’s truly heartwarming to witness this level of passion, resilience, and energy. I must say, it’s something quite unique. Proud of what we are building. Proud of the people making it happen.

🔥 CDUs Are Dead in Native AI Factories. Yes, I said it. Cooling Distribution Units may still make sense in: • Retrofitted legacy data centers • Space-constrained environments • Mixed IT workloads But in purpose-built AI Factories? They are largely irrelevant. Let’s demystify what a CDU really is. A CDU is essentially: • A plate heat exchanger • Pumps • Filters • Sensors (flow, pressure, temp) • Controls Pre-assembled on a skid. Factory tested. Looks clean on paper. But in reality? • Massive pressure drops • Added hydraulic complexity • High maintenance burden • Long lead times • 2x to 10x markup on commodity components • Design constraints across the entire data hall And most vendors are still deploying 1.5–2.5 MW CDUs. Which means: • Dozens (sometimes hundreds) of additional pumps • Dozens of filters • Increased failure points • Higher CAPEX • Dramatically higher OPEX The Real Question: Why fragment your cooling architecture… When you can centralize it? Instead of distributing small CDUs across the data hall, we: • Remove them entirely • Centralize high-efficiency plate heat exchangers in mechanical rooms • Run larger systems with 1–2°C approach temperature (not 4°C+) • Use 4 large optimized pumps instead of dozens of small ones • Centralize filtration • Use redundant control logic for flow, pressure, and VFD management Each degree of approach temperature matters. Each extra pump matters. Each additional failure point matters. At AI scale, every inefficiency multiplies. People often ask: “How are you moving faster than everyone else?” Simple. We eliminate long lead time items. We remove unnecessary complexity. We challenge legacy “best practices.” We redefine what is truly required without compromising SLA or performance. And yes CDUs are one of the first things to go. Your operations team will thank you. Your CFO will thank you. Your lenders will thank you. And most importantly your customers will thank you. Because when CDUs disappear, modularity increases dramatically. With our native Direct Liquid Cooling architecture, customers can deploy 135kW+ racks almost anywhere in the hall with near-zero layout constraints. CDU removal is only step one. If you want to know the rest you know who to call. 😉

Just to give you a sense of how big our engines really are 🤩 Let’s go the largest Bergen Engines plant on Earth is being built! 🚀 And the cherry on top: this plant will operate at close to net-zero pollution.

This is what AI is truly about and far beyond the noise. At DataOne, it’s the driving force behind how and why we build AI Factories. We’re not building them to flood social media with meaningless content, but to create real, measurable impact.

Many people start talking about a “bubble” as soon as they see P&Ls under pressure. More often than not, it simply means they are confusing the investment phase with the return phase. Every major technological disruption follows the same sequence: massive upfront investment, infrastructure build-out, product deployment, gradual client adoption, development of real use cases… and monetization comes last. Those investments only bear fruit after several quarters, sometimes several years. In the data center industry, this timing is even more pronounced. A data center is traditionally designed and built over 3 to 5 years. At DataOne, we deliver projects in 6 to 18 months — an exception, not the norm. So no, this is not a bubble. This is a massive investment cycle, driven by seemingly unlimited client demand. We are living through the largest civilizational shift humanity has ever experienced. In 2000, there was supply but no demand, due to a lack of telecom infrastructure. Today, it’s the exact opposite: everyone is equipped, everyone is connected, everyone is a potential consumer. Add to that a technology where the marginal cost per token follows a steep logarithmic decline, often approaching a 2× reduction over a few-year horizon, and the conclusion is straightforward: short-term P&Ls are mechanically under pressure. But over the long term, the compounding effect of these investments is massive. This is precisely what Joseph Schumpeter described as creative destruction. Except this time… we are experiencing it in an ultra-accelerated form.

Beaucoup parlent de « bulle » dès qu’ils voient des P&L sous tension. C’est souvent le signe qu’on confond phase d’investissement et phase de rendement. Dans toute rupture technologique majeure, la séquence est toujours la même : on investit d’abord, on construit l’infrastructure, on déploie les offres, les clients arrivent, les use cases se développent… et la monétisation suit. Ces investissements ne portent leurs fruits qu’après plusieurs trimestres, parfois plusieurs années. Dans le monde des datacenters, cette temporalité est encore plus marquée. Un datacenter se conçoit et se construit traditionnellement sur 3 à 5 ans. Chez DataOne, nous livrons des projets en 6 à 18 mois une exception, pas la norme. Alors non, il n’y a pas de bulle. Il y a une phase d’investissement massif, portée par une demande client pratiquement sans limite. Nous vivons le plus grand changement civilisationnel que l’humanité ait jamais connu. En 2000, il y avait de l’offre, mais pas de clients, faute d’infrastructures télécom adaptées. Aujourd’hui, c’est l’exact inverse : tout le monde est équipé, connecté, tout le monde est un consommateur potentiel. Ajoutez à cela une technologie dont le coût marginal du token suit une baisse logarithmique, souvent proche d’un facteur 2 à l’échelle de quelques années, et le constat est simple : le P&L est mécaniquement sous pression à court terme. Mais à long terme, l’effet composé de ces investissements est colossal. C’est ce que Joseph Schumpeter appelait la destruction créatrice. Sauf qu’ici, nous en vivons une version ultra-accélérée.

My 2 cents on the current market conditions: CAC40 goes back to 4200 and S&P goes back to 2600, if it breaks those limits you ll see an even greater panic and Central Banks will start buying stocks to calm the turmoil, 50% of retail money on the markets has never been a good sign

Why moving to Paris should be your next career move by @ErikaBatista medium.com/p/why-moving-to-p…

But whatever happen you will still need Datacenter, this is the beauty of our business ;) lnkd.in/gurwFD2

"We’re thrilled to be shortlisted as Technology Provider of the Year for the second consecutive year. BSO was a fina…lnkd.in/gvFSqrU

Are you trading in China?.jpg lnkd.in/dn34MRe

Dear Santa, if you find some time, could you please take a look at our current job openings (lnkd.in/d8KMh52…lnkd.in/fWTyhDH

Always more of the Etix Spirit come and join us we need you !! :) lnkd.in/d-sBNa2