To be honest, recent FUD around Neocloud and debt defaults like CoreWeave’s CDS has led many to once again cry, “AI data centers have peaked” and “compute is oversupplied.” 🧐 Every time there’s a liquidity hiccup, the market rushes in, throwing out the baby with the bathwater—it’s the same old playbook. Today, I just finished reading through @SemiAnalysis_'s latest report, “Datacenters 2027: The Year of Acceleration, +50GW in a Single Year,” cover to cover. I’m re-energized. The conclusion is crystal clear: not only is there no cause for concern about Neocloud’s demand, but the热度 of the data center and compute leasing market far exceeds most people’s imagination. The current consolidation and FUD are, in fact, excellent opportunities to accumulate. Here are the key insights and thoughts I took from the report: 1️⃣ Don’t short based on “intuition”—look at the hard data: Over 50GW of new capacity added in a single year by 2027! Some bears, like Altimeter’s Brad Gerstner, think 43GW of new capacity in 2027 is overly aggressive and predict only 25GW. SemiAnalysis directly counters with satellite imagery and granular data: • The total IT capacity under construction across the world’s top 40 large-scale data center campuses already exceeds 24.3GW. Many of these sites are more than halfway built, with cooling and electrical infrastructure already installed. Note: these 40 campuses represent only a small fraction of all late-stage projects. • Global new capacity added: 5.8GW in 2023, 15.6GW in 2025, projected 25GW in 2026, and a massive 51.2GW in 2027! What does that mean? It’s equivalent to building all the data centers ever constructed globally before 2024—in just one year. Does that look like a peak? 🤔 2️⃣ Trillions in lease agreements are already signed—Neocloud’s thesis is far more resilient than you think! Many worry about CoreWeave’s default, but they’re conflating one company’s leverage timing issues with the entire industry’s leasing demand. Data center developers are pragmatic: without pre-leased customer commitments, they won’t risk major capital expenditure—they typically only commit 5–10% of funding upfront. To secure debt financing from banks, they need long-term leases of 15–20 years (a 1GW contract is worth roughly $30–50 billion). One number in the report made me pause: • Since 2026, over $1 trillion (1.36T, ~13.7GW) in non-cancelable lease agreements have already been executed. • Compare that to just $236 billion signed globally in all of 2024. The tenants are elite: not just AWS, Meta, Microsoft, and Google—but also OpenAI, Anthropic, Nvidia, and AMD—all aggressively securing GW-scale leases, with numerous Neocloud players locking down hundreds of MWs behind them. With leases locked in, financing and construction follow naturally. I view short-term debt market volatility as an overreaction. 3️⃣ Supply chain and power constraints are being brutally solved by “modularization” and “on-site power generation”! Previously, bears loved to argue: U.S. grid interconnection queues take 5–10 years; there’s not enough power or labor. But here’s how the industry is responding now: • Modular prefabrication: AWS, Google, DayOne, and others are pre-assembling electrical and plumbing systems like LEGO blocks—reducing construction timelines from over 18 months to just 9–10 months and cutting on-site labor by 63%. • On-site power generation: If the grid is too slow, build your own! Gas turbines, truck engines, fuel cells—all are being deployed. Over 75GW of on-site generation capacity has already been secured. So the power bottleneck has essentially been broken by technology and capital. 4️⃣ What looks like oversupply is actually a shortage of chips! SemiAnalysis does mention that by end-2028, roughly 10GW of data center capacity may sit idle. But here’s the catch: it’s not because AI demand has faded—it’s because chip production can’t keep up with the pace of data center construction! • The real bottleneck is HBM memory and TSMC’s N3 advanced node. • HBM supply in 2027 is projected to grow by only 27%, far below GPU demand. In short: data centers are being built and waiting for GPUs. Compute remains extremely scarce. Neoclouds holding data center assets and GPU allocations are precisely the scarce assets. To summarize my view: current market panic and emotional selloffs mirror every other tech cycle FUD we’ve seen before. Whether it’s power constraints, construction delays, or leverage noise—none can stop the real-world rollout of a 50GW compute infrastructure wave. Trillions in long-term leases are already on the books. Large models keep evolving—and their appetite for compute will only grow. So my logic is simple: panicking and selling because of CoreWeave CDS issues is almost certainly a strategic mistake. Every emotional selloff creates a prime opportunity to build positions in core compute infrastructure and the Neocloud ecosystem. I’ll continue dollar-cost averaging into $NBIS and $CRWV, as well as affected names like $MU, $AVGO, $MRVL, and $INTC. Of course, high-leverage names are volatile—position sizing must be disciplined. So what do you all think? Let’s discuss your current accumulation targets in the comments. DYOR 🙏 For investing in U.S. stocks, I use the #RWA tokenized U.S. equities platform #MSX—join me in participating in the U.S. market: https://t.co/x4ykUw3WBj Early-stage U.S. stock investors and partners: DM me to fill out a form and gain free access to our U.S. stock discussion community (currently limited to 10 people per week; assistant-reviewed; may take some time—thank you 🙏)!