Dylan Patel has dropped a bombshell on the infrastructure sector: the idea that a private aerospace company could single-handedly deliver ten gigawatts of data center power by 2027 is not a fantasy, but a mathematically viable strategy driven by unprecedented profit margins in AI inference. While the industry fixates on chip shortages, Patel argues the real bottleneck is power, and the solution lies in a radical, speed-first construction model that bypasses traditional permitting and efficiency constraints. This is not just a forecast; it is a blueprint for how capital flows when the return on investment exceeds one hundred billion dollars per gigawatt.
The Economics of Speed
Patel's central thesis rests on a staggering valuation of compute: "large-scale + near-term compute is a remarkably scarce combination, and it's priced at a huge premium – up to $50B/GW/year." He posits that frontier AI labs are not merely buying hardware; they are purchasing a revenue engine capable of generating over "$100B/GW/year of revenue when selling API inference." This reframing of data centers from cost centers to profit factories explains the frantic pace of investment. The author notes that for companies like OpenAI and Anthropic, the margins are so high that they can absorb massive capital expenditures with ease.
The argument gains teeth when applied to Microsoft. Patel writes, "Satya nailed the negotiations with OpenAI: the deal reworked in April 2026 dropped the old 20% revenue share from the equation." This structural change means Microsoft can now capture the full upside of the inference boom. "Put simply, Microsoft has a giant incentive to procure as many MWs as possible, as fast as possible." The implication is that the tech giant is no longer waiting for traditional construction timelines; it is hunting for capacity that can be deployed in months, not years. Critics might note that such aggressive revenue projections rely on sustained, hyper-growth in AI adoption that could face market saturation, but the current data suggests the demand curve is still vertical.
"In times of compute constraints and extremely high AI token margins, a 500mw cluster available in three months, with 90-day cancellation policy, is one of the most scarce and valuable asset in the world."
The SpaceX Playbook
How does a rocket company build a power grid? Patel suggests the answer lies in abandoning the "cost plus" model for "value-based pricing" and prioritizing speed above all else. He draws a sharp contrast with traditional operators: "Most datacenter operators in the world optimize for efficiency and quality – it's the only way to land a 15-20 year take-or-pay hyperscaler datacenter contract. SpaceX will focus on entirely different tradeoffs: it's speed above all."
The author details a strategy of radical improvisation. When faced with supply chain bottlenecks like a two-year backlog on transformers, the solution is to "buy power modules from China, and skip LPTs by delivering medium voltage power from power gen to low voltage transfos." This approach mirrors the agility seen in historical rapid-build projects, such as the Colossus data center, where the team managed to construct 300 megawatts in just 122 days. Patel points out that SpaceX has already expanded a gas plant in Southaven from 27 turbines to 69 in a matter of months, proving that the "speed vs efficiency" tradeoff is not just theoretical.
The financing model is equally unconventional. Patel argues that SpaceX can afford the capital expenditure because the revenue pays back the cost in less than a year. He suggests a combination of vendor financing from Nvidia and operating cash flow from high-margin leases. "SpaceX will continue to sell large-scale compute with 3-5 months lead time, an unbeatable offering, and price it accordingly at 30-50M/MW/year." This creates a self-funding loop where the speed of deployment generates the cash to fund the next wave of construction. A counterargument worth considering is the risk of relying on a secondary market for gas turbines, which could be volatile, but Patel counters that "secondary market prices are very high, but Elon can pay."
The Microsoft-SPACEX Convergence
The convergence of Microsoft's capital and SpaceX's speed creates a unique market dynamic. Patel observes that Microsoft has already signed binding contracts for 10 gigawatts, totaling over $300 billion, yet faces a "near-term gap to fill." SpaceX is positioned to bridge this gap with a 90-day cancellation policy that removes balance sheet risk for the buyer. "This is remarkably easy for Amy Hood to sign off on, given the revenue opportunity."
The author envisions a scenario where Microsoft's Azure revenue growth accelerates from 42% to over 100% by next year, fueled by this new capacity. The deal structure is described as "insane" on paper but logical in practice: "While much of their datacenter capacity currently goes to OpenAI at ~14M/MW/year, they have the opportunity to improve that mix." The key is the ability to monetize inference at rates that dwarf traditional cloud services. Patel's analysis suggests that the market is ready to pay a premium for immediate availability, effectively decoupling power generation from the slow bureaucratic processes that usually govern the industry.
"All typical 'quality' metrics don't matter."
Bottom Line
Patel's most compelling insight is that the rules of data center construction have fundamentally changed; in an era where AI inference margins are astronomical, speed is the only metric that matters. The argument's greatest vulnerability lies in its assumption that supply chains for gas turbines and power modules can sustain such a hyper-accelerated ramp without collapsing, but the financial incentives described are powerful enough to force the market to adapt. Watch for the next quarter's contract announcements, as they will reveal whether this high-stakes gamble on speed can actually deliver the promised ten gigawatts.