Azure Crosses $100 Billion as Microsoft Emerges as SpaceX's Largest AI Power Offtaker
Microsoft's fiscal 2026 results and a reported SpaceX power arrangement show how the AI infrastructure race has shifted from chip procurement to electricity and long-term contracts — while a heavy reported dependence on OpenAI revenue introduces a structural concentration risk.
A report published this week by SemiAnalysis contends that SpaceX's pledge to deploy 10 gigawatts of AI infrastructure power capacity by 2027 is credible, and that Microsoft will be the arrangement's largest offtaker — a pairing that, if it materializes, would embed the rocket company as a significant node in the hyperscale fabric Azure is racing to extend worldwide. The detail matters not merely for its scale but for what it reveals about where the AI buildout now stands: having largely worked through near-term chip supply as the binding constraint, the frontier of competition has migrated to electrons, acres, and contracts signed years in advance.
Microsoft's fiscal year 2026 results, reported on July 30, provide the financial scaffolding for understanding why a commitment of that magnitude is plausible. Full-year revenue reached $331 billion, up 18 percent year-on-year, with Azure crossing $100 billion in annual revenue for the first time — a milestone the division took roughly a decade of double-digit growth to reach. The company's commercial remaining performance obligation stood at $678 billion at quarter close, locking in a long runway of contracted demand. In the same fiscal year, Microsoft brought 88 data centers online globally and disclosed more than $130 billion in new data center lease commitments, pushing total future lease obligations from $196 billion to $329 billion in a single year.
That infrastructure posture is matched by capital allocation to scale. Analysts citing Morgan Stanley place Microsoft's cumulative capex commitment at $175 billion, part of a broader wave that Morgan Stanley itself forecasts will lift global cloud infrastructure spending to $1.2 trillion by 2027 — a figure 30 percent above prior estimates. To fill that physical envelope, CEO Satya Nadella confirmed in late July that Azure will deploy both AMD Helios and NVIDIA Vera Rubin in production AI workloads, a deliberate multi-vendor silicon strategy. In parallel, Microsoft was reported to be testing Kimi K3 — a Chinese open model — for use in Copilot with the goal of reducing inference costs by as much as $600 million, while European GPU cloud operator Nscale locked in a multi-billion-dollar compute supply arrangement with Microsoft ahead of its expected fall IPO. These are not speculative postures; they are the executed moves of a company that understands its infrastructure cost base will define its margin profile for years.
Yet the picture carries a structural tension that no volume of capex can fully resolve in the near term. Multiple outlets reported this week that as much as 70 percent of Microsoft's $241 billion AI revenue in FY2026 may be tied, directly or indirectly, to OpenAI — a figure Microsoft has not officially confirmed, and one that, if accurate, would represent an unusual revenue concentration for a company of its scale. The relationship is symbiotic by design: Microsoft is reportedly providing financial backstop for OpenAI's own data center capital expenditures, and the two organizations remain tightly coupled on model access, Azure hosting, and enterprise distribution. That coupling is a genuine risk. If OpenAI's revenue trajectory disappoints, or if enterprise customers increasingly route AI workloads toward competing frontier models — Anthropic, Mistral (with which Microsoft also signed a sovereign cloud deal in late July), or open-weight alternatives — the portion of Azure's AI premium effectively borrowed from a single third-party relationship becomes exposed.
The geographic and regulatory dimension adds further complexity. In the United Kingdom, Microsoft warned in late July that grid interconnection delays could extend to eight years, placing a $3.2 billion data center investment at risk — a concrete illustration of how the constraint has migrated from chip fabs to substations and planning authorities. This week Microsoft opened its largest data center facility in India, part of a broader Asia-Pacific push where AI infrastructure competition is intensifying. Power, not silicon, now sets the pace: as multiple hyperscalers noted in their most recent earnings calls, electricity availability, network buildout, and construction timelines have become the decisive limiting factors. It is precisely this context that makes the SpaceX arrangement — with its implied access to purpose-built generation capacity — strategically significant beyond the headline number.
Three signals merit close attention over the next two to four quarters. First, whether Microsoft provides clearer disclosure on the composition of its AI revenue: the concentration question will not stay buried, and investor pressure for transparency is building. Second, the pace at which AMD Helios production deployments scale within Azure — meaningful Helios utilization would confirm that the multi-vendor silicon strategy is operational rather than aspirational, reducing exposure to NVIDIA pricing leverage. Third, the shape of data center power procurement deals that become public: if the SpaceX arrangement proves representative of a new model — gigawatt-scale, long-term, purpose-built — it will signal that hyperscalers have internalized energy as a permanent structural input rather than a commodity purchase. Azure has cleared $100 billion in annual revenue; the harder question is what the cost of the next $100 billion will look like, and who controls the inputs that make it possible.