Microsoft AI Capex in 2026: The $190 Billion Bet on Azure, Data Centers, and OpenAI

Microsoft doubled its AI capex to a $190 billion annual run-rate in fiscal 2026 and locked in a $250 billion Azure commitment from OpenAI through 2030 — the two numbers that define its bet on owning AI infrastructure. That's the short answer. The longer answer is that Microsoft still can't build fast enough to keep up with its own demand.

Every hyperscaler earnings call in 2026 has turned into a capex report. Microsoft's is the most interesting one to read closely, because a huge and growing share of that spend is contractually linked to a single customer — OpenAI — rather than spread across a diversified base of internal Azure workloads. That concentration is both the bull case and the risk.

Microsoft's 2025-2026 AI Capex: How Much and Where It's Going

Microsoft spent $34.9 billion on capital expenditures in Q1 FY2026 and $37.5 billion in Q2 FY2026 — a 66% jump year-over-year in that single quarter — putting the company on an annualized run-rate above $150 billion and tracking toward roughly $190 billion in total fiscal-2026 capex. About $25 billion of that increase is attributed to higher chip and component costs alone, not just added volume.

Roughly two-thirds of each quarter's spend goes to short-lived assets — Nvidia and custom GPUs, CPUs, and networking gear that typically depreciate over a few years — while the remaining third funds long-duration infrastructure: land, buildings, and power systems built to last 15+ years. Microsoft added about one gigawatt of new data center capacity in Q2 FY26 alone, and CFO Amy Hood has told investors the company expects to stay capacity-constrained on GPUs, CPUs, and storage through at least the end of 2026.

The $250 Billion OpenAI Azure Commitment

On October 28, 2025, Microsoft and OpenAI signed a new definitive agreement in which OpenAI committed to purchase an incremental $250 billion of Azure services — on top of existing commitments — spanning an estimated six-year window from 2025 through 2030. Spending is expected to start small, around $2 billion in 2025, before accelerating to an estimated $10 billion in 2028, $20 billion in 2029, and $60 billion in 2030 as OpenAI's own compute needs scale with model training and inference demand.

In exchange for anchoring that demand, Microsoft holds approximately 27% of OpenAI's restructured Public Benefit Corporation on an as-converted diluted basis — but it gave up the right of first refusal it previously held on OpenAI's cloud workloads, meaning OpenAI can now also buy compute from Oracle, CoreWeave, and other providers. That trade-off matters: Microsoft locked in a quarter-trillion-dollar demand floor but no longer has exclusive claim on the fastest-growing AI customer on earth. For more on how that restructuring changed the underlying economics, see our breakdown of the Microsoft-OpenAI deal terms.

Microsoft AI Infrastructure Spend vs. Google, Amazon, and Meta

Microsoft isn't spending in a vacuum. Combined 2026 AI capex across the four largest hyperscalers is projected near $725 billion, up roughly 77% from about $410 billion in combined 2025 spending, as each company bets that being under-provisioned on compute is a bigger risk than overspending.

Figures are 2026 estimates blended from Tom's Hardware, CNBC, and company earnings calls covering Amazon, Microsoft, Google, and Meta capex guidance. Individual quarters vary by reporting cadence.

Why Microsoft Is Still Capacity-Constrained at $190 Billion

The counterintuitive part of Microsoft's 2026 story is that record spending hasn't closed the gap between supply and demand. Microsoft carries an estimated $80 billion unfulfilled Azure backlog — contracted revenue it can't yet deliver because it doesn't have the physical capacity to serve it. That's not a demand problem; it's a power and construction-timeline problem. New data centers take 18-36 months to permit, build, and energize, and grid interconnection queues in most U.S. markets now run multiple years.

That's why Microsoft's capex mix skews so heavily toward long-duration assets even in a quarter dominated by GPU purchases: land, substations, and building shells have to be secured years before the racks that go inside them are even ordered. Anyone underwriting Azure's growth in 2027 and beyond should treat power availability, not GPU supply, as the real bottleneck variable. For a broader view of how AI infrastructure spend is being priced across the market, see our AI valuations dashboard.

How Microsoft Is Funding $190 Billion in Capex

Microsoft can absorb this spending in a way most companies can't because it still generates enormous free cash flow — but even Microsoft's cash flow is starting to feel the strain. Operating cash flow has stayed strong, yet free cash flow (operating cash flow minus capex) has compressed sharply as capital spending has outpaced growth in cash generated from the underlying business. That's a big part of why Microsoft, alongside Meta and Oracle, has turned to the corporate bond market and off-balance-sheet financing structures — including special-purpose vehicles used to fund data center construction — rather than paying entirely out of pocket.

The depreciation math matters too. Roughly two-thirds of Microsoft's AI capex goes into GPUs and CPUs that Microsoft depreciates over a relatively short useful life (often 2-4 years for AI accelerators), compared with the 15+ year life assumed for buildings and land. As the GPU-heavy share of the spend has grown, Microsoft's own depreciation expense has climbed accordingly, which is now a headwind analysts flag when modeling Azure segment margins — the infrastructure shows up on the income statement as depreciation well before the associated Azure AI revenue fully ramps.

That's the tension underlying every hyperscaler capex debate in 2026: is this spend translating into revenue fast enough to justify the depreciation and financing cost being taken on today? Microsoft's answer so far has been that Azure's overall growth rate, and specifically its AI-services growth rate, is accelerating enough to make the trade worthwhile — but it's a bet the market is still pricing with real uncertainty, which is visible in how sensitive Microsoft's stock has been to any capex-guidance surprise on recent earnings calls.

What Microsoft's AI Capex Means for Investors

For public-market and late-stage private investors, Microsoft's capex trajectory is the cleanest proxy available for how confident the industry actually is in near-term AI demand — a company doesn't commit $190 billion in a single fiscal year on a hunch. But the concentration risk is real: a meaningful and growing share of that spend is tied to OpenAI's usage commitments specifically, and OpenAI is simultaneously diversifying its own compute sourcing to Oracle and CoreWeave now that Microsoft's right of first refusal is gone.

That means Azure's AI revenue growth over the next several years is partly a bet on OpenAI's own trajectory — its ability to keep raising capital, keep scaling revenue, and keep needing more compute rather than plateauing. For venture investors building or backing companies in the AI infrastructure stack, Microsoft's spend confirms the compute supply crunch is a multi-year phenomenon, not a 2025 blip, and pricing power sits with whoever controls power and land, not just chips. See our big tech earnings tracker for how this shows up in quarterly results across all four hyperscalers.

Microsoft AI Capex Timeline: What's Already Locked In

It's worth separating what's already contracted from what's still a projection. The $250 billion OpenAI Azure commitment, the roughly $80 billion unfulfilled backlog, and the fiscal 2025 baseline of about $88 billion in total capex are all disclosed, reported figures. The $190 billion FY26 estimate and the $150 billion-plus annualized run-rate are extrapolations built from the first two quarters of the fiscal year, so the actual full-year number will depend on how aggressively Microsoft keeps building through Q3 and Q4 FY26 — and on whether Amy Hood's team decides to pull back if Azure growth cools or lean in further if the OpenAI ramp accelerates ahead of the 2028-2030 schedule.

The near-term milestones worth tracking: Microsoft's Q3 FY26 earnings call (expected late April 2026) for updated full-year capex guidance, any incremental Azure commitments from OpenAI as it approaches the $10 billion 2028 spending threshold, and whether the $80 billion backlog shrinks or grows as new data center capacity comes online through the back half of calendar 2026. Each of those data points will tell you whether Microsoft's bet is being validated by actual usage or whether the company is still building ahead of demand it hasn't yet proven out.