Investors parsed Microsoft’s fiscal fourth-quarter results on Thursday with unusual care, and the number that drew the most attention was not revenue. It was a figure buried in the company’s disclosure of data-center leases: $329.1 billion in commitments not yet begun.
That total, disclosed in Microsoft’s quarterly filing, nearly doubled from $196.6 billion three months earlier, after the company signed more than $130 billion in new lease commitments during the quarter, according to the filing. The commitments are a window into how aggressively Microsoft is betting on the AI buildout, and how much it is spending before the revenue arrives.
The headline numbers were strong. Revenue rose 18% to $90.01 billion in the quarter ended June 30, beating the $87.72 billion analysts had projected. Intelligent Cloud revenue of $39.31 billion also topped expectations, and Azure grew 43% excluding currency effects. Shares jumped about 9% in after-hours trading as investors read the results as proof that Microsoft’s AI investments are converting into cloud demand rather than weighing on margins.
The earnings release also contained a first. Microsoft disclosed a $320 million book gain on its investment in Anthropic, the AI startup that has emerged as its closest rival to OpenAI. The company has poured roughly $13 billion into OpenAI and added Anthropic to its portfolio last year as part of a cloud and computing agreement, though it has previously declined to detail the investment.
The disclosure lands as Microsoft’s relationship with OpenAI has grown more complicated. The two companies remain close, with OpenAI running largely on Azure and Microsoft holding a large equity stake, but each has moved into the other’s territory. Microsoft sells its own Copilot assistants to the same enterprise customers that OpenAI courts with ChatGPT, and OpenAI has struck cloud deals with rivals.
TechCrunch, in an analysis published Thursday, argued that Microsoft has effectively crossed from investor to direct competitor, and that a fight for enterprise customers among Copilot on one side and GPT and Gemini on the other is now unavoidable. Microsoft has positioned the strategy as complementary, arguing that customers will use multiple models and that its cloud will serve them all.
The lease commitments tell the story of that ambition. Microsoft’s total undelivered lease obligations, which include capacity contracted but not yet built, jumped to $329.1 billion as of June 30. The figure is not a cash outlay for the quarter; payments begin as facilities come online. It is a measure of how much capacity Microsoft expects to need, and the scale of the jump surprised analysts.
Cloud providers sign such leases years in advance because data-center construction and power procurement take that long. The increase suggests Microsoft’s internal projections for AI demand have grown sharply, and it provides a concrete signal for the equipment makers that supply the industry. For chip makers, the signal landed like a thunderclap. The Philadelphia Semiconductor Index climbed more than 6% on Thursday, with equipment suppliers such as Lam Research and Applied Materials surging on the view that Microsoft’s numbers imply AI computing demand is still accelerating rather than peaking.
The results also relieved a fear that had crept into markets in recent weeks: that hyperscalers would pause their AI buildouts to protect profits. Microsoft has said capital spending will continue to climb this fiscal year as cloud and AI demand outruns supply, echoing statements from Alphabet and Amazon.
The debate among analysts is no longer whether Microsoft can sell AI but how much the selling costs. Operating margins in the intelligent cloud unit narrowed slightly, and the company’s financing arm has expanded to fund the leases, a structure that shifts some cost onto the balance sheet. Analysts pointed out that the arrangement adds fixed obligations that will be hard to unwind if demand cools. Rivals have signed similar leases, spreading the same risk across the industry.
There is also an accounting wrinkle. Microsoft recognizes revenue from its partnership with OpenAI partly through a profit-sharing arrangement, which complicates the comparison between reported growth and underlying demand. The $320 million Anthropic gain illustrates how Microsoft’s AI exposure now runs through equity stakes as well as its own products, giving it multiple ways to benefit from a boom that is also funding its competitors.
The relationship with OpenAI is also financial, and it is changing. The two companies have renegotiated their computing agreements before, with OpenAI committing to buy billions of dollars of Azure capacity while spreading its workloads across other providers. People familiar with the matter said the latest terms give Microsoft a growing share of OpenAI’s inference revenue while OpenAI gains more flexibility over where its models run. Each renegotiation recalibrates the balance between a customer and the investor that also wants to compete with it.
For now, investors are choosing to focus on the growth. Microsoft’s stock, up sharply over the past year, is trading near records, and the quarter gave bulls little reason to change their models. The lease number, though, will be watched closely. When Microsoft reports again in October, investors will look at two figures: how much revenue Azure generated, and how much more capacity Microsoft promised. The gap between them is the price of staying in the AI race.
