Inside a Microsoft data center that is still under construction, there is a certain silence. Not the hollow kind. the anticipatory type. As you pass one of the new locations in Texas or Wisconsin, you’ll notice fleets of idle trucks, copper wound into pickup truck-sized spools, and the constant hum of generators warming up even before the racks arrive. For the most part, eighty billion dollars looks like that. Transformer yards, concrete pads, and hard-hat-wearing engineers staring at clipboards. Not glitzy. Not the way it sounds in the headlines.
In the midst of all that chaos, however, is one of the most important corporate finance choices that any American business has made in a long time. The decade, perhaps. Microsoft has committed to investing about $80 billion in AI-enabled infrastructure in fiscal 2025 alone, and the larger hyperscaler pack has increased total announced capital expenditures for 2026 to over $700 billion, a 69% increase from the previous year. Nation-states used to own numbers like that. They now work for a software company.
It’s worth stopping to consider how strange this is. In the past, Microsoft has been the one with discipline. The company that ran lean, printed cash, paid dividends, and allowed Amazon to overspend on warehouses. The AI wager is the sequel to Satya Nadella’s wager on Azure, which altered that calculus more than ten years ago. Nadella sounded almost composed about it when he spoke at Morgan Stanley’s TMT conference in San Francisco this past March—the kind of person who has already done the math three times. He likened the skepticism of today to the doubts he encountered in 2014, when investors were unsure that cloud spending would ever be profitable. That was successful. This will, too, he’s betting.
Investors are not totally persuaded. The Dow saw its biggest sell-off of 2025 after a TD Cowen note in early 2025 suggested Microsoft had canceled some data center leases, and software stocks fell about 20% in the first two months of 2026. The $80 billion plan was swiftly reiterated by the company, but it added a clause that appeared to be a tiny hedge: it might “strategically pace or adjust” infrastructure in certain areas. Translation: Even Microsoft is unsure of the precise amount of capacity required worldwide. No one does.
It’s not just the size that makes this the decade’s biggest corporate finance story. It’s the framework. Big Tech is increasingly using debt rather than cash to finance this expansion. An off-balance-sheet joint venture is used in Meta’s $27 billion Blue Owl acquisition for its Hyperion campus; Oracle is expected to nearly triple its net adjusted debt by 2028. In contrast, Microsoft is more cautious, combining issuance with its massive cash flow, but the industry-wide trend has changed. AI-related tech companies issued $141 billion in corporate debt in 2025, surpassing the previous full-year record, according to Goldman Sachs. Major tech companies now have about $1.35 trillion in total interest-bearing debt, which is about four times more than it was ten years ago.
As this develops, it seems as though a fundamental aspect of how software companies finance themselves is being rewritten. The model was straightforward for thirty years: strong balance sheets, high margins, and minimal capital expenditures. That is being broken by the AI era. No one can say with certainty whether Microsoft’s $80 billion will eventually resemble Amazon’s warehouses (vindicated) or the dark fiber overbuild of 1999 (a graveyard of optimism). Nadella believes that by extracting more power from each GPU, kernel, and token, software itself will be the lever that makes the math work. He might be correct. He has previously been correct.
Even so, it’s difficult to ignore the discrepancy between the quiet recalibration taking place behind closed doors and the assurance in the press releases. A lot of concrete can be purchased for eighty billion dollars. Additionally, it creates a lot of pressure to be right.
