# OpenAI raises its compute bet to $750 billion but its own CFO isn't sure it can pay the bill > Source: > Published: 2026-07-22 18:06:41+00:00 *OpenAI has lifted its projected compute spending through 2030 to $750 billion, up from roughly $600 billion earlier this year, even as CFO Sarah Friar has privately warned the company may be unable to honor those contracts if revenue growth doesn't accelerate sharply.* The Wall Street Journal reported the revised figure on July 22, and the number is staggering enough on its own. But the more revealing detail isn't the size of the commitment. It's that OpenAI's own finance chief has told leadership the company may not be able to pay for it. The $750 billion spans a web of deals: cloud contracts with Microsoft, Oracle, AWS, and CoreWeave, plus OpenAI's own Project Camellia, a 1,400-acre campus in Effingham County, Georgia, northwest of Savannah, where the company plans to build 3.2 gigawatts of data center capacity at a cost of $20 billion. Development is slated to begin in 2028 and run through 2032, with Georgia Power delivering power in phases. It's OpenAI's first venture as the principal designer and builder of its own infrastructure, and Bloomberg reported on July 22 that the eventual price tag may exceed $30 billion. OpenAI hit a $25 billion annualized revenue run rate in February 2026, up from roughly $20 billion at the end of 2025 and $2 billion just three years ago. That's a growth rate that would make most companies envious. The problem is the spending trajectory is rising faster still, and the commitments are locked in while the revenue isn't. According to reporting from Fortune in April 2026, Friar expressed concern that OpenAI was spending too much on data centers and may not generate enough revenue to cover contracts it had already signed. A single gigawatt-scale facility, fully loaded, runs roughly $50 billion and takes about three years to bring online, meaning the capital decisions made today won't translate into usable compute until 2028 at the earliest. You're writing checks now for capacity you can't run yet, against revenue projections that have already slipped once. OpenAI missed its own internal revenue targets as recently as this spring, according to a Wall Street Journal report. The company's operating margin is deeply negative, burning through capital even as its subscriber base and enterprise deals grow. Revenue is split roughly 65% from ChatGPT subscriptions, 25% from API access, and 10% from partnerships, with enterprise contracts on track to reach parity with consumer by end of year. That mix is improving, but the contracts Friar is worried about don't care about trajectory. They require cash. ## What $750 billion actually means for the hyperscaler arms race Oracle's piece of this alone, a $300 billion cloud contract with OpenAI beginning in 2027 and spanning five years, is larger than most countries' annual defense budgets. Microsoft, Amazon, and CoreWeave are each exposed as well, which is why TipRanks noted that the revised spending figure immediately put those stocks in focus. The hyperscalers are not passive infrastructure vendors here. They're co-betting on OpenAI's ability to generate the revenue that makes their contracts whole. Frankly, that's a different kind of risk than markets are used to pricing. When a cloud provider signs a capacity deal, the credit risk normally sits with a corporate customer running a known business. OpenAI is a different animal: a company with extraordinary growth, a negative operating margin, and a valuation of $730 billion (set at its February 2026 funding round, where Amazon committed $50 billion and Nvidia and SoftBank each put in $30 billion) that rests almost entirely on the assumption that AI demand will outpace AI spend indefinitely. Friar's concern puts a number on what happens if that assumption is off by even a few years. For founders benchmarking AI infrastructure costs, the Project Camellia commitment is useful ground truth: 3.2 gigawatts, 1,400 acres, a decade to full operation, and a price tag that Bloomberg already thinks will exceed $30 billion. The implied cost per gigawatt is roughly $9 to $10 billion before you account for power delivery and cooling infrastructure. OpenAI's closed-loop cooling system, designed to minimize water consumption and reduce demand during peak grid periods, is worth noting as a design choice that signals the company is at least thinking about operating costs, not only capital costs. None of this means OpenAI is in immediate trouble. A $25 billion annualized run rate is real revenue, and the company's user base, over 500 million weekly active users by recent accounts, gives it genuine pricing power. But the WSJ's reporting makes clear there's a scenario, not a remote one, where OpenAI has committed to more compute than its revenue path can support. The CFO said so. That's not a bear-case hypothetical. It's an internal warning from the person whose job is to know. The broader question for investors pricing OpenAI's eventual IPO is whether the company's spending is building a moat or a liability. The difference depends entirely on how quickly enterprise AI adoption converts into durable, contracted revenue. Right now, the commitments are more certain than the cash flows. That gap is what $750 billion looks like up close. **Also read:** [Samsung launches three foldables at once in London and bets Gemini AI can make them mainstream](https://startupfortune.com/samsung-launches-three-foldables-at-once-in-london-and-bets-gemini-ai-can-make-them-mainstream/) • [Monday.com cuts 620 jobs and raises its margin outlook in the same breath](https://startupfortune.com/mondaycom-cuts-620-jobs-and-raises-its-margin-outlook-in-the-same-breath/) • [Meshy raises nearly $400 million to make AI-generated 3D models a standard production tool](https://startupfortune.com/meshy-raises-nearly-400-million-to-make-ai-generated-3d-models-a-standard-production-tool/)