Large U.S. companies just hit the brakes on AI—adoption fell from 14% to 12% in two months, the first decline since tracking began. MIT research explains why: 95% of enterprise pilots deliver zero ROI. The gap between AI hype and workflow reality is widening.
Good Morning from San Francisco, Oracle just inked a $300 billion computing deal with OpenAI. That's not a
Oracle bets $300B on OpenAI's computing future, but the math is stark: OpenAI generates $10B annually while committing to $60B yearly. The deal either transforms Oracle into an AI infrastructure leader—or becomes a cautionary dot-com tale.
Oracle bets $300B on OpenAI's computing future, but the math is stark: OpenAI generates $10B annually while committing to $60B yearly. The deal either transforms Oracle into an AI infrastructure leader—or becomes a cautionary dot-com tale.
💡 TL;DR - The 30 Seconds Version
👉 Oracle signed a $300 billion computing contract with OpenAI over five years, starting in 2027—one of the largest cloud deals in history.
📊 OpenAI generates $10 billion annually but commits to $60 billion yearly payments, while Oracle's debt-to-equity ratio hits 427% versus Microsoft's 33%.
🚀 Oracle's stock surged 43% in one day, adding $100 billion to Larry Ellison's wealth and briefly making him the world's richest person.
🏭 The deal requires 4.5 gigawatts of power capacity—equivalent to two Hoover Dams—for new data centers across five states starting 2027.
⚠️ Analysts see 1999 dot-com bubble energy as Oracle's capital spending jumps to 52% of revenue from 13% last year.
🌍 The arrangement tests whether AI business models can support the $2.9 trillion global infrastructure buildout planned through 2028.
A five-year, 4.5-gigawatt pact sends Oracle’s stock up 43%—and concentration risk through the roof.
Oracle’s $300 billion compute contract with OpenAI promises a dominant AI future; today’s reality is an unprofitable customer and a debt-heavy vendor, with revenue recognition that begins in 2027. The scale is breathtaking, and the timing is late. The deal was first detailed in a Wall Street Journal exclusive report.
OpenAI has committed to buy $300 billion of computing capacity from Oracle over roughly five years, with deliveries starting in 2027. The arrangement calls for 4.5 gigawatts of power—about two Hoover Dams’ output—to feed new U.S. data centers. It’s huge.
The pact marks a strategic break from OpenAI’s historical reliance on Microsoft. OpenAI has added Oracle to reduce supply bottlenecks and diversify vendors, while folding the build-out into its broader “Stargate” program. Oracle’s build partners include data-center specialist Crusoe, with sites under consideration in Wyoming, Pennsylvania, Texas, Michigan, and New Mexico. Geography matters.
Investors reacted first, asked questions later. Oracle shares spiked 43% after the company flagged $317 billion in new contract bookings for the quarter ending Aug. 31, a figure widely understood to be driven in large part by OpenAI. Oracle now reports $455 billion in “outstanding contract revenue,” up more than 350% year-on-year. Euphoria meets execution.
But the financial physics are hard. OpenAI disclosed around $10 billion in annual revenue this summer, far shy of the implied ~$60 billion per year average this contract entails. The startup has signaled losses through at least 2029, with eye-watering cash burn estimates. That gap must close.
Oracle faces its own balance-sheet test. Capital expenditures are running at about $35 billion this fiscal year—roughly 52% of revenue, up from 13% in 2024—while its debt load towers over peers. Microsoft’s debt-to-equity sits near the 30s; Oracle’s is in the 400s. Cheap money helps until it doesn’t.
Oracle leadership laid out an aggressive glide path: Oracle Cloud Infrastructure revenue up 77% to $18 billion by May 2026, then vaulting toward $144 billion by 2030. That would transform a once “overflow” provider into a hyperscaler rival on core AI workloads. Bold doesn’t begin to cover it.
Skeptics, however, see dot-com echoes. One-day market-cap explosions and multi-hundred-billion booking waves are rare for a reason. The projections depend on OpenAI’s adoption curve, competitive moat, and pricing power—and on Oracle’s ability to finance, procure, and stand up capacity at historical speed. That’s a lot of ifs.
This isn’t Oracle picking up scraps the big three can’t handle. It’s a bet to become essential infrastructure for the most prominent generative-AI company. If it sticks, Oracle vaults into the first rank of AI data-center operators, with long-dated, contracted cash flows to match. The narrative flips.
For OpenAI, the deal is a release valve on chronic compute shortages that have throttled product rollouts and model training. It also helps rebalance vendor dependence as OpenAI juggles partnerships, custom-chip ambitions, and regulatory scrutiny. Capacity is strategy.
AI infrastructure has become the economy’s new growth engine. Wall Street research pegs global spending on chips, servers, and data-center gear at roughly $2.9 trillion from 2025 to 2028, with data-center buildouts eclipsing consumer outlays as a driver of U.S. GDP this year. That is unprecedented.
It could be justified if AI-native revenue scales in tandem. It could also be a classic boom where capacity arrives faster than demand, compressing margins and exposing overextended balance sheets. Both readings fit the moment. Choose your parable.
The contract binds both companies tightly. Oracle concentrates an enormous slice of future growth on a single customer’s success; OpenAI signs up for obligations that dwarf its current revenue many times over. Unwinding would be costly for both. That’s the point.
The 2027 start buys time—and adds uncertainty. Oracle must execute a national construction program amid grid constraints, interconnect queues, and supply-chain choke points, all while managing leverage. OpenAI must convert user fascination into durable enterprise spend at scale. Delays will compound.
Three markers will separate hype from reality. First, groundbreakings and power contracts: steel in the ground and megawatts on paper matter more than press releases. Second, OpenAI’s unit economics: inference costs, throughput, and pricing discipline will determine whether revenue keeps up with capacity. Third, revenue mix and disclosures: how much of Oracle’s bookings are cancellable, how recognition ramps in 2027–2030, and how concentrated the top customer list remains. Execution is everything.
Why this matters:
Q: What exactly is Project Stargate that this deal is part of?
A: Project Stargate is OpenAI's broader data center initiative launched with SoftBank and Oracle in January 2025. The partnership aims to build $500 billion in AI computing infrastructure across the US over four years. This Oracle deal covers more than half that target.
Q: Why does Oracle's debt situation matter for this deal?
A: Oracle's debt-to-equity ratio of 427% is dramatically higher than competitors like Microsoft at 33%. The company is already spending more on infrastructure ($27.4 billion) than it generates in cash flow ($21.5 billion), meaning it needs to borrow heavily to build these data centers.
Q: Why doesn't this deal start until 2027?
A: Building 4.5 gigawatts of data center capacity takes years. Oracle needs time to secure land, get power grid connections, obtain permits, and construct facilities across multiple states. The 2027 start also gives OpenAI time to grow revenue before payments begin.
Q: What happens if OpenAI can't make the payments?
A: Oracle would face massive losses since the company is building infrastructure specifically for this contract. With $300 billion committed over five years, Oracle has concentrated enormous risk on one customer. The contract terms and cancellation clauses haven't been disclosed publicly.
Q: How does this compare to other major cloud deals?
A: This ranks among the largest cloud contracts ever signed. For context, OpenAI previously signed an $11.9 billion deal with CoreWeave. Most major cloud contracts are measured in single-digit billions over multiple years, making this $300 billion commitment unprecedented in scale.
Oracle's stock exploded 40% after revealing a $455B AI contract backlog and projections for $144B cloud revenue by 2030. The surge made Larry Ellison briefly the world's richest person—but can the company turn massive bookings into sustainable margins?
Publishers like Reddit and Yahoo launched a new licensing standard to charge AI companies for training data. The Really Simple Licensing protocol lets sites demand payment per crawl or per AI response. No major AI company has agreed to comply yet.
Meta's $72B AI talent hunt is imploding. ChatGPT co-creator nearly quit within a week, forcing tripled compensation. Elite recruits defecting to rivals while existing employees demand parity. The secretive TBD Lab creates new corporate castes.
Arm challenges the smartphone industry's NPU rush with Lumex, betting CPU-based AI can deliver 5x performance gains across 3 billion devices by 2030. The platform's SME2 instructions target developer frustration with fragmented neural engines.
Get the 5-minute Silicon Valley AI briefing, every weekday morning — free.
