Oracle Bet the Farm on AI. Now It’s Telling Investors the Farm Could Burn.
Oracle signed up to spend $300 billion on AI datacenters for OpenAI over five years. That’s not a bet — that’s a conviction that the world’s most famous money-losing startup will figure out profitability before the lease runs out.
This week, in a regulatory filing, Oracle outlined all the ways this could go sideways. It’s worth reading, not because the risks are surprising, but because the company chose to name them out loud in a legal document with SEC teeth.
Let’s walk through them.
The customer might not pay. OpenAI’s tab under the Stargate deal could hit $30 billion annually. OpenAI has never turned a profit. Its ability to pay Oracle depends entirely on its ability to keep raising capital. That’s not a business model — that’s a fundraising treadmill with a very expensive seat. Oracle’s filing says it plainly: “Our business is, and may continue to be, exposed to risks of customer non-payment and non-performance.”
The infrastructure might be worth nothing if OpenAI walks. Oracle leases datacenter capacity from partners (Crusoe, among others) rather than building its own. If OpenAI doesn’t renew, Oracle is on the hook for capacity it can’t easily re-lease. The filing: “If customers do not renew their contracts, we may be unable to re-lease, repurpose or assign such capacity on acceptable terms, if at all.”
Power is harder to find than expected. Oracle admits it’s struggling to secure “reliable and cost-effective power” for its AI buildout. Power prices are volatile. Grid capacity is constrained. And if Oracle signed fixed-price contracts with customers, rising energy costs eat straight into margin.
Construction is a nightmare. Oracle’s filing lists everything that could delay a datacenter build: permitting, zoning, utility interconnection, equipment delivery, contractor performance, government moratoria on new construction, environmental regulations. It’s three dense paragraphs of “anything that can go wrong, will.”
The spending spiral is locked in. Oracle spent $55 billion on capex in FY2026. It plans $70 billion in FY2027. To fund this, it’s raising $40 billion in new debt and equity — on top of $18 billion raised last September. The company warns that if it doesn’t keep spending, it “may fall behind technological developments and evolving industry standards.” Damned if you spend, damned if you don’t.
The market is already voting. Oracle’s stock is down 40% in the last month.
Now, the counterarguments are real too. Oracle has $155 billion in remaining performance obligations beyond the OpenAI deal. AI infrastructure demand is genuinely exploding — every major enterprise is trying to figure out what it needs. And Oracle has survived boom-and-bust cycles before. The .com bubble didn’t kill them.
But here’s the thing that’s different this time: Oracle’s AI bet isn’t diversified across thousands of customers paying for databases and ERP. It’s concentrated in a handful of deals — one of which is with a company that has never, in its entire existence, generated a profit from its core business. If Sam Altman calls tomorrow and says “we can’t make Q3,” Oracle doesn’t just lose a customer. It loses a $300 billion anchor contract that the rest of the infrastructure buildout was designed around.
The most honest line in the whole filing: “If we do not continue to invest significant resources to develop and support our AI products, we may fall behind.” That’s the trap. Oracle has to keep pouring fuel on the fire because stopping means the whole thing collapses. The only alternative is to keep spending and hope the physics of AI demand catches up to the physics of power grids and construction timelines.
Stock markets love a story where a company bets big and wins. But Oracle just filed a document that reads less like a confident wager and more like a list of every way a fire could start — while admitting they have to keep adding fuel.
Sources: The Register — Oracle outlines all the ways it could lose the farm it bet on AI