Last updated: August 1, 2026 · By Vishal Swami, Founder & Lead AI Reviewer, AISagely
Larry Ellison Bet It All on the AI Boom by taking Oracle from a database company into a debt-funded AI data center operator, and the bet is now underwater. Oracle's stock has fallen more than 50% from its September 2025 peak, wiping out roughly $213 billion of Ellison's personal fortune and dropping him from the world's No. 2 billionaire to No. 8 on the Bloomberg Billionaires Index, according to reporting from AI Weekly published July 31, 2026.
Short answer: Larry Ellison borrowed more than $100 billion to turn Oracle into an AI infrastructure supplier, betting on a $300 billion OpenAI computing deal. Oracle's stock is now down over 50% from its September peak, its free cash flow is negative $24 billion, and its backlog leans heavily on one customer — which is exactly the profile of a company that could become the face of an AI bust.
I track AI vendor pricing and infrastructure spending for a living, comparing what tools cost against what they can plausibly sustain, so Ellison's story is one I've been watching closely since the Oracle-OpenAI deal was announced last September. In my testing of AI tools over the past year, the ones with the thinnest margins have been the first to raise prices or throttle usage limits when their underlying compute costs went up, which is exactly the pressure Oracle's debt load could put on everything built on top of it. It's tempting to read this as billionaire gossip. It isn't. Oracle is the company renting compute to OpenAI, Meta, and xAI, so its balance sheet is a preview of whether the AI tools you pay $20 a month for are actually solvent underneath the sticker price.
What Ellison actually did
The 80-year-old Ellison reclaimed operational control of Oracle after ChatGPT launched in November 2022, convinced that generative AI would reshape enterprise computing and that Oracle couldn't afford to sit it out. He committed the company to a "Stargate" buildout of up to $500 billion in AI-focused data centers over four years, targeting a combined 10 gigawatts of power capacity, and in September 2025 Oracle signed a $300 billion multiyear deal to supply computing power to OpenAI.
None of that came cheap. Oracle raised $50 billion in bonds in February 2026, then tacked on another $58 billion in related borrowing within the same two months — more than $100 billion in debt financing in roughly 60 days, per Crypto Briefing’s reporting on the pivot. Fiscal 2026 capital expenditure came in at $55.66 billion, above the company's own $50 billion guidance, and free cash flow went negative $24 billion. Cloud infrastructure revenue did grow 93% year over year, which is the number Oracle leads with on earnings calls. The number it says less about is that its $638 billion remaining performance obligations backlog is heavily concentrated in a single customer whose own IPO has reportedly slipped to 2027.
At the peak, in September 2025, Oracle's stock surge briefly made Ellison the richest person alive, with an estimated net worth between $393 billion and $400 billion. By late July 2026, Yahoo Finance reported shares falling as much as 16.5% in a single trading session after Oracle's quarterly results missed profit and revenue forecasts, even as the company committed to spending still more on AI data centers. Credit-default swap prices on Oracle's debt spiked to three-year highs the same week, a sign that bond investors were pricing in real doubt about whether the borrowing gets paid back on schedule.
Why one company's debt is everyone's problem
The reason this isn't just a stock story is that Oracle sits underneath other people's AI products. If Oracle's bet on OpenAI's growth doesn't pan out on the timeline its bonds assume, the fallout doesn't stay contained to Oracle shareholders — it touches every AI subscription and API that ultimately runs on rented compute, which by 2026 is most of them. I've written before about how the AI industry’s circular deals are reigniting bubble fears and about why AI is getting so expensive to run even as prices to consumers stay flat. Ellison's bet is the clearest single example of that pattern: borrow heavily today against revenue that depends on one customer's growth curve holding.
Oracle isn't alone in taking this kind of swing, either. I covered Apple’s much smaller, much more conservative AI spending as the opposite end of this same trade-off, and the pattern showing up across corporate America pulling back on AI spending this year suggests boards are starting to ask the same debt-versus-payoff question Oracle's bondholders are asking. Zoom out further and it connects to the broader AI mania reshaping how companies make decisions right now — Ellison's bet is an unusually large, unusually well-documented data point in that story, not an isolated one.
What you need before you check your own exposure
You don't need a Bloomberg terminal to run a version of this check on your own AI stack. Grab a list of every AI tool you pay for, whether that's a subscription like ChatGPT Plus or an API key billed by usage, and note which company actually runs the servers behind it. For anything built on OpenAI, Anthropic, or Google's infrastructure, that means looking one layer down at who's financing the compute, not just who's billing your card. Ten minutes and a search for "[vendor] earnings" or "[vendor] debt" is enough for a first pass — you're not predicting a crash, just finding out how exposed your own workflow already is.
Step-by-step: sizing up your own AI vendor risk
1. Separate the infrastructure layer from the app layer
Oracle, Microsoft, Google, and Amazon build and rent the data centers. ChatGPT, Claude, and Gemini are apps sitting on top of that infrastructure. A price change or outage at the infrastructure layer hits every app built on it at once, so know which infrastructure provider sits underneath each tool you rely on.
2. Read the backlog number skeptically
Oracle's $638 billion backlog sounds enormous, but a backlog is a promise to be paid over years, not cash in the bank today. When a vendor brags about contract value, check whether they've also disclosed free cash flow. Oracle's is negative $24 billion right now — growth and profitability are not the same claim.
3. Check who's on the other side of the big contracts
A backlog concentrated in one or two customers is fragile. Oracle's is leaning heavily on OpenAI, whose own IPO has reportedly moved to 2027. If your AI vendor's growth story depends on one other company's growth story, you're carrying two risks stacked on top of each other.
4. Track debt and credit signals, not just the stock price
Stock charts move on sentiment. Bond yields and credit-default swap prices move on whether investors think a company can actually service its debt. Oracle's CDS prices hit three-year highs the same week its stock cratered — that's the more useful signal if you're trying to judge staying power.
5. Decide how much single-vendor exposure you can tolerate
There's no universally right answer here. A hobbyist using a free ChatGPT tier has almost no exposure. A small business running its entire support workflow through one paid API has a lot. Match your fallback plan to how much it would actually cost you if that one vendor's pricing or availability changed overnight.
6. Keep one working alternative on hand
You don't need a backup for every tool, just for the one your work would actually stop without. My free AI tools roundup is a fast place to line up a no-cost fallback before you need it in a hurry.
Example prompts you can copy
- Map your dependency chain: "Here's my main AI subscription: [tool]. Tell me which cloud infrastructure provider it most likely runs on, and what I'd need to know to judge that provider's financial health."
- Stress-test a single vendor: "If [AI tool]'s price doubled or the service went down for a week, what in my workflow would actually break versus what would just be annoying?"
- Cut through a debt headline: "Summarize this article about [company]'s AI spending in three sentences: how much debt, what it's for, and what has to go right for it to pay off — no marketing language."
- Build a fallback plan: "I rely on [tool] for [task]. Suggest two realistic alternatives I could switch to within a week, including at least one free option."
Common mistakes to avoid
The mistake I see most often is treating a big contract announcement, like Oracle's $300 billion OpenAI deal, as if it were cash already collected. It isn't; it's a bet that a customer's growth continues on schedule. Second is confusing revenue growth with financial health — Oracle's cloud revenue grew 93%, and its free cash flow still went negative $24 billion in the same year, and both of those things are true at once. Third is ignoring customer concentration: a backlog that leans on one buyer is a single point of failure no matter how large the headline number is. Fourth is overreacting to any single "bubble" headline by canceling tools that have nothing to do with the company in question, instead of checking your own specific exposure first.
Oracle's AI bet: September 2025 vs. July 2026
| Metric | September 2025 peak | July 2026 |
|---|---|---|
| Oracle market valuation | Near $1 trillion | Down more than 50% from peak |
| Ellison's Bloomberg Billionaires rank | No. 2 (briefly No. 1, ~$393–400B net worth) | No. 8 (–$213B from peak) |
| AI-related debt raised | — | $50B bonds (Feb 2026) + $58B more within 60 days |
| Fiscal-year capital expenditure | $50B guided | $55.66B actual (above guidance) |
| Free cash flow | Positive | –$24 billion |
| Cloud infrastructure revenue growth | — | +93% year over year |
| OpenAI compute deal | $300B signed | OpenAI's own IPO reportedly pushed to 2027 |
| Remaining performance obligations backlog | — | $638 billion, concentrated in OpenAI |
Tools that make this easier
Sizing up a vendor's financial health is a habit worth building whether or not Oracle's bet ever pays off. My AI tool ratings hub tracks where each major assistant is genuinely strong versus overhyped, which is the same lens worth applying before you build a workflow around any single provider. If you're specifically trying to reduce single-vendor risk, my guide to using Claude covers getting a second, independent opinion instead of leaning on one assistant for everything work depends on. And if the debt story above has you rethinking your budget, the free AI tools roundup is where I'd start building a cushion.
My take
Ellison didn't do anything irrational. Sitting out the AI infrastructure race the way Apple did carries its own risk of falling behind for a decade. But betting over $100 billion of new debt on one customer's growth curve is the kind of concentrated exposure that turns a bad quarter into a genuine crisis, and Oracle just had a bad quarter. Whether Ellison ends up remembered as the face of the AI bubble or as the guy who got the infrastructure build-out right depends entirely on whether OpenAI's revenue catches up to Oracle's backlog before Oracle's bond payments come due. That's not a bet I'd want to be personally exposed to, and it's worth checking whether, one or two layers down, you already are.
Frequently Asked Questions
Is Larry Ellison actually going to become "the face of the AI bubble"?
No one knows yet — that's the framing the New York Times Magazine used in its July 31, 2026 profile of Ellison's Oracle bet. What's not in dispute is the exposure: Oracle has raised over $100 billion in debt against a backlog concentrated in one customer, which is a riskier position than most of its AI-spending peers.
How much money has Larry Ellison lost in the Oracle AI bet?
Oracle's stock has fallen more than 50% from its September 2025 peak, which has cut roughly $213 billion from Ellison's fortune and dropped him from No. 2 to No. 8 on the Bloomberg Billionaires Index as of late July 2026.
Why did Oracle take on so much debt for AI?
Oracle committed to a Stargate data center buildout of up to $500 billion over four years and signed a $300 billion compute deal with OpenAI in September 2025. Building that capacity ahead of revenue required borrowing — $50 billion in bonds in February 2026 alone, plus $58 billion more within the same two months.
Does this mean the AI bubble is about to burst?
Not necessarily. Oracle's cloud infrastructure revenue is still growing 93% year over year, and a large backlog can convert into real revenue if OpenAI's growth holds up. The risk is concentration and timing: free cash flow is already negative $24 billion, and the backlog leans on one customer whose own IPO has slipped to 2027.
What should I do if I rely on AI tools built on Oracle or OpenAI infrastructure?
List what you actually pay for, check which infrastructure provider sits underneath each tool, and keep one free or low-cost fallback ready for whichever tool your work would genuinely stop without. That's the same protection against concentration risk that Oracle itself is short on right now.