Oracle’s $300 Billion Bet on OpenAI: The Hidden Risk Investors Are Missing
In September 2025, Oracle did something no enterprise software company had ever done before. It signed what is arguably the largest commercial cloud contract in corporate history — a deal reportedly worth $300 billion over five years, handing OpenAI a primary infrastructure partner after years of OpenAI flirting with AWS's reliability, Microsoft's deep pockets, and Google's hardware.
Wall Street celebrated. Oracle's shares surged 43% on the announcement, hitting an all-time high of $344. The narrative was seductive: Oracle, a legacy database company that many had written off as irrelevant in the cloud era, had just become the backbone of the most talked-about AI company in the world. Larry Ellison looked like a genius.
Six months later, the stock had nearly halved. The company signed a $300 billion agreement with OpenAI, but concerns over dependency on a single client led to a $315 billion drop in market capitalisation.
So what changed? And more importantly, what risk were investors slow to see?
The deal, decoded
To understand the risk, you first need to understand the sheer scale of what Oracle committed to.
Under the terms of the deal, Oracle committed to reserving 4.5 gigawatts of power — roughly the capacity of seven large nuclear power plants — and deploying approximately 500,000 high-end AI acceleration cards. The contract runs five years, starting in 2027, and is part of the broader Stargate project — a joint AI infrastructure campaign involving OpenAI, Oracle, and partners like SoftBank, collectively aimed at building up to 30 gigawatts of AI computing capacity in the US.
The financial commitment to make this happen has been staggering. Oracle's capex jumped from just $6.9 billion in FY24 to $21.2 billion in FY25. For FY26, the company has guided for $50 billion — more than double the previous year's spend. To fund this buildout, Oracle could not afford to do business with OpenAI unless it borrowed money, raising over $38 billion in bonds in late 2025 and announcing plans to raise an additional $45–50 billion through debt and equity in 2026. Total debt has since crossed $150 billion.
On paper, the revenue promise makes this look worth it. Oracle's remaining performance obligations — contracted revenue not yet recognised — surged 359% to $455 billion following the OpenAI agreement. By Q3 FY26, that figure had climbed to $553 billion. Cloud revenue grew 44% year-on-year. The backlog is real, and the demand is genuine. But the backlog is also where the hidden risk lives.
The one-customer problem
Here is the number that should keep Oracle investors up at night: analysts estimate that around 58% of Oracle's future order backlog could come from OpenAI alone.
In any other industry, a company with over half its contracted revenue tied to a single customer would face immediate scrutiny. In the AI infrastructure boom, this concentration risk has largely been overlooked in favour of the headline backlog growth story.
The backlog is simultaneously Oracle's greatest asset and its biggest vulnerability. A massive RPO signals demand, but it also raises concentration risk. If a significant portion of future revenue depends on a single customer, any disruption to that relationship would ripple through Oracle's financials.
To fund this buildout without waiting for revenues that won't arrive until 2027, Oracle is doing something it has never done at this scale before — cutting people to pay for machines.
The more uncomfortable truth behind the layoffs is what triggered them. According to research, multiple US banks pulled back from financing Oracle-linked data centre projects, with lenders roughly doubling the interest rate premiums they charge Oracle for project financing since September — pushing borrowing costs to levels typically reserved for non-investment-grade companies.
With external financing tightening, Oracle had to find capital from within. Options on the table included a headcount reduction of 20,000 to 30,000 to free up $8–10 billion in cash flow, and potentially selling off assets, including Cerner, the health tech specialist Oracle acquired for $28.3 billion in 2022. The workers being let go are not being replaced by AI. They are being replaced by capex that banks no longer want to fund.
Also, Analysts at RBC Capital have already flagged that Anthropic is gaining ground in the enterprise AI market and could chip away at OpenAI's dominance over time. If OpenAI's competitive position weakens, the downstream effect on Oracle is not a rounding error — it is existential.
OpenAI's own revenue problem
The concentration risk becomes more acute when you look at OpenAI's own financials.
At $10 billion ARR, OpenAI's monthly revenue is roughly $0.8–1.0 billion. A $60 billion per year compute bill — what Oracle expects to receive annually — equates to roughly $5 billion per month. At current revenue, paying Oracle would consume five times OpenAI's monthly earnings.
This is not a small gap. This is a structural mismatch that OpenAI must close through explosive revenue growth — or through continued access to external capital. OpenAI has wrapped up a funding round of $122 billion, including $50 billion from Amazon, $30 billion from Nvidia, and an equal amount from SoftBank, which gives it a war chest to fulfil obligations in the near term. But the math only works if OpenAI's commercial model scales dramatically between now and 2027, when the Oracle contract kicks into full gear.
Moody's flagged the counterparty risk associated with the deal, noting that Oracle's AI infrastructure business has "tremendous potential" but that exponential growth and already high debt burden "could result in an extended period of high leverage and negative cash flow." The agency stopped short of a rating action — but the warning was clear enough.
Cracks are already showing
The first real-world sign of friction appeared at Oracle and OpenAI's flagship Stargate data centre in Abilene, Texas. Oracle and OpenAI cancelled plans to expand their flagship Abilene Stargate data centre campus, with the expansion complicated by financing challenges and OpenAI's frequently changing demand forecasting and shifting view of Stargate. A multi-day outage caused by winter weather impacted liquid cooling equipment, reportedly damaging relations between OpenAI and Crusoe, the data centre operator.
The Abilene pullback was eventually explained away — OpenAI noted it chose to put additional capacity in other locations, and the broader 4.5-gigawatt agreement remains on track across multiple US sites. Oracle itself pushed back firmly on reports of delays. But the episode exposed something important: Oracle's infrastructure schedule is not entirely within its own control. It depends on third-party operators, power grid timelines, and crucially, OpenAI's shifting demand forecasts.
Oracle doesn't build or own the data centres it operates — the Abilene facility is built and owned by Crusoe, but furnished and operated by Oracle. Roughly half the cost of a modern AI data centre is land and infrastructure, with the remainder split between network, storage, and compute. This asset-light model limits Oracle's capital outlay on construction but also limits its control over delivery timelines.
The lease commitment nobody is talking about
Beyond the debt, there is another financial layer that has received less attention.
Oracle disclosed $248 billion in additional lease commitments, substantially all related to data centres and cloud capacity arrangements. These lease commitments are expected to start between Q3 FY26 and FY28, with 15 to 19-year terms that are not yet on the balance sheet.
Read that again. Fifteen to nineteen-year commitments — for infrastructure built to serve a five-year contract. The duration mismatch is extraordinary. If the OpenAI relationship changes after five years, Oracle is still sitting on nearly two decades of lease obligations. The company has essentially traded the certainty of a software subscription business for the capital structure of a utility, without the regulated returns that utilities receive.
Analysts worry that Oracle has traded its stable software-as-a-service revenue for a more volatile, capital-intensive utility model.
The bull case is real — but so is the uncertainty
None of this means Oracle's bet will fail. Q3 FY26 delivered 20% revenue growth and a 44% cloud segment surge. Demand for AI compute is genuinely outpacing supply. Oracle's co-CEO Clay Magouyrk has stated on earnings calls that demand for AI infrastructure continues to exceed supply, and that this is "directly visible" in the $553 billion remaining performance obligations.
The project is expected to contribute $30 billion to Oracle's revenue annually beginning in 2027. If OpenAI's commercial trajectory holds — and the $122 billion fundraise suggests investors think it will — Oracle is positioned to collect massive, recurring infrastructure revenues for years. The upside scenario is genuine.
But the downside scenario is also genuine, and it is underpriced in most analyst coverage. If OpenAI slows, stumbles competitively, or revises its demand forecasting again, Oracle is left holding $150 billion in debt, $248 billion in long-term leases, and data centres built for a customer that may no longer need them at full capacity.
What this means for investors
Oracle is no longer a software company with a cloud business on the side. It has fundamentally re-rated itself as an AI infrastructure utility — a picks-and-shovels play on the AI arms race, with the leverage to match.
The investment thesis now rests on three things that must all go right simultaneously: OpenAI's revenue scaling dramatically by 2027, the broader AI compute demand sustaining at current levels for 15+ years, and Oracle successfully diversifying its customer base beyond OpenAI before the concentration risk becomes a crisis.
The first two are plausible. The third is the one Oracle does not fully control.
Oracle's future is now inextricably linked to the success of a handful of hyperscale clients — and if OpenAI's demand were to plateau, Oracle would be left with billions of dollars in specialised infrastructure and a mountain of debt.
That is the hidden risk. Not the debt itself. Not the layoffs. Not even the capex. It is the fact that Oracle has signed what amounts to a 15-year structural commitment to serve a customer whose own business model is still being proven out — and done so at a scale that leaves very little margin for being wrong.
The $300 billion number is the headline. The fine print is what investors are only now beginning to read.
