On July 9, 2026, S&P Global Ratings cut Oracle's credit rating to BBB-. One notch above junk. The reason was not a bad quarter or a product failure. It was the sheer cost of building AI infrastructure.
Oracle's capital expenditure will reach $90 to $95 billion in fiscal 2027. S&P now forecasts a free operating cash flow deficit of nearly $42 billion. The agency said it "underestimated the scale of the investments required to expand the AI business."
This is not an Oracle story. It is a preview of what happens when the AI cost escalation hits the balance sheet of the companies building the compute layer. And the companies building the compute layer are betting that advertising revenue will eventually cover the bill.
The problem: the ad revenue is projected to miss by 90%.
The Numbers That Bent a Credit Rating
Let us walk through the mechanics.
Oracle's cloud infrastructure business made up 27% of revenues in fiscal 2026. S&P projects it will reach almost 60% by fiscal 2028. That sounds like growth. What it actually means is Oracle is shifting from a stable, recurring enterprise software model to a capital-intensive infrastructure business where returns come in over years.
The agency added $260 billion in lease commitments to its debt calculation, plus $13 billion in purchase obligations tied to data center power arrangements. These are not conventional borrowings. They are the financial reality of building AI compute at scale.
And then there is the concentration risk. OpenAI makes up roughly half of Oracle's $638 billion in remaining performance obligations. S&P wrote that OpenAI's ability to meet its contractual obligations "depends on continued AI tailwinds and its models remaining market leaders." If OpenAI cannot pay, Oracle gets stuck with massive data center leases it cannot exit.
The Ad Revenue Bridge
Here is where marketing enters the picture.
OpenAI projected $2.5 billion in ad revenue for 2026 and $100 billion by 2030. According to Emarketer data reported by Adweek, the entire U.S. chatbot ad market, including ChatGPT, Microsoft Copilot, Google AI Mode, and Amazon Alexa for Shopping, will generate less than $1 billion this year and just $5.41 billion by 2030.
That is a 90% miss on a $100 billion target. And that $100 billion target is not just an investor pitch. It is the financial bridge between spending $90 billion on data centers and actually making the economics work.
If the ad revenue falls short, someone has to eat the difference. The infrastructure companies have already borrowed against it. The AI labs have already raised capital on the assumption of it. The ChatGPT ad revenue collapse is not a side story. It is the missing revenue that makes the entire AI compute stack insolvent on a long enough timeline.
What Happens When the Bridge Is Too Short
The math is simple and uncomfortable.
Oracle is spending $90 billion to build compute capacity. OpenAI, its largest customer, is betting on $100 billion in ad revenue to justify what it pays Oracle. The analyst consensus says the ad market will deliver $5.41 billion. That leaves a gap of roughly $95 billion that has to come from somewhere.
There are three places it can come from. Higher subscription prices for AI tools. Higher advertising rates for brands that want to reach AI users. Or lower payments to the infrastructure layer, which triggers the exact default scenario S&P described.
The most likely path is the one marketers will feel first. AI platforms will push harder on advertising because they have no choice. The ChatGPT ads network will expand. Google's AI Mode will integrate more ad formats. Every AI surface will become an ad surface, not because users want ads in their AI conversations, but because the compute bill has to get paid.
This is already happening. OpenAI started its ad trial in February. By May, it was pitching advertisers at Cannes with the framing that ChatGPT users are "super intentional." The urgency is not market demand. It is financial necessity.
The Cost Pass-Through Problem
Here is what should worry marketing leaders.
The AI infrastructure layer is running on debt that was underwritten based on advertising projections that analysts now say are off by 90%. When those projections miss, the cost does not disappear. It gets pushed down the stack.
Infrastructure companies raise lease rates. AI labs raise subscription prices and ad rates. Brands pay more for the same or worse inventory, because the pricing is driven by the need to service debt, not by the value of the impression.
This is the AI budget illusion in its sharpest form. Marketing teams are being told that AI tools will reduce costs. The financial reality is that AI tools are being priced to cover a $42 billion cash deficit that S&P can see and marketers cannot.
A similar pattern emerged in the token cost shock earlier this year, where inference costs exceeded entire marketing team salaries. That was a warning. The Oracle downgrade is the confirmation.
What Brands Should Watch
The Oracle S&P report is a leading indicator. When the company building the compute layer gets downgraded because of AI spending, and the company buying the compute layer is projected to miss its revenue target by 90%, the cost will find its way to the end of the chain. Brands are the end of the chain.
Watch ad rate increases on AI platforms. If ChatGPT, Google AI Mode, and other AI surfaces start pushing aggressive ad pricing, it is not because the inventory is valuable. It is because the infrastructure debt needs servicing.
Watch subscription price increases across AI tools. When the per-seat cost of an AI marketing platform goes up 30% in a single quarter, the reason is not product investment. It is the compute bill.
And watch vendor consolidation. The S&P report noted that SpaceX is now leasing compute capacity to Anthropic and Alphabet, and Meta may follow. More suppliers means more competition, but it also means more debt spread across more companies, all chasing the same ad revenue pool that analysts say is a fraction of what was promised.
The uncomfortable truth is that the AI infrastructure buildout was financed on the assumption that marketing budgets would eventually absorb the cost. The ad revenue projections were the collateral. Now the collateral is worth 10 cents on the dollar, and the debt is still on the books.