Three AI-linked stocks closed lower on Monday: Nvidia fell 3% from $216.61 to $213.48, AMD lost 4%, down $11.02 to $323.61, and Oracle dropped 4% from $172.96 to $165.96.
The trigger was not a fresh problem at those companies. It was a crack in OpenAI's growth story.
One number: 900 million, not 1 billion
OpenAI had set a clear internal target: ChatGPT should pass 1 billion users by the end of 2025. The number now being discussed is different: 900 million weekly active users and 50 million paid subscribers.
That is still a massive audience, but against OpenAI's own target it is a missed goal, not a beat. The more important issue is what it does to CFO Sarah Friar's earlier warning that OpenAI could struggle to pay for future compute contracts if revenue did not grow fast enough.
That warning once sounded like cautious messaging. With user growth below plan and paid conversion still limited, the contracts signed in 2025 look heavier: $500 billion with Nvidia, $300 billion with Oracle and $270 billion with AMD. Together they add up to $1.07 trillion, while OpenAI's annual revenue is still measured in the tens of billions.
On Monday, the market priced that tail risk first: if OpenAI might not be able to pay, the valuation premium on chip suppliers has to be trimmed.
The market is testing the AI capex ceiling
The uncomfortable part is that this is not only about OpenAI. Nvidia, AMD and Oracle have traded for two years on the assumption that frontier model companies will keep buying compute without limit. OpenAI is the largest buyer inside that story. If its growth curve bends, even slightly, the demand curve for the upstream compute chain has to be redrawn.
Friar's logic is especially sensitive because compute contracts are signed before the cash arrives. OpenAI is effectively pledging future cash flow for today's capacity. The structure has an echo of pre-2008 CDO logic, with GPU clusters replacing subprime mortgages as the collateral.
If OpenAI ever cannot pay, the consequences would travel fast: Nvidia's backlog would look less firm, Oracle's cloud revenue expectations would need another pass, and AMD's MI-series orders could be delayed or cancelled.
The sell-off may still be overdone
Motley Fool analyst Danny Vena argued that the sell-off was overdone, and the numbers give that view some support.
| Company | Year-on-year revenue growth |
|---|---|
| Nvidia | +73% |
| AMD | +34% |
| Oracle | +22% |
A 3% to 4% move looks emotional against that backdrop. And slower OpenAI growth does not mean AI spending disappears. It may simply move elsewhere: Google Gemini has 750 million monthly active users, while Microsoft Copilot has 150 million users. More Gemini usage still burns TPU capacity, and when TPU supply is tight, Nvidia GPUs remain the bridge.
Markets, however, do not always wait for the fundamentals. Sentiment-led repricing can last longer than the rational case suggests.
What to watch next
The Q4 IPO script is still alive. If OpenAI reports another set of missed expectations, the proposed $852 billion valuation will face a tougher market test.
Sam Altman still has cards to play. GPT-5.5 launched only weeks ago, and enterprise ChatGPT, Workspace Agents and Codex are all meant to fill the revenue gap. Enterprise ARPU can be dozens of times higher than consumer ARPU, so 5 million paid enterprise users may matter more to the financial model than 100 million free consumers.
Next quarter, investors will be watching two lines: paid enterprise seats and the revenue share from Codex and Agents. If those hold, Friar's warning remains a warning. If they do not, the $1.07 trillion of compute commitments signed in 2025 becomes the stone on OpenAI's valuation.
Six months should be enough to tell.
Sources: CocoLoop; Nvidia, AMD, and Other AI Chip Stocks Are Swooning. Blame OpenAI. (The Motley Fool)