Chipmakers fell on October 8 after a Financial Times report put OpenAI’s annualized revenue at roughly $50 billion as of late September, about $20 billion below the $70 billion figure that had circulated in press and market commentary just days earlier.
Shares of Nvidia closed down about 3 percent, Micron fell 4.8 percent, Broadcom dropped 4.4 percent, Intel slid 5 percent, and Super Micro Computer lost close to 5 percent. The Philadelphia Semiconductor Index sank 3.4 percent, its steepest day since mid-August. The broader market fell less, with the S&P 500 down about half a percent and the Nasdaq down roughly 1.25 percent, but the damage concentrated hard in AI-exposed names.
Why such a big reaction to a private company’s revenue? Because the whole AI trade has been built ontop of a small number of customers, and OpenAI is the largest. Nvidia’s data center revenue, Oracle’s cloud backlog, and CoreWeave’s entire business model all lean on compute contracts signed with model companies. If OpenAI’s actual revenue is materially lower than the market thought, so is the ceiling on what those companies can pay for chips and data centers.
A $20 billion gap with a footnote
The discrepancy stems from accounting method, not from an actual collapse in revenue. According to reporting that followed the Financial Times story, part of the $20 billion gap turns out to be how OpenAI and rival Anthropic count partner revenue rather than a genuine slowdown. Anthropic historically reported figures that included some partner channel revenue. Once you adjust for that, the two companies sit closer together than the raw print suggests.
What seems clearer is that OpenAI’s revenue genuinely did not grow as fast between late August and late September as many analysts had penciled in. Growth remains steep by any normal standard, but the pace at which OpenAI can afford to absorb new compute contracts is what matters for the supply chain, and that number just got revised down.
Skeptics of the selloff pushed back on the interpretation. Some analysts immediately chalked the move up to narrative rather than fact. “Broader reaction is overblown in my opinion,” one analyst wrote in a widely shared note, calling the accounting-method story a red herring. Traders did not wait to read the fine print. By Thursday afternoon the tape had already repriced the AI complex.
The bigger story is the funding gap
The revenue print lands at a particularly bad time for another reason. On October 8, The Wall Street Journal reported that Broadcom is lining up roughly $50 billion in financing for OpenAI, with Oracle also seeking an undisclosed amount. The financing is meant to pay for the compute OpenAI has contracted, not for its day-to-day operating losses. That depended on OpenAI drawing significant revenue from the same agreements the financing would fund. A revenue number $20 billion lower changes how cushions both the equity and the debt sides of that deal look.
Sit next to that and you can see why the market wobbled. A company at the center of the AI buildout has just told the market its cash flow is going to be lower than everyone had modeled, just as its supply chain partners go looking for $50 billion or more in debt financing to pay for capacity it will buy. The pieces are not independent. OpenAI’s revenue is the income statement side of a supply chain that Oracle, Broadcom and others are financing on the balance sheet side.
Selloffs like this are also a reminder that the AI trade is not one trade anymore. It is a web of interlocking balance sheets. Nvidia sits at the center. It owns a stake in some of its biggest customers. It has committed capital to a $500 billion GPU lending facility with Wall Street banks. Broadcom has announced multi-year design wins to build custom accelerators for a specific set of buyers. Oracle has signed compute agreements worth tens of billions with model companies it cannot ship product to without the chips. A single revision in one counterparty’s revenue assumptions ripples through all of them.
Which names got hit and why
The selloff was not uniform. Nvidia and Broadcom, the two companies with direct OpenAI supply contracts, fell in line with the sector. Oracle and CoreWeave, which sell the cloud capacity that houses those chips, fell harder, with CoreWeave dropping about 8 percent and Oracle down nearly 6 percent. The pattern makes sense. The companies closest to OpenAI’s cash flow, the cloud providers that would pay for chips and resale them, are most exposed to a revision in how much OpenAI can actually afford.
Memory and optical stocks fell in sympathy even though most do not sell directly to OpenAI. Samsung Electronics, in contrast, reported a record quarterly profit forecast in the same period, and shares in South Korea still closed lower on Thursday, a sign that the OpenAI story was not the only thing dragging down AI names.
What to watch next
The annualized revenue figure is a projection of current monthly revenue scaled by 12, not a done deal. Actual revenue this quarter and next will tell the real story. OpenAI is widely expected to close a new funding round at a valuation in the region of $500 billion to $800 billion, and investors will want to see a credible revenue trajectory attached to that price. Any sign that OpenAI’s growth is accelerating past the current pace, or that it is pulling forward big new compute contracts, would reset the story.
Separate headlines also hit tech this week. The Nasdaq Composite had already fallen the day before on Treasury yields and oil. President Trump said on October 8 that the US will not strike Iran before the midterm elections, pulling crude back from highs and easing some of the macro pressure. That helped equity futures on Thursday morning point to a modest recovery, but the sector is still digesting a $20 billion revenue revision.
The main question this week is not whether AI demand is real, but whether the financing behind it can keep pace with the cash flow coming in to service it. Thursday’s selloff says the market is no longer willing to take that on faith.