Chip Stocks Post Worst Stretch Since May as OpenAI's Numbers Don't Match the Story Investors Were Told

For most of 2026, the chip trade ran on a simple assumption: whatever revenue number got attached to OpenAI, Anthropic or the next frontier lab, the market would underwrite the hardware buildout behind it without asking too many questions. That assumption broke down over the week of October 5-13, when the iShares Semiconductor ETF (SOXX) strung together its longest losing streak since May 2025, and 22 of the 30 stocks in the Philadelphia Semiconductor Index were sitting more than 20% below their spring highs. The proximate cause wasn't a rate shock or an export-control headline. It was a number.

The $20 billion gap

On October 8, the Financial Times reported that OpenAI had told investors its annualized revenue was nearing $50 billion at the end of September — well below the roughly $70 billion figure that had circulated in press reports and investor materials for months. That $20 billion gap matters less for what it says about OpenAI specifically than for what it says about how the entire AI-infrastructure trade has been priced: hyperscaler capex plans, chip order books and financing packages have all been built on run-rate figures that, by the market's own admission, nobody outside a small circle of investors had actually verified. Goldman Sachs is still projecting roughly $800 billion in 2026 hyperscaler capex, rising to about $1.1 trillion in 2027 — numbers that assume demand grows into the compute being built, not that the revenue underwriting it gets revised down by nearly a third.

The single-day reaction on October 8 was blunt: Intel fell about 5.3%, Micron roughly 4.8%, Broadcom 4.3%, Nvidia around 3%. SK Hynix and Cerebras each lost close to 20% over the following week. None of this erased the sector's 2026 gains — the SOX index remains up more than 70% year to date even after the slide — but it marked the first time this year that a disclosed fundamentals number, rather than a macro variable, moved the whole chip complex at once.

Broadcom's CEO calls his own customers' investors gamblers

Four days later, at the DBS Asia Future Forward Conference in Singapore on October 12, Broadcom CEO Hock Tan added a second, stranger ingredient: he publicly questioned the rationality of the people funding his company's biggest growth engine. Tan said many AI investors "have no clue what they are investing in," compared the dynamic to a casino — "your parents give you $10 of tokens, you finish it in 10 minutes and you haven't learned anything" — and specifically flagged doubts about OpenAI's long-term profitability, even while citing the same ~$50 billion run-rate figure that had just rattled the sector. He also argued that access to financing, not chip supply, is emerging as the industry's real constraint — a notable admission from the CEO of a company selling the chips.

"They set very extreme expectations only to find out that it's almost like they are in a casino... many AI investors have no clue what they are investing in." — Broadcom CEO Hock Tan, DBS Asia Future Forward Conference, October 12, 2026

The remarks are hard to read as pure skepticism, because Broadcom is one of the largest direct beneficiaries of the buildout it's describing as speculative. The company's fiscal Q3 AI semiconductor revenue roughly tripled year-over-year to about $17 billion, Q4 AI revenue is guided to around $22 billion, and Tan has set a target of $230 billion in annual AI chip revenue by fiscal 2028 — close to a fourfold increase from the current ~$58 billion annual pace. A CEO guiding to a 4x ramp while calling his customers' backers gamblers is less a warning about AI than a hedge: if the music stops, Tan is on record as having said so.

The insider-selling tell

The timing of that hedge looks better once set against Broadcom's own insider filings. Henry Samueli, Broadcom's co-founder and board chair until June 2026, sold roughly $750 million of AVGO stock across three tranches this year: about 782,000 shares in March near $320, about 654,000 shares in June near $382, and about 702,000 shares in September near $356 — a combined 2.14 million shares sold into a stock that has nonetheless trailed the S&P 500 by 11 percentage points and the Nasdaq 100 by 18 points year to date. Tan himself sold roughly $24 million of stock in early January. None of this is illegal or even unusual for long-tenured executives diversifying concentrated holdings, but it sits uncomfortably next to a public narrative in which the same company is promising investors a quadrupling of AI revenue within two years. Insiders holding the clearest view of Broadcom's AI order book were net sellers through the year Broadcom's AI story was supposedly accelerating fastest.

Why this is a different kind of correction

Every prior AI-chip wobble in 2026 has been traceable to financing stress at the edges of the ecosystem — neocloud bond yields, debt-financed capex, credit-rating pressure on companies like Oracle. This one is different: it's the first time the market reacted to a gap between what a frontier lab told investors privately and what had been priced in publicly, and it arrived packaged with a chip supplier's own CEO casting doubt on the buyers of his product. That combination — disclosed-revenue risk plus a credibility gap at the top of the supply chain, with insider selling as a side signal — is a harder thing for the market to underwrite with capex forecasts alone. Upcoming Q3 earnings from Nvidia, Micron and TSMC will now be read less for revenue growth, which has been reliably strong all year, and more for whether management's forward commentary matches what their own largest customers are quietly telling investors behind closed doors. The $70 billion number didn't just overstate OpenAI's revenue by $20 billion — it exposed how much of the chip rally has been priced off numbers nobody outside a small circle had actually checked.