Broadcom Jalapeño Launch|Director Sold 1M Shares the Same Day

The Best AI Chip News in Broadcom's History Came With a Problem

Broadcom shares are down 16% over the past month. That number matters because the cause is not a miss — Q2 AI semiconductor revenue came in at $10.8 billion, up 143% year over year, and Q3 guidance landed at $16 billion, a 200%-plus increase. The stock fell anyway. Then on June 24, Broadcom and OpenAI unveiled Jalapeño, the first custom inference chip OpenAI has ever built, co-designed with Broadcom from the ground up in nine months. Broadcom CEO Hock Tan told Reuters the chip matches Nvidia's Blackwell in performance and cuts inference costs by roughly 50%. Forty-four Wall Street analysts carry buy ratings with a consensus price target of $524. The stock is at $377. The gap between those two numbers is the question this video is about — and the place to start is not the chip. It is the fact that on the same day Jalapeño was unveiled, Broadcom Director Henry Samueli disposed of more than one million shares. The most informed person in the room sold on the day of the most significant chip announcement in the company's history. The provisional explanation is not that Samueli is wrong. It is that the milestone the market is celebrating and the milestone that matters to him may not be the same one.

What the Insider Sees That the Analyst Consensus Does Not

Wall Street's $524 consensus target rests on a specific assumption: that Broadcom's AI chip revenue will reach $100 billion in fiscal 2027, and that the company can sustain margins while scaling to that number. The $16 billion Q3 guidance is not the problem — 200% year-over-year growth is extraordinary. The problem is that $16 billion came in below the $17.2 billion that analysts had modeled. That $1.2 billion miss is small in absolute terms but large in what it signals. Broadcom's forward P/E sits at 36 times next year's expected earnings, a multiple that prices in flawless execution. A guidance miss at the Q3 level on a name trading at that premium is not a rounding error. It is the first crack in the projection that justifies the valuation. Now hold the margin piece. Broadcom CEO Hock Tan told Reuters that profit margins on custom AI chips are currently lower than margins on Broadcom's other products — specifically networking switches — because the AI chips require large amounts of high-bandwidth memory from SK Hynix and Samsung, and the cost of that memory is squeezing gross margin. Broadcom's 68% adjusted EBITDA margin is the structural proof point analysts cite. What Tan confirmed is that the AI chip product line that is supposed to drive $100 billion in revenue by 2027 is also the line with the compressed margins. A director who has watched Broadcom's margin profile for decades knows exactly how hard it is to sustain 68% EBITDA as the revenue mix shifts toward a lower-margin product. Samueli's sale does not prove the thesis is wrong. It introduces a question the buy ratings do not answer: at what margin does $100 billion in AI chip revenue actually land?

The Jalapeño Revenue Path and the Variable That Decides It

Jalapeño begins deployment in late 2026 in small prototype form, scales through 2027, and goes full production in the first half of 2028. That timeline means Jalapeño contributes essentially nothing to Q3 FY2026 revenue, which closes around August 2026. The $16 billion Q3 AI chip target is entirely a function of Broadcom's existing XPU customers — the six hyperscalers including Google, Meta, and now OpenAI on earlier-generation silicon. Jalapeño is a 2027–2028 revenue story dressed up as a 2026 announcement. The capital market reaction treated the unveiling as a near-term earnings catalyst. It is not. What Jalapeño actually does for the near-term investment case is different and more precise. It confirms that OpenAI is locked into Broadcom's silicon for at least two chip generations — Tan cited a multi-generation roadmap and visibility "all the way to 2028." That extends the revenue backlog, which Broadcom disclosed at $30 billion in the most recent quarter. But the transmission from backlog to earnings runs through a specific choke point: high-bandwidth memory allocation. Tan was direct — SK Hynix and Samsung supply is the constraint. If HBM supply tightens or memory prices spike, Broadcom's gross margin on the Jalapeño-class product will compress further. The variable that actually decides whether the $100 billion 2027 target lands at a margin the stock's multiple can support is not OpenAI's commitment. It is HBM pricing in the back half of 2026.

The Two Conditions That Separate an Entry from a Trap

The counter-evidence against the full bull case is the margin compression Tan himself disclosed, the Q3 guide that missed estimates by $1.2 billion, and the insider sale at the moment of maximum positive sentiment. None of those individually breaks the thesis. Together they introduce a conditional: the thesis holds if Broadcom can demonstrate that AI chip margins expand as Jalapeño scales, not compress further. For a holder who entered above $450 and is sitting on a 16% drawdown, the question is whether the Q3 print — due around September 2026 — confirms the $16 billion figure and shows gross margin stability or improvement in the AI chip segment. If Q3 delivers $16 billion with flat or expanding gross margin in semiconductors, the stock's multiple is justified and the Samueli sale reads as personal portfolio rebalancing. If Q3 lands below $16 billion or margin contracts further, the AI chip revenue path to $100 billion looks slower and more expensive than modeled — and the 36x forward P/E becomes an anchor, not a floor. For a watcher who has not entered, the earliest leading signal is not Q3 earnings but HBM pricing data from SK Hynix and Samsung over the next 60 days. A memory price increase signals margin pressure before the quarterly results confirm it. A memory price decline or allocation expansion signals that Broadcom's cost structure on Jalapeño-class chips is improving. Jalapeño's reveal was real. The margin question underneath it is also real. The move becomes an entry when HBM pricing holds or falls and Q3 gross margin in semiconductors holds or expands. It becomes a trap if the memory cost structure worsens while the stock continues pricing in 2028 earnings at 2026 multiples.