Microchip Says the AI Bubble Is Its Bull Case

Sanghi calls the data center unwind while his own supply chain is being cannibalized by it — and walks back the margin promise he made a quarter ago

MCHP · Earnings Call · 2026-11-05

Steve Sanghi says the quiet part out loud

Most semiconductor CEOs this earnings season described the data center build as strong and stable. Microchip's chairman went the other way — unprompted, near the end of the Q&A, and flatly: data center is in the bubble. He had already put the market's own tape into words minutes earlier: “the second derivative on data center has to come down substantially. And I think you see market voting on it. To me, it's not new. I think it was predictable.” — Steve Sanghi, Chairman · 2026-11-05The tape is indeed voting. The global theme cluster AI data centers is a 90-day decliner at roughly -31%, with 57 negative constituents against 7 positive — one of the broadest rotations in this dataset, joined by adjacent plumbing themes like high bandwidth memory and co-packaged optics. Sanghi is not predicting the unwind so much as narrating one already in progress, and he is doing it from a stock that has already fallen 31% in ninety days and 37% from its May 6 peak of $102.92.

A shortage caused by somebody else's boom

What makes this more than a soundbite is the inversion he built on top of it. Microchip's binding constraint this quarter is not demand — bookings were strong, book-to-bill "well above 1," and October backlog kept building. The constraint is that AI has taken the supply. Lead times on the standard catalog, historically 4-to-8 weeks, have stretched to what management now calls new lead times of 6 to 32 weeks. The mechanism is explicit and unusually concrete:“some of that capacity at our major OSAT suppliers has been displaced. AI companies, you know which ones, came in and took over the whole factory and they pushed everybody out” — Steve Sanghi, Chairman · 2026-11-05He later added that test capacity had been "confiscated by the AI crowd," that minimum batch size demands at some assembly houses are now larger than Microchip's volume on certain packages, and that requalifying a new OSAT takes roughly six months. So the punchline:“I think a deceleration on the data center will probably have a positive effect on our business. Because, you know, some of that capacity frees up.” — Steve Sanghi, Chairman · 2026-11-05That is a rare and genuinely contrarian framing — a $35B broad-line chipmaker positioned as a beneficiary of an AI correction because it never had the upside in the first place. Data center is only 16.7% of sales, and half of that is catalog analog, power, timing and security. The non data center franchise — industrial +50.3% year-over-year, aerospace and defense +41%, automotive +19.3% — is doing the heavy lifting. Even the capacity constraint story cuts both ways: he argues the squeezing is a transient requalification problem, not a multi-year capital hole, and explicitly ruled out revisiting a 300-millimeter fab.

The margin walk-back is the real damage

Strip out the noise and the quarter was fine. The $53 million of infrequent items — two lumpy patent licensing deals, a $21M distribution price-protection recapture, a $10M customer contract — all landed at 100% gross margin. Excluding them, Sanghi argued the December guide is up about 2.3% sequentially against a normal seasonal decline of 3% to 5%, and about 1.5% excluding the Halo and Virago acquisitions. Product backlog and bookings, he insisted, are not the problem. Revenue of $1.622B, up 42.2% year-over-year, is real.Total revenue has climbed from $970M to $1.622B across six quarters.But the gross margin is. One quarter ago, on the August call, the message was that ~66.5% would hold:“full quarter of increased prices will balance some of these one-time costs that you talked about... and gross margin could be still in that range, but do not expect it to go higher.” — Steve Sanghi, President and Chief Executive Officer · 2026-08-06His CFO went further, saying they were “comfortable that we're going to be in this -- about this range for at least the next few quarters.” — J. Bjornholt, Chief Financial Officer · 2026-08-06 Fourteen weeks later:

I think that turned out to be more aggressive than the reality has turned out. So we have backed off from that assumption.
Steve Sanghi, Chairman · 2026-11-05

December gross margin guides to 64%–65%, and the language has been re-anchored to the long term model of 65% — not above it. The culprits are the ones that always bite in a supply-constrained upcycle: unsupported high margin products that cannot ship, lower internal utilization as assembly and test deliberately slow down, and a fresh round of foundry and OSAT price increases landing in January that management says it intends to absorb rather than pass on. Non-GAAP gross margin was 66.5% in September and is guided back to 64–65% for December. The operating leverage story is intact — operating margin has recovered from -10% to 22.7% — but the slope just got flatter than investors were promised.

The setup, and the things that did not change

Two threads carried over almost verbatim. First, the supply squeeze is not new: on the August call Sanghi was already explaining that “many of our internal products get assembled and tested at the subcontractors on advanced packages and different test platforms. And that's where we have some of the constraints for the internal fab capacity.” — Steve Sanghi, President and Chief Executive Officer · 2026-08-06 The difference now is that the squeeze has spread from foundry to OSAT to substrates to internal packaging, and it is costing revenue rather than merely threatening to. Second, capital returns remain frozen. Net debt to adjusted EBITDA fell to 2.36x from 2.85x, but the board still wants it lower; Eric Bjornholt said plainly that the dividend stays put and buybacks wait until next fiscal year at the earliest. There is no buyback, no raise, and the effective net cash position remains deeply negative at -$5.1B.That is the tension in the story. The business is compounding — 24 PCIe Gen 6 switch and retimer design wins accumulated and a data center unit heading toward $1B in calendar 2026 — while the valuation has been gutted. At roughly $65 a share, on Sanghi's own conservative fiscal-2028 math of "over $4 of earnings," the stock trades near 15x. Price to operating income has compressed from 1,175x to 62x. His closing line was as close to a pitch as a CEO gets: he was, he said, tempted to buy the stock himself.Whether that reads as conviction or capitulation depends on one number the company refused to give — how much of the unsupported backlog is AI-adjacent versus the industrial and defense recovery Sanghi swears is untouched. He did concede the point that matters for the bear case: the AI bubble bursting is only cleanly good news for Microchip if the freed-up capacity arrives faster than any second-order hit to industrial demand and pricing. Where the tape and the transcript agree is that the AI premium is coming out of semiconductor multiples. Where they diverge is whether Microchip's 37% drawdown prices that correctly, or overprices it against a franchise that grows faster if the bubble deflates.