Broadcom's $3.5 Trillion Bet: The Market Is Pricing Its Own AI Forecast as Fiction

Broadcom's most recent quarter ended with AI chip revenue up 221% year over year to $16.7 billion, and management guided the next quarter to climb again to $21.7 billion. The same management then published a multi-year ladder: about $58 billion of AI revenue in fiscal 2026, roughly $115 billion in fiscal 2027, and around $230 billion in fiscal 2028. Run those numbers through the market's usual multiple and Broadcom stops being a very large chipmaker and starts being a $3.5 trillion company — the neighborhood of Microsoft, just below Apple and Nvidia.

So here is the strange part. When Broadcom reported those results, investors sold. The stock fell in extended trading on September 2, and today it sits near $364, about 26% below its 52-week high of $495, up only about 5% year to date while the S&P 500 has gained 12%. Record numbers, punished stock. That divergence is the disagreement underneath this article, and it is worth putting in writing before events move.

The contract: roughly $3.5 trillion by October 2028

Broadcom trades near a $1.7 trillion market cap today. My call is specific: by the end of fiscal 2028 — the year that closes in October 2028 — Broadcom's market value reaches roughly $3.5 trillion, about a double from here. That lands it alongside Microsoft's roughly $3.55 trillion, though short of the $5 trillion club held by Apple and Nvidia. It would, in plain terms, make Broadcom one of the world's most valuable technology companies.

Let me be honest about the conviction. I do not think this is a certainty; I rate the $3.5 trillion outcome at roughly 40%. The reason it is a bet with any edge at all is the gap between what management has already committed to and what the price currently pays for. A $1.7 trillion valuation is what the market reserves for a company it expects to keep growing, but not to triple its AI revenue in two years. The crowd that drove the drawdown from $495 to $364 is pricing the ladder as mostly fiction — or betting that the multiple will be crushed before the revenue ever lands.

The clock and what has to happen

The mechanism is a three-link chain, and each link has a dated check.

Link one — the quarterly prints. AI chip revenue must keep compounding the way it is guided. The first test is the fiscal fourth quarter, where management is targeting $21.7 billion of AI revenue, up 236% year over year. If that land as guided, the $58 billion full-year figure is on track, and the arithmetic for $115 billion in fiscal 2027 is coherent.

Link two — the customers must stay or grow. Broadcom designs custom accelerators for a group of frontier AI labs that management has expanded to six names, and it has said demand for both custom chips and the networking that ties them together outpaces supply. KeyBanc analysts expect Anthropic and OpenAI to become Broadcom's largest customers in fiscal 2028. This is the link that can break in either direction: new names scaling up force the re-rating, while a lost hyperscaler program blunts it.

Link three — the multiple must hold. This is the fragile link, and it deserves the most skepticism. Even if Broadcom delivers the revenue ladder, a forward multiple of roughly 15 times on "this is an AI peak" logic would not produce $3.5 trillion. The bet is not only that the numbers arrive; it is that investors keep paying for them. Management has helped its own case here by locking up the inputs that usually cap growth — leading-edge wafers, high-bandwidth memory, and substrates — through fiscal 2028.

What kills the call

The honest countercase is competition on the customer link. Google recently expanded its TPU program with Marvell, and analysts have flagged that Broadcom could lose share on Google's accelerator roadmap while staying in it. With so much of the value tied to a handful of hyperscalers, one customer shifting at scale does real damage. And the entire edifice depends on AI infrastructure spending that can turn on a single hyperscaler's guidance call — the same kind of announcement that could force the skeptical crowd to capitulate in the other direction if it comes out strong.

So the kill switch has a number. Management has previously said fiscal-2027 AI revenue would be "over $100 billion," and it has since raised that to roughly $115 billion. If the December fiscal-Q4 report shows AI revenue falling short of the $21.7 billion guide, or if management trims the $115 billion target, the call is dead: the market will use that as license to cut the multiple, and $3.5 trillion stays out of reach. If the December report instead delivers the guided ramp and re-affirms fiscal 2027, the gap between the market's $1.7 trillion and what the ladder implies starts to close, and the forced re-buying begins.

Mark the calendar for December. The number on the page that matters is whether AI chip revenue meets $21.7 billion — and whether the company still says $115 billion comes next. That is the date the forecast stands or fails.