Broadcom's AI Revenue Doubled. Its Stock Still Crashed 12%. Here's Why.

Broadcom just reported the best quarter in its history. AI revenue hit a record $10.8 billion — up 143% from a year ago. Total revenue grew 48% to $22.19 billion. Profit margins hit 69%.

The stock fell 12%.

That's where we are in the AI trade right now.

What Broadcom Actually Reported

The numbers from Tuesday's earnings call were, by any reasonable standard, extraordinary. Broadcom's AI semiconductor business generated $10.8 billion in Q2 — up 143% year-over-year. Total revenue came in at a record $22.19 billion, up 48% from last year. Adjusted EBITDA hit $15.2 billion, a 69% margin that most companies in any industry would find exceptional.

AI chips now account for 49% of Broadcom's total revenue. The company that was primarily known as a networking and storage chip maker has quietly become one of the defining players in artificial intelligence infrastructure.

And yet the stock dropped 12.6% — because Broadcom missed total revenue estimates by $80 million on a $22 billion quarter, and guided Q3 AI revenue to $16 billion versus analyst expectations of $17.2 billion.

The Number That Puts This in Perspective

Broadcom's market cap before earnings was approximately $2.34 trillion. The 12.6% drop wiped out roughly $300 billion in market value — triggered by a miss of just $80 million.

That's the market destroying $3,750 for every $1 of missed revenue. Only Nvidia and Microsoft have posted larger single-stock market cap losses in the megacap era. And it happened to a company that just reported 143% AI revenue growth.

The Expectations Problem

This is the defining tension of AI investing in 2026: the growth is real, the numbers are enormous, and Wall Street has priced in something even bigger.

Broadcom maintained its full-year AI chip revenue forecast at $56 billion — up roughly 180% from fiscal 2025. Analysts had modeled $57.6 billion. The company didn't raise guidance. In a market where AI stocks are priced for perpetual acceleration, "maintained" reads as a warning.

CEO Hock Tan's long-term target of $100 billion in AI chip revenue by fiscal year 2027 is backed by a $73 billion committed customer backlog and long-term supply agreements. Those are signed contracts. But the market wanted $100 billion to become the floor, not the ceiling.

Who's Actually Buying

Broadcom isn't selling chips to consumers or small companies. Its confirmed XPU customers include Google, Meta, OpenAI, Anthropic, Fujitsu, and ByteDance — all signed up for long-term custom chip partnerships.

What Broadcom makes for these companies are custom AI accelerators — called XPUs or ASICs — designed specifically for each customer's workloads. Google's TPUs. Meta's MTIA. OpenAI's first-generation custom chip, scheduled to ship in 2027.

Tan confirmed 1 gigawatt of Anthropic TPU deliveries in 2026 alone, scaling to 3.5 gigawatts from 2027. The Google TPU deal runs through 2031. This isn't a business built on quarterly purchase orders. It's a decade-long infrastructure buildout with the most valuable companies in the world as locked-in customers.

The VMware Drag

The part of the business that actually disappointed was software. Broadcom's infrastructure software segment — which includes VMware, acquired for $61 billion in 2023 — generated $7.18 billion, missing estimates of $7.32 billion.

VMware's integration has been rocky. Broadcom pushed hard to move customers from perpetual licenses to subscriptions, creating friction with enterprise clients. Some large customers have pushed back or delayed renewals. The software miss, combined with the AI guidance shortfall, gave investors two reasons to sell on the same day.

What This Actually Means for the AI Trade

The was built on the assumption that AI infrastructure spending would keep accelerating without pause. Broadcom's report doesn't disprove that thesis — it just shows what happens when real numbers meet inflated expectations.

AI revenue grew 143% year-over-year. In isolation, that should be celebrated. In the context of what the market had priced in, it became a reason to sell.

The companies buying Broadcom's chips are deploying AI across entire categories of work and driving real enterprise demand. That demand is genuine. The question is whether the rate of growth can match a market that always looks one quarter further than the data can reach.

The Number to Watch

Broadcom guided Q3 total revenue to $29.4 billion — up 84% year-over-year — with AI chips at $16 billion. If those numbers land, the company will have grown its AI revenue by over 200% in a single year.

The stock will probably be judged not on whether that happened, but on whether Q4 guidance raises the bar high enough.

That's the AI trade in 2026. The fundamentals are strong. The expectations are stronger. Understanding how that gap plays out is the most important thing to watch in tech stocks for the rest of the year.