Broadcom Beats Revenue by $22B but Misses on AI — Stock Drops 15%

Broadcom reported $22.19 billion in Q2 FY2026 revenue — up 48% year-over-year, a record quarter, and $10.26 billion in free cash flow. Then the stock dropped 15% pre-market. The reason: AI revenue specifically missed Wall Street expectations.

This is the first time a major AI infrastructure — as explored in the economics of AI compute infrastructure — company has beaten on total revenue while disappointing on the AI component. It matters because Broadcom’s valuation — like Nvidia’s, like TSMC’s — is built on the assumption that AI spending accelerates indefinitely. When the AI line misses, the entire thesis gets repriced.

The Numbers

Revenue: $22.19B (+48% YoY). Non-GAAP EPS: $2.44. Free cash flow: $10.26B at a 46% margin. Q3 guidance: $29.4B — implying 84% year-over-year growth next quarter. By every traditional metric, this was an exceptional quarter.

But the AI semiconductor revenue line — the number that drives the multiple — came in below analyst models. Broadcom doesn’t break out AI revenue as a separate line item, but the company had previously guided to $10.7B in AI semiconductor revenue for Q2. Whatever the actual number was, the market decided it wasn’t enough.

What This Signals

Three possible readings:

Timing, not demand. Broadcom’s AI revenue is driven by custom ASIC design wins with hyperscalers (Google TPU, Meta MTIA, OpenAI, Anthropic, Apple — as explored in the interface layer wars reshaping consumer tech — ). These are project-based — revenue is lumpy by nature. A single delayed tape-out or qualification cycle can shift hundreds of millions between quarters. The $29.4B Q3 guidance (+84%) suggests the demand is there but the revenue recognition shifted.

Custom ASICs are harder to scale than GPUs. Nvidia ships the same GPU to every customer. Broadcom designs a different chip for each hyperscaler. That’s higher margins but lower predictability. As the custom ASIC business scales from $15B to $40B+, the execution complexity grows non-linearly. Miss one customer’s timeline, and the quarter misses.

The market is pricing perfection. Broadcom trades at 25-30x forward revenue. At that multiple, beating by 48% isn’t enough — you need to beat on the specific revenue line that justifies the multiple. When Nvidia reports 85% growth and the stock barely moves, and Broadcom reports 48% growth and drops 15%, the message is clear: the market expects acceleration, not just growth.

The Bigger Picture

Broadcom’s miss comes on the same morning that CrowdStrike dropped 11% despite beating estimates, PVH crashed 21% on guidance cuts, and Nasdaq futures are sliding on Iran concerns. The risk-off rotation is real — and AI stocks are no longer immune.

For the AI infrastructure thesis, this is a stress test. If Broadcom’s Q3 delivers the guided $29.4B and AI revenue reaccelerates, the Q2 miss becomes a blip. If it doesn’t, the entire AI capex narrative — the $300B+ in committed hyperscaler spending, the $75B SpaceX IPO, the $3.5T in AI mega-IPOs — faces its first real credibility check from public markets.

The SpaceX roadshow starts today. The timing is not ideal.

For the full structural map of the AI economy, read The Map of AI Redrawn on Business Engineer.