Broadcom reported record revenue, 143% AI chip growth, and beat estimates. The market erased $1.3 trillion. The selloff reveals what the AI chip trade is actually pricing: not growth, but the second derivative of growth.
Broadcom reported fiscal Q2 2026 earnings on June 3. Revenue: $22.2 billion, a company record, up 48% year over year. AI semiconductor revenue: $10.8 billion, up 143% year over year. Earnings per share: beat analyst estimates. Adjusted EBITDA: $15.2 billion, 69% of revenue. The stock fell 12.6% the next day. By Friday June 5, the Philadelphia Semiconductor Index had posted its worst session since March 2020, and roughly $1.3 trillion in market value had been erased across the chip sector. Marvell lost 17%. Micron lost 13%. ARM, Intel, and AMD each fell between 11% and 13%. Nvidia shed roughly $200 billion.
The cause was a number that was never spoken. CEO Hock Tan guided third-quarter AI chip sales to $16 billion. Analysts had expected $17.2 billion. More importantly, Tan did not raise Broadcom's full-year 2026 AI semiconductor forecast. He held it flat. For two years, every major AI chip company had raised guidance every quarter. Broadcom was the first to stop.
The $1.3 trillion in destroyed value had nothing to do with bad results. The market was reacting to the absence of acceleration. The semiconductor trade has been priced on the second derivative for two years. Not growth, but the rate of change of growth. Each quarter, the price required a beat that exceeded the prior beat. Broadcom delivered 143% AI revenue growth and the market treated it identically to a contraction, because the guidance for next quarter fell short and the full-year target was not raised. In the mathematics of compound expectations, the first flat quarter and the first declining quarter produce the same signal. Both tell you the curve has changed shape.
Tan's response was instructive. In the same announcement, he disclosed that Broadcom would pivot to selling custom AI chips only, abandoning plans to build complete AI systems. The CEO of the company that had just demonstrated 143% AI revenue growth chose to retreat from the higher-ambition business. This was not confusion. A chip supplier growing at 30% is healthy. A systems integrator growing at 30% is failing. Tan recognized that the same growth rate signifies health or crisis depending on where the market placed the expectations bar. He moved to where the bar is lower.
Every company in the AI chip trade now faces this dynamic. Nvidia trades at roughly 23 times forward earnings. Marvell carries a premium built on Jensen Huang's prediction at Computex that it would be the next trillion-dollar company. These valuations do not require growth. They require acceleration. And every exponential in the physical economy eventually bends into a sigmoid. Fabrication capacity is finite. Power grids have throughput limits. Enterprise procurement has natural cadences. The bend is not a question of whether but when. Broadcom was first to the inflection. The penalty was $1.3 trillion across the sector for a single company's flat forecast.
The question for the next four quarters is whether the market's pricing framework can survive the transition from exponential to linear. Track which CEO raises guidance next earnings season and which holds flat. Track whether holdouts get the Broadcom treatment or whether the market recalibrates voluntarily. The historical analog is the early-2000s networking hardware cycle, when the first flat guidance at the sector's peak triggered a repricing cascade that fundamentals did not justify but valuations demanded. Broadcom's 143% AI revenue growth proved that demand is real. The selloff proved that real demand, absent acceleration, is now the worst possible answer.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)