AI Chip Selloff Tests Capital Discipline, Not Just AI Hype

A repricing after a huge run

The June 5 selloff in U.S.-traded chip stocks was large enough to matter, but the clean “AI bubble popped” framing is too simple. Reuters reported that chipmakers lost more than $1 trillion in market value, with Nvidia, Micron, AMD, Broadcom, and Marvell among the hardest hit. The PHLX Semiconductor Index was down almost 8.5% in afternoon trading and more than 10% over two sessions, but it was still up about 75% year to date, according to the same Reuters report.

That matters. A violent drawdown after a huge run is not the same thing as proof that end demand has collapsed. It is evidence that investors were pricing in very little room for disappointment.

Broadcom was the immediate trigger. Its shares fell after quarterly revenue and AI chip guidance came in slightly below elevated expectations, and after CEO Hock Tan did not raise the company’s prior 2027 AI chip sales forecast. Reuters said Broadcom’s Q2 revenue missed Wall Street’s $22.27 billion estimate, while Q3 AI chip revenue guidance of $16 billion was below Visible Alpha’s $16.36 billion estimate.

But Broadcom’s own June 3 results were not weak in the ordinary operating sense: record Q2 fiscal 2026 revenue of $22.187 billion, up 48% year over year; AI semiconductor revenue of $10.8 billion, up 143%; free cash flow of $10.262 billion; and Q3 revenue guidance of about $29.4 billion. The market was not punishing a shrinking supplier. It was punishing a stock and sector priced for a smoother ramp.

Rates make the AI buildout harder to justify

The macro backdrop also mattered. The June 5 U.S. jobs report showed nonfarm payrolls up 172,000 in May and unemployment at 4.3%, reviving concern that the Federal Reserve could stay tighter or even hike later in 2026. Higher rates change the math for AI infrastructure. Data centers, GPUs, networking, power procurement, cloud leases, and semiconductor capacity are capital-heavy bets. When financing costs rise, vague “AI optionality” is worth less than contracted revenue, high utilization, and visible margins.

That is the practical issue behind the selloff. The question is not whether AI chips are useful. It is whether the 2026-2028 investment wave earns enough return after depreciation, component inflation, power constraints, and customer concentration risk.

This is especially important because semiconductors had become a major driver of the broader equity rally. A chip repricing is therefore also a test of index concentration, not just a referendum on Nvidia or Broadcom.

Demand evidence remains real, but not self-validating

The strongest bearish version also overreaches. Big buyers are still signaling enormous spending. Reuters reported Bridgewater Associates’ estimate that Alphabet, Amazon, Meta, and Microsoft would invest about $650 billion in AI-related infrastructure in 2026.

Microsoft said on its April 29 earnings call that quarterly capex was $31.9 billion, roughly two-thirds for short-lived assets such as GPUs and CPUs. It guided to more than $40 billion of Q4 capex and about $190 billion of calendar-2026 capex. Microsoft also said Azure demand continued to exceed available capacity and that it expected to remain capacity-constrained at least through 2026.

Meta likewise raised its 2026 capex guidance to $125 billion-$145 billion from $115 billion-$135 billion, citing higher component pricing and additional data-center costs.

Those are not signs of vanished demand. But capex is not the same as proven economic return. The useful test now is whether spending turns into durable workload revenue across training, inference, memory, networking, and custom silicon — not just one more purchase cycle from a small set of hyperscalers.

The selloff is healthy only if it forces that discipline. Investors should watch 2026 capex execution, GPU and HBM availability, data-center power access, cloud gross margins, free cash flow, and customer concentration. The uncertainty is not whether AI infrastructure is being built. It is whether enough of it will be used profitably before higher rates and depreciation start taking their share.