Broadcom’s $349 Billion Wipeout Exposes Cracks in the AI Infrastructure Trade
A $1.2 billion guidance miss triggered a 15% after-hours plunge, signaling demand normalization and valuation reset risks across the semiconductor complex.
Broadcom erased $349 billion in market value during after-hours trading on 3 June 2026 after projecting third-quarter AI chip revenue below analyst expectations, despite posting record quarterly results that beat estimates on every metric.
The selloff wasn’t about present performance — second-quarter AI semiconductor revenue hit $10.8 billion, up 143% year-over-year, topping forecasts. The trigger was future disappointment: third-quarter AI chip guidance of $16 billion missed the consensus estimate of $17.2 billion by roughly 7%. Shares dropped 15.4% in extended trading, wiping out months of gains in minutes.
What spooked markets wasn’t the $1.2 billion shortfall alone. CEO Hock Tan refused to raise Broadcom’s full-year fiscal 2027 AI revenue target beyond “in excess of $100 billion” — language unchanged from prior guidance — even as the company reported receiving over $30 billion in AI chip orders during the quarter while delivering only $10.8 billion. That gap, per Tan, reflects customers securing future supply rather than immediate demand, with orders contingent on preparatory work including power infrastructure buildouts.
The Demand Normalization Signal
Broadcom’s stumble arrives as hyperscaler Capital Expenditure reaches historic extremes. The five largest US Hyperscalers plan combined capex of $660-690 billion in 2026, according to CreditSights, consuming nearly 100% of operating cash flows versus a decade-long average of 40%. Microsoft alone carries an $80 billion Azure backlog, yet GPUs sit idle due to power grid constraints.
The revenue guidance miss suggests AI Infrastructure spending is hitting physical and financial bottlenecks. Tan acknowledged on the earnings call that customer orders increasingly depend on securing power capacity before deployment — a multi-quarter process that decouples bookings from near-term revenue recognition. Free cash flow of $10.3 billion in the quarter (46% of revenue) remains robust, but forward visibility has clouded.
“The bookings that are coming are not for immediate delivery. Some of these orders reflect their desire to secure future supply, but they all acknowledge the reality that significant preparatory work—such as securing power.”
— Hock Tan, President and CEO, Broadcom
Valuation Reset Risks Across the Complex
Broadcom’s 40% year-to-date gain through 2 June reflected consensus belief that AI infrastructure capex could sustain triple-digit growth indefinitely. The after-hours collapse suggests that assumption is being repriced. Semiconductor stocks anchored 2024-2025 market rallies on the premise of a multi-year “supercycle” driven by generative AI workloads, with the global industry projected by Deloitte to reach $975 billion in 2026, up 26% year-over-year.
But if the largest custom AI chip supplier serving hyperscalers can’t meet growth expectations even with $30 billion in order backlog, the cyclical nature of semiconductor demand is reasserting itself. Analog chip markets are already nine quarters into an inventory correction cycle, per BNP Paribas, with restocking only beginning now — a reminder that even AI-adjacent segments remain exposed to traditional boom-bust dynamics.
| Metric | Broadcom Guidance | Analyst Consensus | Miss |
|---|---|---|---|
| Q3 AI Revenue | $16.0B | $17.2B | -7.0% |
| FY2026 AI Revenue | $56.0B | $57.6B | -2.8% |
| FY2027 AI Target | >$100B | (No raise) | Unchanged |
Geopolitical and Supply Chain Exposure
Broadcom’s vulnerability extends beyond demand cycles. The company derives roughly 35% of revenue from China, exposing it to tariff escalation and export control tightening. Production depends heavily on TSMC, which controls 70% of the global foundry market and operates at full capacity utilization for Broadcom’s custom AI accelerators. Any geopolitical friction around Taiwan or US-China semiconductor restrictions directly threatens output.
Tan maintains competitive insulation in custom-optimized transport (COT) silicon, stating in March that “we will not see competition in COT for many years to come.” But moat depth matters less if hyperscaler customers delay deployments due to infrastructure constraints rather than competitive dynamics.
Broadcom’s custom AI accelerators serve a concentrated customer base of hyperscalers (Google, Meta, ByteDance) building proprietary silicon to reduce reliance on Nvidia. These chips require 18-24 month design cycles and multi-billion-dollar upfront commitments, making revenue lumpy and highly dependent on customer-specific deployment timelines.
What to Watch
Regular-session trading on 4 June will confirm whether the $349 billion market cap loss holds or if buyers step in at lower valuations. More structurally, watch for evidence of hyperscaler capex guidance cuts in upcoming Earnings — Microsoft, Google, and Amazon report throughout June. Any acknowledgment of power or deployment delays will ripple across semiconductor suppliers.
Monitor TSMC’s July earnings call for commentary on advanced packaging capacity and customer order patterns. If Broadcom’s experience reflects broader demand softening, TSMC will signal it first through utilization rate guidance. China revenue exposure becomes increasingly critical if US export controls expand — Broadcom’s 35% exposure is among the highest in the semiconductor complex.
Finally, track whether Broadcom raises its fiscal 2027 AI revenue target above $100 billion in coming quarters. Failure to do so despite record bookings would confirm that infrastructure bottlenecks are constraining revenue conversion, not just delaying it — a distinction that determines whether this is a guidance hiccup or the beginning of a broader sector repricing.