KEY POINTS
- Broadcom shares cratered 15% on June 4 after the company's Q3 AI chip revenue forecast of $16 billion missed Wall Street's $17.2 billion consensus by 7%.
- The selloff reflects a market that has priced AI semiconductor stocks for perfection, punishing even double-digit revenue growth when it falls below buy-side whisper numbers.
- Traders should watch whether the $16 billion Q3 target holds as a floor and whether Broadcom's six hyperscaler customers accelerate or defer XPU orders in the back half of fiscal 2026.
Broadcom erased roughly $50 billion in market capitalization on Thursday after its fiscal second-quarter earnings report revealed an AI chip business growing at triple digits but still not fast enough for a market addicted to upside surprises. The stock closed down 15%, its worst single session since January 2025, dragging the Philadelphia Semiconductor Index lower and puncturing a tech rally that had pushed the S&P 500 above 7,600 just two days earlier.
The numbers themselves were not weak. Revenue hit $22.19 billion for the quarter ending May 4, up 48% year over year and slightly above the $22.13 billion consensus. Earnings per share landed at $2.44, beating the $2.39 estimate. AI semiconductor revenue specifically came in at $10.8 billion, a 143% surge from a year ago and now nearly half of total company sales. By any historical standard, that is an extraordinary growth clip for a company of Broadcom's size.
The Miss That Mattered
None of those backward-looking beats mattered once CEO Hock Tan delivered forward guidance. Broadcom projected $16 billion in AI semiconductor revenue for the fiscal third quarter running through July. Analysts had penciled in $17.2 billion on average, and buy-side whisper numbers ran even higher. Tan also declined to raise the company's full-year AI chip sales target of $100 billion for fiscal 2027, a figure the Street had treated as a bare minimum rather than a stretch goal.
The gap between $16 billion and $17.2 billion is roughly $1.2 billion in a single quarter. That is real money, but it represents a miss of about 7% against expectations that had been revised upward five times in the prior twelve months. The punishment was wildly asymmetric: a 7% revenue shortfall on one line item triggered a 15% decline in the entire equity.
That asymmetry tells a clear story about positioning. Broadcom had rallied more than 40% year to date heading into the print, making it one of the best-performing large-cap semiconductor names of 2026. Institutional holders had crowded into the stock on the thesis that its six core custom chip customers — a roster that includes Anthropic, Google, Meta, and OpenAI — would drive revenue acceleration through the second half of the year. When Tan signaled sequential growth rather than sequential acceleration, the unwind was immediate.
Custom Silicon Under Scrutiny
Broadcom's AI business is structurally different from Nvidia's. Where Nvidia sells general-purpose GPUs at scale, Broadcom designs bespoke AI accelerators — called XPUs — tailored to specific hyperscaler architectures, plus the networking silicon that connects them. This custom model generates high margins and deep customer lock-in, but it also creates lumpy revenue recognition. Orders from six customers, no matter how large, can shift between quarters based on data center build timelines, power availability, and internal testing cycles.
Tan acknowledged this on the earnings call, noting that AI bookings exceeded $30 billion and that the pipeline remains robust. But bookings are not revenue, and traders who bought the stock for near-term momentum do not have the patience to wait for pipeline conversion. The market wanted confirmation that the June-through-October corridor would deliver a step function in revenue. It got a ramp instead.
The read-through for the broader AI trade is significant. If Broadcom — with guaranteed demand from six of the largest AI spenders on earth — cannot deliver a blowout quarter, the bar for every other AI-adjacent name just got higher. Applied Materials, Marvell Technology, and Arista Networks all traded lower in sympathy on Thursday.
Software Weakness Added Fuel
The AI forecast miss dominated the narrative, but Broadcom's infrastructure software segment also underwhelmed. Software revenue came in roughly flat sequentially, reflecting slower enterprise spending on VMware licenses and bundled infrastructure products. That segment had been a quiet contributor to Broadcom's margin expansion story since the VMware acquisition, and any deceleration there raises questions about whether the company's non-AI business can sustain its premium multiple.
Total operating margin remained north of 40%, and free cash flow generation was strong at $8.3 billion for the quarter. Those are not the metrics of a broken business. They are the metrics of a company growing rapidly into expectations that grew even faster.
What Comes Next
The $16 billion Q3 AI target is now the line in the sand. If Broadcom meets or exceeds it on the August earnings call, the selloff will look like a buying opportunity in retrospect. If it misses again, the narrative shifts from "expectations too high" to "demand slowing," and the stock faces a deeper re-rating. Watch for commentary from Google and Meta on their capital expenditure plans during upcoming investor events — those are the customers whose spending cadence will determine whether Broadcom's $100 billion fiscal 2027 target remains credible. The July FOMC meeting also matters: if the Fed signals further delay on rate cuts, the multiple compression trade that hit high-growth tech in 2022 could resurface.
