Broadcom reported Q2 FY2026 after the close tonight. Revenue $22.2 billion, up 48%. EPS $2.44, beating $2.40 consensus. AI semiconductor revenue $10.8 billion, up 143% year-over-year. Q3 guided to $29.4 billion in total revenue — above the $28.5 billion Street consensus — with AI semiconductors alone at $16 billion.
The stock fell 3% after hours to $462.
The reason isn't in the AI numbers. It's in the other $7 billion.
The Scorecard
| Metric | Expected | Actual | Verdict |
|---|---|---|---|
| Revenue | $22.11B | $22.19B | +$80M |
| Non-GAAP EPS | $2.40 | $2.44 | +1.7% |
| AI Semiconductor Rev | $10.7B | $10.8B | +$100M |
| Semiconductor Rev | $14.8B | $15.0B | +$200M |
| Software Rev | $7.32B | $7.18B | −$140M |
| Adj EBITDA Margin | ~69% | 68.7% | In line |
| Q3 Revenue Guide | $28.53B | $29.4B | +$870M |
Six lines green. One line red. The stock followed the red line.
The Software Gap
Infrastructure software — VMware, mostly — came in at $7.18 billion against a $7.32 billion expectation. That's a $140 million miss on a $22 billion revenue print. It's 0.6% of total revenue. It's also below Broadcom's own ~$7.2 billion guide from last quarter.
For a stock that closed at an all-time high of $481.57, that 0.6% shortfall was enough. Bernstein's Stacy Rasgon put it simply on CNBC: it wasn't the AI beat that mattered tonight — it was the forward AI guidance failing to exceed the most aggressive buy-side models. When you've rallied 37.7% year-to-date into earnings night, "good" isn't the trade. "Perfect" was the trade. And perfect had no room for a software miss.
The Number That Matters
Broadcom booked $30 billion in AI semiconductors in Q2. It shipped $10.8 billion. That's a 3:1 book-to-bill ratio in the AI segment alone.
This is the number the after-hours tape is ignoring. Bookings represent committed demand from hyperscalers who are building custom silicon with Broadcom — Google's TPUs, Meta's MTIA XPUs, and custom accelerators for Anthropic, OpenAI, and two unnamed customers. These aren't speculative purchase orders. They're multi-generational design wins with 18-month-plus deployment timelines.
The constraint isn't demand. It's fabrication. Every custom chip Broadcom designs still has to go through TSMC, and TSMC capacity is the bottleneck — demand is reportedly 3x supply. The 3:1 book-to-bill isn't just a demand signal. It's a manufacturing queue.
The Q3 Guide
The Q3 guidance tells you how fast Broadcom is working through that queue:
A 48% sequential jump in AI revenue in a single quarter. From $10.8 billion to $16.0 billion. That's $5.2 billion of incremental AI semiconductor revenue in 90 days. It implies TSMC allocation is expanding fast — and that Broadcom's six hyperscaler customers are accelerating their deployment schedules.
For context, I was watching whether Q3 AI guidance would exceed $13 billion. UBS, the most bullish sell-side estimate, had penciled in $13.6 billion. Broadcom guided to $16 billion — $2.4 billion above even the most aggressive analyst estimate. The buy-side, evidently, had been even more aggressive than that.
The Customer Map
Six confirmed custom silicon customers: Google, Meta, Anthropic, OpenAI, plus two unnamed. This is Broadcom's real moat — not just networking switches, but the multi-year design partnerships where hyperscalers commit to Broadcom's silicon architecture for their AI training clusters.
Meta's commitment is the most concrete new disclosure: multiple generations of MTIA XPUs with expectations to deploy 3 gigawatts of compute through end of 2028. Initial 1-gigawatt orders — including both custom XPUs and networking — begin delivery in H2 2027. That's not a purchase order. That's an infrastructure buildout denominated in power consumption.
AI networking now represents approximately 40% of AI revenue, up from 33% in Q1. That's roughly $4.3 billion in networking revenue this quarter — Tomahawk switches and 200G SerDes are scaling into production deployments across all six customers.
What the Tape Missed
The after-hours reaction is a positioning story, not a fundamental one. Here's what it looks like when you strip away the noise:
- ▲ AI revenue beat guide and grew 143% YoY
- ▲ Q3 total revenue guided $870M above consensus
- ▲ Q3 AI guide of $16B crushed sell-side estimates by $2.4B
- ▲ $30B AI bookings = 3:1 book-to-bill
- ▲ Free cash flow $10.3B (+60% YoY), 46% of revenue
- ▼ Software revenue missed by $140M (0.6% of total)
- ▼ Stock at ATH into print — perfection was priced in
Two negatives, five positives, and the negatives drove the tape. That's what happens when a stock rallies 37.7% into earnings at an all-time high. The bar wasn't "beat expectations." The bar was "beat the buy-side's most aggressive whisper number on every single line item." Software broke the streak.
The Broadcom Identity Question
There's a deeper tension in these results. Broadcom is a $22 billion quarterly revenue company. Semiconductors — $15 billion — are growing 79% annually, driven almost entirely by AI. Software — $7.2 billion — is growing 9% annually, driven by VMware subscription conversions. The market is valuing Broadcom as an AI company. But 32% of its revenue is an enterprise software business growing at a single-digit rate that just missed estimates.
As long as AI semiconductor revenue is accelerating at 143% year-over-year and bookings are 3x shipments, the market will forgive a software miss. But the question for the next twelve months: does VMware become a stable cash flow base that supports AI investment, or does it become the line item that introduces doubt every quarter? Tonight's tape suggests the market hasn't decided.
Broadcom Q2 FY2026 reported June 3, 2026 after market close. Stock closed at $481.57, traded to ~$462 after hours (~−4%). Q3 FY2026 earnings expected September 2026. All non-GAAP figures per company reporting.