2026-06-06 — views · Broadcom Inc.

Broadcom's 143% AI Beat Wasn't Enough: A Guidance "Miss" That Cut the Chip Sector

Broadcom posted record Q2 FY2026 results on June 3 — revenue up 48% to $22.19B, AI semiconductor revenue up 143% to $10.8B — yet the stock fell roughly 12-15% the next day because Q3 AI guidance of $16B landed under the ~$17.2B some analysts wanted. The reaction dragged the

What happened

Broadcom reported fiscal Q2 2026 results (quarter ended May 3, 2026) after the close on June 3, and on paper it was a blowout. Consolidated revenue rose 48% year over year to a record $22.19 billion, GAAP net income was $9.31 billion, non-GAAP diluted EPS was $2.44, and adjusted EBITDA hit $15.24 billion, or 69% of revenue. Free cash flow was $10.26 billion. The headline number investors care about most — AI semiconductor revenue — grew 143% year over year to $10.8 billion, ahead of the company’s own forecast.

Then the stock fell roughly 12% to 15% the following session.

Why a 143% grower sold off

The selloff was about the forward number, not the quarter. For fiscal Q3, Broadcom guided total revenue to about $29.4 billion (around 84% year-over-year growth) and AI semiconductor revenue to “over 200 percent year-over-year” growth, reaching $16.0 billion. That $16B figure landed below the roughly $17.2 billion some analysts had penciled in. After a long run in which AVGO repeatedly cleared a rising bar, an in-line-to-slightly-light AI guide read as a disappointment. On the call, CEO Hock Tan put full-year fiscal 2026 AI semiconductor revenue at about $56 billion (up ~180% from fiscal 2025) and reiterated — rather than raised — a fiscal 2027 AI target “in excess of $100 billion.” The lack of an upward revision was the trigger.

Metric	Q2 FY2026 actual	Q3 FY2026 guide
Total revenue	$22.19B (+48% YoY)	~~$29.4B (~~+84% YoY)
AI semiconductor revenue	$10.8B (+143% YoY)	$16.0B (>200% YoY)
Non-GAAP EPS	$2.44	n/a
Adjusted EBITDA	$15.24B (69% of rev)	margin ~67% op

The sector ripple

Because Broadcom is now a bellwether for custom AI silicon, the disappointment spread. The Philadelphia Semiconductor Index dropped more than 6% in the session, and peers including Nvidia, AMD, Arm and Marvell fell as investors questioned how durably hyperscaler AI spending can keep accelerating. Compounding the move, stronger-than-expected May jobs data revived rate-sensitivity in growth names the same week.

The business underneath the tape

Tan reiterated that Broadcom stays a chip-and-networking supplier rather than a systems vendor, selling XPUs (its custom AI accelerators) plus the networking silicon that clusters them. Management pointed to a small set of large customers with multi-year commitments anchoring the AI ramp. That concentration is the double-edged sword: it produces eye-watering growth rates but makes each customer’s order cadence — and the Street’s quarter-to-quarter modeling of it — unusually load-bearing for the stock.

Practitioner note

When a name has compounded on serial guide-raises, the market starts pricing the raise, not the result. A 143% grower that guides “only” to triple-digit growth can still drop double digits because the surprise — not the level — is what moves the tape. For anyone watching AI-chip names, the actionable read is to track consensus AI-segment revenue and the size of the implied beat, not just headline EPS; the gap between guide and whisper is where the volatility lives.

Under-considered angle

Most coverage framed this as “AI spending may be slowing.” The more interesting question is about revenue lumpiness in custom-silicon versus merchant GPUs. Custom-accelerator revenue is tied to a handful of hyperscaler programs with discrete tape-outs and ramp schedules, so a single program’s timing can swing a quarter’s guide by a billion dollars without saying anything about end demand. That makes Broadcom’s quarter-to-quarter AI line a noisier signal about the broader AI capex cycle than the market treated it as — the sympathy selloff in merchant-GPU names may have over-extrapolated from what is partly a scheduling artifact.