Broadcom Inc. delivered what should have been a landmark fiscal second-quarter 2026 earnings report on Wednesday: record revenue of $22.2 billion, up 48% year-over-year and accelerating from Q1's 29% growth rate; AI semiconductor revenue of $10.8 billion that surged 143% from a year earlier and came in above the company's own forecast of $10.7 billion; record free cash flow exceeding $10 billion in a single quarter for the first time in the company's history; and an EBITDA margin of 69% that expanded from the 68% both management and analysts had guided. The market's response was a 14% plunge in after-hours trading.
The disconnect between the report's underlying strength and the stock's reaction is the defining signal of the current AI infrastructure investment cycle: when a company is priced for perfection after a sustained multi-quarter run to record highs, even blockbuster results trigger selling if the forward guide fails to clear the highest buy-side expectations. For Broadcom, the trigger was Q3 AI revenue guidance of approximately $16 billion — a figure that implies over 200% year-over-year growth and would be a record quarter — but fell short of the $17.2 billion that the most aggressive institutional estimates had modeled. CEO Hock Tan confirmed on the earnings call that the company also elected not to raise its full-year AI revenue target, which the market interpreted as a ceiling signal rather than a floor.
The Q3 total revenue guide of $29.4 billion — representing 84% year-over-year growth — is itself an extraordinary forward commitment for a company of Broadcom's scale. Infrastructure software, which contributed the second-largest revenue segment after AI semiconductors, came in slightly below estimates, providing the one concrete data miss that bears could point to in an otherwise exceptional report. Non-GAAP EPS of $2.44 topped the roughly $2.32 consensus.
Broadcom's custom AI accelerator model — designing XPUs co-engineered with specific hyperscaler clients rather than selling off-the-shelf GPUs — creates the high switching costs and multi-year revenue visibility that differentiate its AI business from Nvidia's more commoditized GPU approach. CEO Hock Tan described demand for custom AI accelerators and AI networking as driving the Q2 beat, and noted that the networking side of the AI data center — moving data between thousands of chips — is growing as fast as the compute side. The company's three major hyperscaler XPU customers are widely understood to include Alphabet, Meta, and Apple, though Broadcom does not name them by contract.
The Thursday session saw Broadcom shares extend the after-hours decline in regular trading, dragging the broader semiconductor sector lower and contributing to a rotation in which the Dow Jones Industrial Average hit a fresh record high of 51,561.93 — led by UnitedHealth, JPMorgan, and Walmart — while the Nasdaq lost 0.09% and the S&P 500 rose just 0.41%. The pattern illustrated the sector rotation dynamic that Broadcom's earnings triggered: institutional capital moving from chip names into non-tech large caps on valuation concerns.
Friday's Broadcom price action will be closely watched as a gauge of whether the Thursday selloff represented panic selling at a technically extended level — and therefore a potential buy-the-dip opportunity for investors with conviction in the custom-silicon AI thesis — or the beginning of a broader multiple compression in AI chip names as the market recalibrates expectations ahead of Nvidia's August earnings cycle.