Broadcom posts $8.4B AI revenue

Broadcom’s latest numbers show how much of the AI buildout now sits outside the GPU itself. In first-quarter fiscal 2026 results released on March 4, the company said AI semiconductor revenue reached $8.4 billion, up 106% from a year earlier, driven by demand for custom AI accelerators and AI networking. Broadcom also guided to $10.7 billion in AI semiconductor revenue for the second quarter, indicating that the run rate is still rising. That matters because Broadcom is not selling a single AI product category. The company is supplying both custom silicon for hyperscalers and the networking gear needed to connect very large clusters, a mix that CEO Hock Tan said was behind the quarter’s outperformance. Total company revenue for the quarter was $19.311 billion, up 29% year over year, while adjusted EBITDA was $13.128 billion, or 68% of revenue. (investors.broadcom.com) ### Why does the $8.4 billion figure stand out? The March 4 earnings release put the AI number in unusually direct terms. Tan said first-quarter AI revenue was “above our forecast” and tied the increase to “robust demand for custom AI accelerators and AI networking.” Broadcom did not frame the gain as a one-off order spike; it paired the result with a second-quarter AI semiconductor forecast of $10.7 billion. (investors.broadcom.com) A 106% year-over-year increase also means AI is growing much faster than Broadcom’s overall business. Consolidated revenue rose 29% in the quarter, well below the AI segment’s pace, according to the company’s release. CNBC, citing the results, reported that Tan had projected a doubling of AI revenue in December and that the quarter exceeded that mark. (investors.broadcom.com) ### Where does the reported $73 billion backlog fit in? Broadcom’s reported AI backlog points to how far hyperscaler commitments now extend beyond one quarter’s shipments. Secondary reports tied to the company’s earnings cycle said Broadcom’s AI order book had climbed above $73 billion, giving investors a measure of future demand tied to custom silicon programs. Because Broadcom’s official first-quarter release did not spell out that backlog figure in the text surfaced here, the backlog is best treated as part of the broader earnings discussion rather than the headline release itself. (investors.broadcom.com) The significance is practical: custom accelerator programs are booked over longer cycles, and those programs require matching investments in switching, optics and interconnect. That is an inference from Broadcom’s product lineup and reported demand, not a company quote. (stockanalysis.com) ### What does Tomahawk 6 tell us about Broadcom’s AI position? Broadcom said on June 3, 2025 that it had begun shipping the Tomahawk 6 switch series, which it described as the world’s first 102.4 terabits-per-second switch chip. The company said the product doubles the bandwidth of any Ethernet switch then on the market and is aimed at scale-up and scale-out AI networks. The chip is designed for AI clusters with more than one million XPUs, Broadcom said, and includes support for 100G and 200G SerDes as well as co-packaged optics. (investors.broadcom.com) Ram Velaga, Broadcom’s senior vice president and general manager of the Core Switching Group, said customer and partner demand had been “unprecedented.” Bloomberg Intelligence analyst Kunjan Sobhani, quoted in the release, said the product addressed a network bottleneck as AI clusters scale. (broadcom.com) ### Why are networking and custom chips being discussed together? Broadcom’s own wording links the two. Tan said first-quarter AI growth came from “custom AI accelerators and AI networking,” not from one category alone. That pairing reflects how hyperscalers are increasingly buying full AI system ingredients: accelerator silicon, switch silicon, optical connectivity and the software stack needed to keep large clusters utilized. (broadcom.com) Tomahawk 6 also shows why networking revenue can rise with accelerator deployments. A larger training or inference cluster needs more bandwidth, more ports and tighter congestion control, especially as operators push Ethernet deeper into AI fabrics. Broadcom said Tomahawk 6 was built for exactly those scale-up and scale-out environments. ### What comes next in Broadcom’s disclosures? (investors.broadcom.com) Broadcom said on March 4 that it expected about $22.0 billion in second-quarter fiscal 2026 revenue and about $10.7 billion in AI semiconductor revenue for that period. The company’s next formal update will come with its fiscal second-quarter results and any accompanying earnings call materials posted on its investor relations site. (investors.broadcom.com) (broadcom.com)