Broadcom's fiscal third quarter confirms that the largest AI spenders are building dedicated chip pipelines at a scale that reshapes the semiconductor supply chain. AI semiconductor revenue rose 221% to $16.7 billion4,5, the company named Anthropic and OpenAI as expanding customers alongside incumbents Google and Meta, and HPE separately signed a $3.5 billion server deal with a cloud firm while raising its outlook for two fiscal years3. ANALYSIS Taken together, the results show hyperscalers locking in custom ASIC capacity and the infrastructure to run it, a diversification that runs parallel to, not merely supplementary to, the GPU-centric buildout.
Why it matters
Broadcom's total quarterly revenue hit $29.59 billion, up 86% year over year, beating the $29.36 billion consensus. Adjusted earnings came in at $3.32 per share versus the $3.24 expected. Net income more than tripled to $13.09 billion from $4.14 billion a year earlier. ◆ Those numbers place Broadcom's AI semiconductor line alone at a run rate that would have ranked it among the largest chip businesses in the world just two years ago. The growth is not a rounding error on a diversified portfolio; AI chips now dominate the company's semiconductor segment.
The big picture
The customer list tells the strategic story. Broadcom designs custom chips for Google, Meta, and OpenAI. It expects to speed up shipments of Google Ironwood tensor processing units to Anthropic and TPU 8i chips to Google. It expects production shipments of Meta's custom MTIA accelerator optimized for inference and recommendation at scale. And OpenAI is getting ready to tape out its second-generation chip while putting together plans with Broadcom for a third version. Broadcom said it developed the custom Jalapeno chip with OpenAI.
ANALYSIS Each of those programs represents a multi-year, multi-gigawatt commitment that ties a hyperscaler's roadmap to Broadcom's design and packaging capacity rather than to a merchant GPU vendor's product cycle. The breadth is notable: Google, Meta, OpenAI, and Anthropic account for a large share of global AI compute demand, and all four now have named custom-silicon programs running through Broadcom.
Broadcom is looking at a 1.3 gigawatt deployment of Jalapeno in 2027, with a line of sight of over 5 gigawatts for Jalapeno and the second generation. For Anthropic, the company is looking to deploy 5 gigawatts of TPU 8i chips in 2027, with line of sight to deliver another 10 gigawatts. Broadcom will deliver tens of billions of dollars of processors to Google each year for the next several years. ◆ The gigawatt figures translate custom silicon from a niche optimization into a primary infrastructure layer, one where power allocation, not chip availability, becomes the binding constraint.
Meanwhile, HPE reported Q3 revenue up 34% year over year to $12.2 billion, versus the $11.9 billion estimate. The company boosted its sales outlook for this fiscal year and next, citing artificial intelligence workloads. It signed a $3.5 billion server deal with a cloud firm. ◆ HPE's results reinforce the demand signal from the infrastructure side: the servers, networking, and storage that surround custom ASICs are scaling in tandem.
Between the lines
Wall Street's reaction was cooler than the operating results. Broadcom's stock dropped 5% in extended trading after the company forecast Q4 revenue of $34.8 billion, below the $35.03 billion analysts expected. HPE and NetApp shares also fell in late trading despite both beating expectations and raising full-year outlooks2. ◆ The sell-the-news pattern across all three companies suggests investors are pricing in the possibility that near-term growth rates have peaked even as absolute spending continues to climb. Broadcom's longer-range ambitions remain aggressive: the company is targeting AI revenue of $115 billion in fiscal 2027 and $230 billion in fiscal 2028. Infrastructure software contributed $8.75 billion in the quarter, and the company said it is on target to produce over $30 in earnings per share.
Apple also said it would spend more with Broadcom for U.S. chip production, adding a non-AI revenue anchor.
What's next
Broadcom's fiscal fourth quarter, with its $34.8 billion revenue guide, will test whether the gap between guidance and consensus was a one-quarter air pocket or a sign that custom-silicon ramps follow lumpier delivery schedules than GPU shipments. OpenAI's second-generation chip tapeout and the planning for a third version set a cadence: ANALYSIS if those programs proceed on roughly annual cycles, Broadcom's design-win pipeline extends well into the end of the decade, embedding it as a structural counterweight in the AI chip supply chain.