Broadcom targets $100B AI chip revenue

Broadcom’s role in the AI buildout is no longer confined to Ethernet switches and back-end networking. Company filings and customer announcements across the past year show the Palo Alto, California-based chip designer sitting inside several of the largest custom-compute programs in the industry, including work tied to Meta, OpenAI, Google’s TPU ecosystem and Anthropic’s latest capacity expansion. The disclosures also give investors a clearer view of the economics: Broadcom said on March 4 that first-quarter AI semiconductor revenue reached $8.4 billion, up 106% from a year earlier, and it forecast $10.7 billion for the second quarter. IBTimes reported on May 16 that Broadcom is targeting $100 billion in AI chip revenue in 2027 and carrying roughly $73 billion of backlog. That backlog figure and the 2027 target were not detailed in the Broadcom materials reviewed for this article, but the company and its customers have publicly described multi-year, gigawatt-scale programs that help explain why investors are focused on custom silicon as Broadcom’s fastest-growing business. (investors.broadcom.com) ### Which AI companies have actually named Broadcom? Meta and OpenAI have both publicly named Broadcom in custom-chip announcements. Broadcom and Meta said on April 14 that they had expanded their partnership to support Meta Training and Inference Accelerator chips, with plans extending through 2029 and an initial commitment of more than 1 gigawatt in the first phase. OpenAI and Broadcom said on October 13, 2025 that they would collaborate on 10 gigawatts of OpenAI-designed AI accelerators. (ibtimes.com) The companies said Broadcom would co-develop accelerator and Ethernet systems, with targeted deployment beginning in the second half of 2026 and completion by the end of 2029. Anthropic also named Broadcom in a separate compute agreement. Anthropic said on April 6 that it had signed a new agreement with Google and Broadcom for multiple gigawatts of next-generation TPU capacity expected to begin coming online in 2027. (broadcom.com) ### Where does Google fit into this picture? Google has not, in the materials reviewed here, issued a stand-alone announcement saying Broadcom designs all of its TPU chips. But Google has continued to roll out new TPU systems, including TPU 8t and TPU 8i in April, while Anthropic’s April 6 announcement explicitly tied Google and Broadcom together in a next-generation TPU capacity agreement. (openai.com) That means the cleanest verified formulation is narrower than some broad market claims: Google is a named participant in an Anthropic compute expansion that also names Broadcom, and Google continues to expand its TPU roadmap. (anthropic.com) A broader claim about Broadcom’s role across all Google AI chips would require additional direct sourcing. ### What has Broadcom itself said about the size of the AI business? Broadcom gave the clearest official numbers in its fiscal first-quarter release on March 4. (blog.google) The company said total quarterly revenue was $19.311 billion, AI semiconductor revenue was $8.4 billion, and second-quarter AI semiconductor revenue was expected to reach $10.7 billion. Hock Tan, Broadcom’s chief executive, said in that release that demand was being driven by custom AI accelerators and AI networking. (anthropic.com) Broadcom has also used product releases to describe its XPU platform, optical components and Ethernet products as building blocks for gigawatt-scale AI clusters. ### Why are custom chips such a big part of this story? Meta, OpenAI and Anthropic have each described programs that pair custom accelerators with large-scale networking and systems work. (investors.broadcom.com) Meta said its MTIA roadmap is intended to optimize performance and total cost of ownership at scale, while OpenAI said its Broadcom partnership is aimed at scalable, energy-efficient AI infrastructure. Anthropic said its Google-Broadcom agreement would support frontier Claude models and extraordinary customer demand. Those statements matter because they show Broadcom selling more than a single chip. The company is being named in longer-dated infrastructure programs that include accelerator design, packaging, interconnect and cluster networking. That is an inference drawn from the structure of the disclosed deals and Broadcom’s own product descriptions. ### What should readers watch next? June 3 is the next scheduled date on Broadcom’s calendar. (broadcom.com) The company said it will report fiscal second-quarter 2026 results that day, and investors will be looking for updated AI semiconductor revenue, customer concentration details if any are disclosed, and any change to Broadcom’s demand outlook. The next visible milestones after that sit with named customers. Anthropic said its new Google-Broadcom TPU capacity is expected to start coming online in 2027, while OpenAI and Broadcom said their 10-gigawatt deployment is targeted to start in the second half of 2026 and finish by the end of 2029. (broadcom.com) (anthropic.com) (investors.broadcom.com)