Broadcom targets $100B AI chip market

Broadcom is trying to define the custom AI-chip market in numbers. On March 4, 2026, CEO Hock Tan said the company had “line of sight” to more than $100 billion in AI-chip revenue in 2027 as customers expanded orders for custom accelerators and networking gear. (cnbc.com) Broadcom has paired that target with a reported $73 billion AI backlog and a series of named partnerships with Meta, Google, Anthropic and OpenAI. (fool.com) The company’s pitch is not that it will replace every general-purpose AI chip. Broadcom has instead been building custom silicon and the networking that connects it, letting large cloud and model companies design systems around their own workloads. (investors.broadcom.com) Hock Tan described that business in Broadcom’s earnings commentary as being driven by “custom AI accelerators and Ethernet AI switches.” ### Where does the $100 billion figure come from? March 4, 2026, is the date Broadcom put the target into the market. CNBC, citing Broadcom’s first-quarter earnings report and comments from Tan, said the company had line of sight to more than $100 billion in revenue from AI chips. (cnbc.com) Broadcom’s investor site separately shows that first-quarter 2026 results were released that day. December 2025 is when Broadcom laid out the order book behind that claim. In its fourth-quarter and full-year 2025 earnings release, Tan said Broadcom expected first-quarter AI semiconductor revenue of $8.2 billion, up more than 100% from a year earlier. (investors.broadcom.com) A transcript of the same earnings call said the company expected its $73 billion AI backlog to be delivered over the next 18 months. ### Which customers has Broadcom actually named? Meta and Broadcom announced an expanded partnership on April 14, 2026. Broadcom said it would deliver technology supporting Meta’s MTIA chips and Ethernet-based rack-scale interconnects, with plans extending through 2029. (cnbc.com) The initial commitment exceeded 1 gigawatt, according to the release. Google and Broadcom widened their work on April 6, 2026. CNBC reported that Broadcom agreed to produce future versions of Google’s AI chips, while Anthropic would get access to about 3.5 gigawatts of computing capacity based on Google processors under an expanded arrangement. (investors.broadcom.com) Anthropic confirmed the same month that it was deepening work with Google Cloud and Broadcom while continuing to train and run Claude across AWS Trainium, Google TPUs and Nvidia GPUs. OpenAI and Broadcom disclosed a separate partnership on October 13, 2025. OpenAI said the companies would collaborate on 10 gigawatts of custom AI accelerators, with Broadcom set to deploy racks of accelerator and network systems starting in the second half of 2026 and completing by the end of 2029. (investors.broadcom.com) ### What exactly is Broadcom selling into AI systems? Broadcom said at OFC 2026 that its AI infrastructure portfolio spans XPU, Ethernet, optics, SerDes, DSP and PCIe products for gigawatt-scale clusters. On March 12, 2026, the company also said it was shipping its 102.4-terabits-per-second Tomahawk 6 switch in production volume. (cnbc.com) Those products sit alongside custom accelerators that customers design for their own training or inference needs. Google’s TPU roadmap shows why that matters. Google said in April 2025 that Ironwood was its seventh-generation TPU and introduced eighth-generation TPU products in April 2026 for training and inference workloads. (openai.com) Broadcom has not publicly claimed ownership of Google’s TPU architecture, but its April 2026 agreement to produce future versions of Google AI chips places it inside that supply chain. That is an inference from Google’s TPU announcements and Broadcom’s disclosed manufacturing role. ### Is this only about chips, or also about networking? Ethernet is central to Broadcom’s AI business. (investors.broadcom.com) Tan said in December that first-quarter AI growth would be driven by “custom AI accelerators and Ethernet AI switches,” and Broadcom’s Meta release tied the partnership to rack-scale interconnects for MTIA deployments. OpenAI’s October announcement also said Broadcom would co-develop systems that include accelerators and Ethernet solutions for scale-up and scale-out. Anthropic’s own statements point the same way. In April 2026, the company said it was using a range of hardware platforms so it could match workloads to the chips best suited to them, while keeping resilience across providers. (blog.google) That leaves room for Broadcom to sell both custom silicon support and the networking fabric around it, depending on the customer. June 3, 2026, is the next scheduled checkpoint. Broadcom’s investor calendar says the company will report fiscal second-quarter 2026 results that day, giving investors the next update on AI revenue, backlog and customer programs already named by Meta, Google, Anthropic and OpenAI. (investors.broadcom.com 1) (investors.broadcom.com 2) (anthropic.com)