Anthropic and OpenAI to Overtake Google as Broadcom's Top AI ASIC Clients
Broadcom announced that AI labs Anthropic and OpenAI are projected to surpass Google as its two largest custom AI ASIC customers by 2027–2028. The semiconductor giant reported $16.7 billion in AI chip revenue for FY2026 Q3, driven by shipments of custom hardware including Google's TPU v7 Ironwood, OpenAI's Jalapeño, and Meta's MTIA accelerators. This shift underscores a major structural transformation in the AI hardware landscape as leading AI research labs move toward custom application-specific silicon (ASICs) to lower compute costs and scale massive model workloads. Driven by this aggressive expansion, Broadcom projects its annual AI semiconductor revenue to reach $115 billion in 2027 and $230 billion in 2028. Anthropic plans to expand its TPU deployment scale to 1GW of Ironwood in 2026, 5GW of TPU v8i in 2027, and an additional 10GW in 2028. Meanwhile, OpenAI plans to deploy 1.3GW of its Jalapeño inference chip in 2027 and over 5GW of Jalapeño and its successor in 2028, alongside Meta deploying 3GW of MTIA accelerators by late 2028.
## BACKGROUND
Application-Specific Integrated Circuits (ASICs), often termed XPUs by Broadcom, are chips tailored specifically for distinct workloads like AI training and inference rather than general-purpose computing. Major technology and AI companies collaborate with physical design partners like Broadcom to create custom silicon—such as Google's Tensor Processing Units (TPUs), Meta's MTIA, and OpenAI's Jalapeño—to achieve superior energy efficiency and performance compared to traditional GPUs.