Broadcom (AVGO) recently disclosed plans for significant AI compute deployments by major tech companies, indicating a growing demand for its custom AI accelerators (XPUs) in the coming years. According to Broadcom, demand for its XPUs is expected to accelerate between 2027 and 2028, driven by infrastructure expansions from Anthropic, OpenAI, and Meta.
Broadcom CEO Hock Tan outlined Anthropic’s deployment plans during the company’s latest earnings call. Anthropic will deploy 1GW of Ironwood compute capacity in 2026, scaling to 10GW by 2028, making it Broadcom’s largest XPU customer in both years. OpenAI, meanwhile, has begun shipping its first generation Jalapeño chip, targeting over 5GW of deployments by 2028. This makes OpenAI a strong contender to become Broadcom’s second largest XPU customer in the same period.
Meta’s custom chip program is also progressing. Broadcom will deliver three generations of MTIA accelerators to Meta through 2027, with cumulative deployments expected to reach 3GW by 2028. The two companies previously announced a multi year partnership to jointly develop multiple generations of MTIA chips and supporting AI networking infrastructure.
These strategic partnerships underscore Broadcom’s solid positioning in the high growth AI compute market. The company expects significant expansion in the demand for its XPUs, driven by the increasing compute requirements of major AI infrastructure players.