Broadcom has secured multi-year agreements with OpenAI, Google, and Meta to develop custom AI accelerators, positioning itself as a key alternative to Nvidia in the custom silicon market. The OpenAI partnership, announced on October 13, 2025, covers co-development of 10 GW of custom AI accelerators, with rack deployments planned from late 2026 through the end of 2029. A deal with Google, disclosed on April 6, 2026, includes custom Tensor Processing Units (TPUs) and networking components through 2031, with Anthropic gaining access to approximately 3.5 GW of TPU capacity beginning in 2027. On April 14, 2026, Broadcom expanded its collaboration with Meta on the Meta Training and Inference Accelerator (MTIA), with the partnership initially exceeding 1 GW of custom silicon. These custom chips, or ASICs, complement rather than replace Nvidia hardware, as hyperscalers run mixed environments using Nvidia GPUs for general training and custom chips for specialized tasks. The multi-year agreements, stretching to 2029 and 2031, provide Broadcom a subscription-like revenue floor in its AI business. Competitors like Marvell Technology also operate in the custom ASIC space, but Broadcom’s long-term partnerships with three major AI developers raise the competitive bar.