Via tech-insider.org
The semiconductor giant is quietly becoming the custom silicon backbone for the world's biggest AI players.
Broadcom has secured major multi-year agreements with three of the most important names in artificial intelligence: OpenAI, Google, and Meta. Together, these deals cover custom AI accelerator development measured in gigawatts.
The combined scope of these partnerships positions Broadcom as a critical alternative to Nvidia in the custom silicon market.
The deals at a glance #
The OpenAI partnership, announced on October 13, 2025, centers on the co-development of 10 GW of custom AI accelerators. Rack deployments are planned to begin in late 2026 and continue through the end of 2029.
Then there’s Google. Disclosed on April 6, 2026, Broadcom’s long-term deal with the search giant covers the development and supply of custom Tensor Processing Units (TPUs) and networking components. The agreement extends through 2031.
A notable wrinkle in the Google partnership: Anthropic, the AI safety startup behind Claude, will gain access to approximately 3.5 GW of TPU capacity beginning in 2027.
On April 14, 2026, Broadcom expanded its collaboration with Meta. The focus is Meta’s in-house chip program, the Meta Training and Inference Accelerator, or MTIA. The partnership initially exceeds 1 GW of custom silicon.
Why custom silicon is eating the market #
Custom silicon, sometimes called ASICs (application-specific integrated circuits), offers hyperscalers hardware tailored precisely to their workloads. Google pioneered this approach with its TPUs years ago. Now OpenAI and Meta are pursuing similar strategies, and Broadcom is the company they’re calling to help design and manufacture the chips.
These deals are designed to complement, not replace, existing Nvidia hardware. Most hyperscalers run mixed environments, using Nvidia GPUs for general training workloads and custom chips for specific inference or specialized tasks.
What this means for investors and the competitive landscape #
Multi-year agreements stretching to 2029 and 2031 give Broadcom something closer to a subscription-like revenue floor in its AI business.
Companies like Marvell Technology also compete in the custom ASIC space. Broadcom’s ability to land three of the most prominent AI developers as long-term partners raises the bar for everyone else trying to win similar contracts.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our