Broadcom–Meta AI pact
Broadcom expanded a multi‑year partnership with Meta to develop what it calls the industry’s first 2nm AI accelerator and to deploy multi‑gigawatt AI infrastructure, starting with an initial commitment of over 1GW through 2029. (stocktitan.net) Hock Tan is reported to be moving into an advisor role on Meta’s board as part of the broader tie‑up. (x.com)
Broadcom and Meta have widened their artificial intelligence chip alliance through 2029, tying Meta’s custom silicon plans to a multi-gigawatt buildout. (broadcom.com) The companies said on April 14, 2026 that they will work over the next three years on new generations of Meta’s Meta Training and Inference Accelerator chips, starting with an initial commitment of more than 1 gigawatt by 2029. Broadcom said the effort includes what it calls the industry’s first 2-nanometer artificial intelligence compute accelerator. (broadcom.com) Meta Training and Inference Accelerator is Meta’s in-house chip family for running and training artificial intelligence systems inside its own data centers instead of relying only on general-purpose processors from outside suppliers. Meta said in March it was developing and deploying four new generations of these chips within two years, after first introducing the family in 2023. (about.fb.com) A 2-nanometer chip refers to a newer manufacturing generation that packs more circuitry into a small area, which companies pursue to raise performance and lower power use. Broadcom said its role is to provide the custom accelerator platform and Ethernet networking needed to link Meta’s expanding artificial intelligence clusters without bottlenecks. (broadcom.com) The scale in this deal matches the scale Meta is already building on the ground. Meta said in February that it had broken ground on a 1 gigawatt data center campus in Lebanon, Indiana, a project it valued at more than $10 billion. (about.fb.com) Meta has been shifting toward custom chips for years as recommendation systems, advertising models, and generative artificial intelligence features demand more compute. In April 2024, Meta said its second-generation Meta Training and Inference Accelerator had more than doubled compute and memory bandwidth from the prior version and was already serving models in production. (about.fb.com) Broadcom’s announcement also ties the hardware push to Meta’s consumer products. The company said the roadmap is meant to support real-time generative artificial intelligence features and what Meta calls “personal superintelligence” across WhatsApp, Instagram, and Threads. (broadcom.com) One widely shared report said Hock Tan would move into an advisory role as part of the broader arrangement, but Meta’s latest official filing I could verify shows Tan was elected to Meta’s board on February 13, 2024 and joined its Audit and Risk Oversight Committee at that time. I did not find a new Meta filing or company statement confirming a board change as of April 15, 2026. (sec.gov) The immediate next step is not a single chip launch but a longer build cycle. Broadcom described this week’s deal as the first phase of a sustained rollout, with Meta locking its custom silicon roadmap more tightly to the data centers it is now building at gigawatt scale. (broadcom.com)