Meta and Broadcom on Tuesday announced a comprehensive agreement that extends the existing partnership between the two companies on the design of Meta’s custom in-house AI accelerator through 2029.
At the same time, Mr. Mehta said in the filing that Broadcom CEO Hock Tan told Mr. Mehta last week that he had decided not to stand for re-election to Mr. Mehta’s board of directors. Mr. Tan joined Meta’s board of directors in 2024.
Meta has committed to initial deployment of 1 gigawatt of training and inference accelerators, according to a statement. The deal will see Meta eventually deploy multi-gigawatt chips based on Broadcom technology.
Broadcom said in a statement that the MTIA chip will be the first AI silicon to use a 2-nanometer process.
“Meta is partnering with Broadcom in chip design, packaging, and networking to build the large-scale computing infrastructure needed to bring personal superintelligence to billions of people,” Meta co-founder and CEO Mark Zuckerberg said in a statement.
Broadcom shares rose 3% in extended trading after the announcement. Meta stocks were flat.
“Contrary to recent analyst reports, the roadmap for Meta’s custom accelerator, MTIA, is still alive and well. It’s currently shipping and, in fact, is on track to scale to multi-gigawatts in 2027 and beyond for next-generation XPUs,” Tan said on Broadcom’s March earnings call.
Meta announced four new versions of its MTIA chip in March. The company first announced custom silicon in 2023, following similar chip programs from Google and Google. Amazon.
Hyperscalers are seeking alternatives to expensive and constrained graphics processing units from Nvidia and AMD to power AI data centers.
They manufacture GPU replacements called application-specific integrated circuits (ASICs). It is smaller and cheaper than the flagship GPU for general-purpose AI, but is limited to performing a narrower range of tasks.
Google first got into the custom ASIC game, releasing its first Tensor Processing Unit in 2015. Amazon then announced its first custom chip in 2018. While these tech giants include their AI chips as part of their respective cloud computing platforms and make them accessible to customers, Meta’s MTIA chip is used entirely for internal purposes.
The deal comes two weeks after Broadcom announced a long-term deal with Google for TPU production, giving Anthropic access to 3.5 gigawatts worth of Google’s in-house chips.
Broadcom stock is up 10% so far in 2026, while the S&P 500 is up about 2% over the same period.
Tracy Travis, who stepped down as Estée Lauder’s finance chief last year, will leave the company’s board after assuming the board seat in 2020, Meta said.
Meta has made a flurry of deals since pledging to spend up to $135 billion on AI in January, as it seeks to keep pace with its mega-cap peers, Anthropic and OpenAI.
Tan said during an earnings call in March that he is looking for OpenAI to deploy a first-generation AI chip that will be operational in 2027.
Meta’s AI deals over the past few months include promises to deploy up to 6 gigawatts of AMD GPUs, millions of Nvidia chips, and new custom chips from chip architecture company Arm Holdings.
Overall, Meta plans 31 data centers, including 27 in the United States.
