Meta and Broadcom have agreed to expand their partnership on custom AI chips through 2029, covering multiple generations of Meta’s Training and Inference Accelerator, known as MTIA — the silicon that powers AI across Meta’s apps and services.
A commitment to deploy more than 1 gigawatt of computing capacity anchors the agreement, with Meta framing that figure as an opening installment in what it expects to become a multi-gigawatt buildout. Under the arrangement, Broadcom’s XPU platform will be used to develop custom AI accelerators, with collaboration spanning the design, packaging, and networking layers of the hardware stack. Broadcom’s Ethernet technology will also connect Meta’s expanding AI compute clusters.
As part of the expanded deal, Broadcom CEO Hock Tan will leave Meta’s board of directors and move to an advisory role focused on Meta’s custom silicon roadmap. Tan joined Meta’s board in 2024, according to CNBC. According to a regulatory filing, Tan informed the company last week of his intention to step down from the board rather than seek another term.
“Meta is partnering with Broadcom across chip design, packaging, and networking to build out the massive computing foundation we need to deliver personal superintelligence to billions of people,” Meta founder and CEO Mark Zuckerberg said in a statement. Hock Tan, Broadcom’s president and CEO, called the initial MTIA deployment “just the beginning of a sustained, multi-generation roadmap.”
The MTIA chips will be the first AI silicon to use a 2 nanometer process, according to Reuters. Meta unveiled four new versions of MTIA in March, with Of the four chips unveiled in March, the MTIA 300 is already running Meta’s ranking and recommendation systems, and the remaining three are slated to arrive by 2027.
Broadcom stock rose about 3% in extended trading after the announcement. Meta stock was little changed.
The Broadcom deal is one piece of a broader hardware push. Meta has struck a multiyear agreement with AMD to deploy up to 6 gigawatts of AMD Instinct GPUsand has also committed to millions of Nvidia chips and custom chips made by chip architecture firm Arm Holdings. The company announced in January that it expects to spend up to $135 billion on AI infrastructure in 2026, and it is developing 31 data centers to support that buildout, 27 of which are located domestically.
In a separate disclosure, Meta announced that Tracey Travis, a board member since 2020, will step down and not seek another term when shareholders meet at the company’s annual meeting.