Meta Platforms Inc. has announced a new deal with Broadcom Inc., extending a partnership that already sees the two companies work together on the design of the social media giant’s in-house artificial intelligence accelerators.

The company said in an announcement that it is committing to an initial deployment of one gigawatt of its Meta Training and Inference Accelerators, a custom-designed chip for AI workloads that Meta runs in its own data centers. Ultimately though, the extended partnership envisions Meta deploying multiple gigawatts of the chips, which are based on Broadcom’s technology.

Broadcom stressed in a separate announcement that the new MTIA chips will be the first custom silicon for the AI industry that uses a 2-nanometer process. The chipmaker’s stock rose more than 3% in late trading today after the news was published. It’s now up more than 10% in the year to date, outpacing the broader S&P 500 Index, which has gained just 2% over the same timeframe.

Meta co-founder and Chief Executive Mark Zuckerberg was quoted as saying that the MTIA chips will use Broadcom’s chip design, packaging and networking technologies to “build out the massive computing foundation we need to deliver personal superintelligence to billions of people.”

Reports earlier this year had suggested that Meta was struggling to bring the latest generation of its MTIA chips to market, but Broadcom CEO Hock Tan said in an earnings call last month that that’s not the case. “Contrary to recent analyst reports, Meta’s custom accelerator, MTIA roadmap is alive and well,” he said. “We’re shipping now and, in fact, for the next generation of XPUs, we will scale to multiple gigawatts in 2027 and beyond.”

Last month, Meta announced it was developing four new versions of the MTIA chips. It debuted the original MTIA chip back in 2023, following in the footsteps of rivals such as Google LLC and Amazon Web Services Inc., which have also developed their own AI processors.

The MTIA chips give Meta an alternative to the costly and hard-to-buy graphics processing units made by companies such as Nvidia Corp. and Advanced Micro Devices Inc. as it scrambles to build new AI data centers. Like Google’s and Amazon’s chips, they’re a kind of application-specific integrated circuit, or ASIC, which are smaller and cheaper to manufacture than GPUs, because their design limits them to a narrow range of computing tasks. In contrast, GPUs are general-purpose processors that can run almost any kind of task.

Google delivered its first custom ASICs way before the start of the AI boom, back in 2015. Its tensor processing units were originally designed for standard cloud computing workloads, and were followed by Amazon’s first custom chips in 2018. Both companies relied on Broadcom to help develop their silicon.

Broadcom has announced a string of deals for its XPUs, or custom processors, in recent months. The most recent came just eight days earlier, when Anthropic PBC said it had penciled an agreement with the chipmaker and Google to secure 3.5 gigawatts of TPUs starting next year.

As part of today’s announcement, Meta revealed that Tan, who has sat on its board of directors since 2024, has opted not to stand for reelection. Instead, he will be moving to an advisory role at the company, offering input on the company’s future custom chip strategy only. Meta’s board is also facing a second departure, with Reuters reporting the news that Tracey Travis, the former Chief Financial Officer of Estée Lauder, will give up the seat she has held since 2020.

The social media giant has announced a string of multibillion-dollar chip deals this year as part of its commitment to spend more than $135 billion on capital expenditures in fiscal 2026. Previously, it committed to deploying six gigawatts of AMD’s GPUs, millions of chips from Nvidia, and also new custom processors designed by the chip architecture designer Arm Holdings Ltd. It has further committed to spending billions of dollars on renting chips from suppliers such as CoreWeave Inc. and Nebius N.V.

Image: Meta Platforms