Meta and Broadcom deepen chip partnership as Broadcom CEO exits Meta board

The deal

Meta and Broadcom announced an expanded partnership Tuesday to co-develop multiple generations of Meta’s in‑house MTIA AI accelerators through 2029. Meta has committed to an initial deployment of 1 gigawatt of Training and Inference Accelerators, and the deal is expected to scale to multiple gigawatts in future generations, the companies said. It has been reported that Broadcom CEO Hock Tan told Meta last week he will not stand for reelection to the company’s board, according to a filing — a move that follows his 2024 appointment and comes amid the new agreement.

Why it matters

Mark Zuckerberg framed the tie‑up as building “the massive computing foundation we need to deliver personal superintelligence to billions of people,” saying Broadcom will help across chip design, packaging and networking. Hock Tan has pushed back against recent doubts about Meta’s MTIA roadmap, saying the program is “alive and well” and that Broadcom will help the next‑generation XPUs scale to multiple gigawatts in 2027 and beyond. Market reaction was muted: Broadcom shares rose about 3% in extended trading after the announcement, while Meta’s stock was flat.

Context and stakes

Hyperscalers are hustling for GPU alternatives as Nvidia and AMD chips remain expensive and supply‑constrained. The result: a race to application‑specific ASICs — smaller, cheaper and task‑focused than general‑purpose GPUs. Google and Amazon blazed earlier trails with TPUs and custom chips; Meta’s MTIA chips, unveiled and expanded this year, are for internal use only. The Broadcom pact follows the chipmaker’s recent deals with Google and Anthropic, and comes as Meta pushes a big AI spending plan — up to $135 billion committed this year — and builds out dozens of data centers.

Change at the board. Gigawatts of silicon. A bellwether moment in the AI hardware arms race. Who gains the upper hand? That answer will come down to chips, cooling, cabling — and whether these bespoke accelerators can outmuscle the GPU behemoths when it matters.

Sources: cnbc.com