MW How Meta's little-known chip business fits in with Zuckerberg's 'superintelligence' ambitions
By Christine Ji
Taking a page from Google's and Amazon's books, Meta is racing to vertically integrate its AI infrastructure via a new Broadcom deal
Meta plans to debut four new generations of its MTIA chips over the next two years to support ranking and general AI workloads.
After spending billions on data centers, Meta Platforms is focusing on another area of artificial-intelligence investment: custom chips.
Meta first introduced its custom silicon, called Meta Training and Inference Accelerator (MTIA) chips, in 2023. The chip division has been a relatively obscure part of the business and hasn't garnered much investor attention. The MTIA chips are also lesser known than Amazon.com's (AMZN) Trainium chips or Google's $(GOOG)$ $(GOOGL)$ tensor processing units.
But the Facebook parent company is betting big on its own silicon. Meta (META) and Broadcom $(AVGO)$ announced Tuesday an extension of their partnership to co-develop over a gigawatt of chips through 2029.
Custom silicon has become an increasingly attractive and cost-efficient undertaking among hyperscalers, providing an alternative to the highly in-demand and expensive Nvidia (NVDA) chips. However, Meta has been a relative newcomer to building in-house chips compared with cloud providers such as Google and Amazon.com.
"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 CEO Mark Zuckerberg said in the press release. "As we roll out more than [one gigawatt] of our custom silicon to start and then multiple gigawatts over time, this partnership will give us greater performance and efficiency for everything we're building."
Custom hardware is increasingly becoming a key part of Zuckerberg's roadmap to "personal superintelligence." The company announced last month that it plans to develop and deploy four new generations of MTIA chips in two years to support ranking and recommendations, as well as general AI tasks.
MTIA chips are designed for inference workloads, and Meta aims to release a new version every six months or less to iterate rapidly and adapt to evolving AI technologies. The partnership will also provide Meta with networking and other hardware solutions to efficiently scale up AI compute clusters.
See more: AMD's stock rockets as Meta deal serves as major validation point for investors
Google began renting out its tensor processing units on a large scale to Anthropic last year. Amazon's custom-chip business - which encapsulates its Trainium, Graviton and Nitro products - has also picked up momentum in recent quarters. Amazon CEO Andy Jassy shared last week that the business is growing at triple-digit rates, surpassing $20 billion in annual run rate. Meta has not offered financial details regarding its MTIA chip business.
Meta shared last month that it plans to use multiple chip types to meet its diverse workloads, and MTIA appears to be a significant area of focus. Meta is slated to spend roughly $135 billion on capital expenditures in fiscal 2026. The company has inked multibillion-dollar chip deals with Advanced Micro Devices $(AMD)$ and Nvidia, and has also used neoclouds such as CoreWeave (CRWV) and Nebius (NBIS) to rent cloud capacity.
For Broadcom, the Meta deal is the latest sign that the company maintains a critical position in the AI-infrastructure trade. Broadcom CEO Hock Tan will transition off Meta's board and move to a specialized advisory role, a shift that Bernstein analyst Stacy Rasgon said could be a positive read on Broadcom's MTIA momentum. The deal's 2029 time frame also provides more visibility into MTIA demand, Rasgon wrote in a Wednesday note.
The Meta deal comes after Broadcom secured a fresh agreement with Google and Anthropic last week to provide 3.5 gigawatts of tensor processing units over the next five years.
Read: Broadcom's stock is rising. Here's why its new Google and Anthropic deals are so significant.
-Christine Ji
This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.
(END) Dow Jones Newswires
April 15, 2026 10:12 ET (14:12 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments