Broadcom Secures Meta Deal to Supply Custom Silicon Through 2029 as CEO Hock Tan Leaves

Broadcom’s new long‑term supply contract with Meta cements the chipmaker’s role as a primary hardware partner for the social‑media giant’s AI ambitions. According to Tom’s Hardware, the agreement will see Broadcom deliver “multiple generations of custom‑designed Meta Training and Inference Accelerator (MTIA) hardware through 2029,” a rollout that will initially exceed 1 GW of compute capacity and eventually consume “multiple gigawatts of power.” The scale of the deal implies that Meta will rely on hundreds of thousands of Broadcom‑built AI processors, a volume that dwarfs most single‑customer semiconductor contracts and signals a deepening of Meta’s in‑house silicon strategy.

The deal is built around Broadcom’s XPU platform, a modular architecture that blends custom silicon blocks with off‑the‑shelf logic, memory and high‑speed I/O. Tom’s Hardware notes that this approach “greatly improve[s] efficiency and lower[s] the cost of such bespoke processors,” positioning Broadcom to provide both the specialized tensor engines needed for large‑scale model training and the standard components that keep the overall system affordable. In addition to the accelerators, Broadcom will supply Meta with Ethernet networking solutions designed for “scale‑up, scale‑out, and scale‑across requirements,” and will likely integrate optical interconnects that are already part of Microsoft, Meta and OpenAI multi‑source agreements for unified compute fabrics.

A notable technical detail is the use of modified RISC‑V cores from Andes Technology within the MTIA accelerators for scheduling and orchestration. While the core compute blocks—tensor engines, vector engines and systolic arrays—are not RISC‑V‑based, the inclusion of RISC‑V cores “will likely benefit both Andes Technology and the RISC‑V ISA ecosystem in general,” according to the Tom’s Hardware report. This design choice reflects a broader industry trend toward open‑instruction‑set architectures for flexibility, and it could give Meta a licensing advantage as it scales its custom silicon across future generations.

The agreement also prompted a governance shift: Broadcom CEO Hock Tan is stepping down from Meta’s board of directors and moving into an advisory capacity. Tom’s Hardware frames the move as a conflict‑of‑interest mitigation, noting that Tan will “guide Meta’s custom silicon roadmap and influence its future infrastructure investments” while remaining Broadcom’s chief executive. This separation allows Meta to retain strategic input from Broadcom’s leadership without the appearance of direct corporate control, a balance that may become a template for other large‑scale supplier‑customer relationships in the AI hardware space.

From a market perspective, the partnership underscores Meta’s commitment to “personal superintelligence” at scale, as CEO Mark Zuckerberg told Tom’s Hardware: the rollout of “more than 1 GW of our custom silicon to start and then multiple gigawatts over time” is intended to deliver “greater performance and efficiency for everything we are building.” By locking in a supplier that can provide both custom accelerators and the networking fabric needed for massive compute clusters, Meta reduces its exposure to supply‑chain volatility and positions itself to compete more aggressively with rivals such as OpenAI, Google and Microsoft, all of which are also pursuing vertically integrated silicon strategies.

Overall, the Broadcom‑Meta pact illustrates how leading semiconductor firms are transitioning from pure component suppliers to collaborative partners that co‑design and co‑manage the full AI stack. The longevity of the contract—extending to 2029—offers Broadcom a stable revenue stream while giving Meta a predictable hardware roadmap, a combination that could accelerate the rollout of next‑generation AI services across the company’s billions of users.