Meta and Broadcom Extend Custom AI Chip Partnership to 2029 | $2.3B Paid, 2nm MTIA 500 Confirmed

Meta Platforms and Broadcom announced on April 14, 2026 that they have extended their custom silicon partnership through at least 2029, cementing what is now the largest known custom AI chip collaboration in the industry outside of Apple. The deal spans four generations of Meta's MTIA (Meta Training and Inference Accelerator) chip family, with the flagship MTIA 500 set to be fabricated on a 2-nanometer process — the most advanced node ever used for a custom AI accelerator.

$2.3 Billion Paid in 2025 | Scale of the Financial Commitment

Meta disclosed in securities filings that it paid Broadcom $2.3 billion in 2025 alone for chip design services and hardware. The new agreement backs Meta's 2026 capital expenditure budget, which the company has projected at between $115 billion and $135 billion — the largest single-year infrastructure spend in Meta's history. The initial deployment commitment under the extended deal exceeds one gigawatt of computing capacity, roughly equivalent to the power consumption of 750,000 U.S. homes.

The financial scale reflects a deliberate shift in Meta's strategy: rather than buying Nvidia GPUs at market rate, Meta is building proprietary accelerators tuned specifically for its ranking, recommendation, and generative AI workloads, reducing long-term unit cost and dependency on third-party supply chains. See our earlier coverage of TSMC and ASML Q1 2026 earnings driven by AI chip demand for broader context on the custom silicon wave.

Four MTIA Generations | Technical Roadmap Through 2029

The MTIA chip family runs on open-source RISC-V architecture. Meta defines the logic and performance requirements; Broadcom handles silicon design, advanced packaging, and networking integration. The four-generation roadmap confirmed under the extended deal:

The MTIA 500 represents the most technically ambitious milestone. The jump from the current MTIA 300 to the 500 delivers a claimed 25x increase in compute performance and a 4.5x increase in HBM (High Bandwidth Memory) bandwidth — figures that, if achieved, would make the MTIA 500 competitive with or superior to flagship Nvidia datacenter GPUs for Meta's specific inference workloads.

Broadcom XPU Platform | CoWoS Packaging, Proprietary Networking, 2nm TSMC

Broadcom is providing Meta with its XPU custom accelerator platform — not just a chip design, but a full-stack solution including:

• CoWoS packaging (Chip on Wafer on Substrate) — the same advanced stacking technology used in high-end Nvidia H100 and H200 GPUs, enabling tighter memory-logic integration.
• Proprietary networking silicon — Broadcom's datacenter networking IP is embedded directly into the AI cluster architecture, reducing inter-chip latency.
• 2nm fabrication at TSMC — the MTIA 500 will be the first custom AI accelerator manufactured at the 2nm node, giving Meta a process advantage over most commercially available AI hardware.

The combination positions Broadcom as not merely a chip designer for Meta, but as the systems integrator for Meta's entire AI hardware stack — a role that creates significant switching costs and deepens the strategic dependency in both directions.

Hock Tan Steps Off Meta Board | Moves to Dedicated Advisory Role

To manage the growing conflict-of-interest risk as the financial relationship scales, Broadcom CEO Hock Tan will step down from the Meta Board of Directors when his current term expires at the 2026 annual shareholder meeting. He will transition into a dedicated advisory role focused specifically on executing the silicon roadmap through 2029 — maintaining his operational involvement without the governance complications of simultaneous board membership.

The move is consistent with how other deep corporate partnerships have handled executive cross-representation as deal sizes grew past a point where independence could credibly be claimed. With $2.3 billion paid in 2025 and multi-year commitments now locked in, the formal separation of Tan's board seat from his advisory function is a straightforward governance response to an extraordinary commercial relationship.

For Meta, the extension locks in supply chain certainty for its AI infrastructure buildout at a moment when hyperscaler demand for advanced packaging and 2nm capacity is outpacing global fab supply. For Broadcom, it secures one of the largest and most technically visible custom silicon engagements in the industry through the end of the decade.