News
Meta to accelerate custom AI chip roadmap with biennial releases utilizing TSMC 3nm and 4nm nodes
Thursday, March 5, 2026 at 09:53 PM
Meta plans to accelerate its custom AI silicon roadmap by launching a new MTIA chip every six months over the next two years. The MTIA 300 is currently in production, while the MTIA 400 is in the testing phase. These RISC-V based chips are being co-developed with Broadcom and will utilize TSMC's 3nm and 4nm manufacturing nodes. The supply chain involves ASE for advanced packaging and KYEC for testing services. Future iterations including the MTIA 400 and 450 are scheduled for release in 2027.
Context
On March 11, 2026, Meta announced an aggressive acceleration of its custom silicon roadmap, committing to a biennial release cycle that will see new AI chips launched every six months over the next two years. Developed in partnership with Broadcom, the MTIA (Meta Training and Inference Accelerator) series utilizes RISC-V architecture and is fabricated on TSMC’s 3nm and 4nm nodes. This rapid iteration strategy aims to keep Meta’s hardware aligned with the explosive growth of generative AI and recommendation workloads while reducing reliance on expensive third-party vendors like Nvidia.
The roadmap includes the MTIA 300, which is currently in production for ranking tasks, and the MTIA 400, which has completed testing for imminent deployment. Two additional generations, the MTIA 450 and MTIA 500, are scheduled for release in 2027 with specific optimizations for high-bandwidth memory and low-precision AI processing. To support this massive hardware rollout, Meta is leveraging Taiwan’s semiconductor ecosystem, including KYEC for chip testing and ASE for advanced packaging, as it expands its global data center footprint in locations like Louisiana, Ohio, and Indiana.
Sources (10)
Four MTIA Chips in Two Years: Scaling AI Experiences for BillionsMeta rolls out in-house AI chips weeks after massive Nvidia ... - CNBCMeta unveils plans for batch of in-house AI chips | ReutersMeta Is Developing 4 New Chips to Power Its AI and Recommendation Systems | WIREDExpanding Meta's Custom Silicon to Power Our AI WorkloadsMeta Outlines New MTIA Accelerator Roadmap for its Next- ...AI Semiconductor Market Growth Insights | PDF | Graphics Processing Unit | Artificial IntelligenceFabless Companies and Business Models-defining and detecting their strength - THE WAVES
Related Companies
Meta
META