Meta’s Next-Generation MTIA AI Chips Set to Enter Production in September
Meta is on track to begin manufacturing the newest version of its in-house AI chip, the MTIA (Meta Training and Inference Accelerator), starting in September, according to TechCrunch.
The chip and its partners
Meta designs the MTIA chip in-house but doesn’t build it alone: the design work is done in partnership with Broadcom, while actual fabrication will be handled by TSMC, the world’s largest contract chipmaker. This mirrors the pattern several other AI labs are following — designing the chip yourself, but leaning on Broadcom and TSMC for the parts of the process that require decades of specialized manufacturing expertise.
Why Meta wants its own chips
Meta runs enormous AI workloads across its own products — ad ranking, content recommendation, and its Llama family of models among them — and buying every accelerator from Nvidia at market price is expensive at that scale. A custom inference chip, tuned specifically for Meta’s own workloads, can be cheaper to run per unit of compute than general-purpose GPUs, even after accounting for the up-front cost of designing it.
Why it matters
Meta joins Google (TPU), Amazon (Trainium/Inferentia), OpenAI (Jalapeño, with Broadcom), and reportedly Anthropic and DeepSeek in the industry-wide move toward custom AI silicon. For a company the size of Meta, even modest efficiency gains per chip translate into massive savings at the scale of its global data center footprint.
Source: TechCrunch.
