Meta's Silicon Bet: Custom Chips Aim for 44% Cost Reduction
Meta is betting big on custom silicon to reduce costs and energy use in its AI workloads, targeting a 44% reduction in total cost of ownership and 40% improvement in power efficiency compared to Nvidia's GPUs.
The company has already started production with the MTIA 300 chip generation, which powers inference tasks for Instagram feeds and Facebook ads. The next generation, codenamed Iris (MTIA 400), is set to enter production in September 2026 after clearing initial testing phases.
Meta aims to release new chip generations every six months, a pace the semiconductor industry rarely attempts. This is made possible by a modular chiplet design, which allows Meta to mix and match building blocks rather than redesigning entire chips from scratch each time. Broadcom collaborated on this program, with TSMC handling fabrication.
Meta's projected capital expenditures for 2026 are between $115 billion and $135 billion, with plans to scale its computing capacity to 14 gigawatts by 2027. The MTIA line is designed specifically for inference tasks, sacrificing flexibility in exchange for efficiency. Training AI models still relies on Nvidia and AMD GPUs.
While Meta continues to maintain partnerships with both Nvidia and AMD worth billions of dollars, the MTIA chips are positioned as a complement to external chips, not a replacement.