Meta on Wednesday debuted a new model of its AI chip, which it said is three times more efficient than its first one.
Along with Microsoft, Facebook parent Meta is one of the two largest buyers of Nvidia’s AI chips, known as GPUs, graphics processing units. But Meta launched its own AI chip last May, the MTIA v1 (Meta Training and Inference Accelerator), to support its generative AI efforts. So did did Nvidia’s other customers. Microsoft announced its first AI accelerator in November. And Meta’s next generation MTIA comes just a day after Google and Intel showed off their own new AI chips. The three big AI chip arrivals could signal a threat to Nvidia’s dominance of the AI accelerator market, of which it holds a 75% share.
Meta said producing its own hardware allows it to make chips more compatible with its software. “Because we control the whole stack, we can achieve greater efficiency compared to commercially available GPUs,” the company said in its announcement.
But the MTIA isn’t actually being used to power AI models yet — it’s used to support ranking and recommendation ads models. The company said its AI chips will eventually help it power and innovate its AI models. Google’s TPU v5p AI chip, on the other hand, is already being used to train and serve large language models and its own Gemini 1.5 Pro chatbot.
“MTIA will be an important piece of our long-term roadmap to build and scale the most powerful and efficient infrastructure possible for Meta’s unique AI workloads,” Meta said.
More AI chips news
Google’s new chips look to challenge Nvidia, Microsoft and Amazon
The AI frenzy powered TSMC to its fastest monthly sales growth since 2022
Intel says its new AI hardware is ‘way ahead of Nvidia’
Nvidia is an AI superpower. It started at a Denny’s