Meta launches new self-developed AI chip to reduce dependence on external semiconductor suppliers

Meta is deploying a new in-house chip to help power its artificial intelligence services, aiming to reduce its reliance on chips from outside companies such as Nvidia.

The chip, announced Wednesday, is the latest version of Meta’s training and inference accelerator (MTIA), which helps rank and recommend content on Facebook and Instagram. Meta released the first MTIA product last year.

Meta's shift to AI services has brought greater demand for computing power. Last year, the social media giant released its own version of an AI model to compete with OpenAI's ChatGPT. It has also added new generative AI features to its social apps, including custom stickers and chatbot characters with celebrity faces.

In October, the company said it would invest up to $35 billion in infrastructure to support AI, including data centers and hardware. “AI will be our largest investment area through 2024,” CEO Mark Zuckerberg told investors that month.

Much of that spending is likely to continue to go to Nvidia, which makes the popular H100 graphics card used to power AI models. Earlier this year, Zuckerberg said the company would buy 350,000 of those chips, which cost tens of thousands of dollars each.

But the movement for tech giants to develop chips in-house is growing. Meta is joining Amazon.com Inc. AWS, Microsoft Corp. and Google parent Alphabet Inc. in trying to move away from this costly reliance. However, it’s not an overnight solution. So far, these efforts have not had a noticeable impact on the industry’s huge demand for Nvidia’s AI accelerators.

The AI ​​boom has helped Nvidia become the world's third-largest technology company, behind Microsoft and Apple. Its sales to data center operators totaled $47.5 billion in fiscal 2024, up from $15 billion the year before. Analysts predict that figure will further double in fiscal 2025.