Meta challenges Nvidia — the company’s AI chip is already tested

Meta is developing and testing an AI training chip as part of its plan to reduce its dependence on suppliers such as Nvidia. The company plans to increase production for widespread use if the tests are successful.

Reuters writes about it.

Everything we know about the development of Meta’s AI chip

Meta was motivated to develop its own chip by the idea of reducing huge infrastructure costs, as the company is placing high stakes on AI tools to drive growth. In 2025, Meta plans to spend up to USD 65 billion on AI infrastructure.

The company’s training chip is designed to perform only AI-specific tasks, which will make it more energy efficient than integrated GPUs that are commonly used for AI.

The chip is being developed jointly with Taiwanese chipmaker TSMC. Typical testing costs tens of millions of dollars and takes three to six months with no guarantee that the test will be successful.

Meta’s management has stated that it plans to use its own chips for training by 2026. Their goal is to start with recommendation systems and later use them for generative AI, such as the Meta AI chatbot.

Meta has previously abandoned the development of its own chip after it failed during a test deployment similar to the one it is currently conducting.

As a reminder, Meta announced the launch of the Waterworth project. It involves the construction of the 50,000 km long submarine optical internet cable.

We also wrote that the first autonomous AI agent, Manus, was unveiled in China. It can make decisions independently and perform complex multi-stage tasks.