Meta Starts Testing Its First AI Chip to Reduce Dependence on Nvidia | WhatsApp Meta Download | Meta Facebook login | Meta Instagram | Turtles AI

Meta, Facebook’s parent company, has begun testing its first in-house chip dedicated to training AI systems marking a significant step in the company’s strategy toward designing custom hardware and reducing dependence on outside vendors such as Nvidia. 

Key points:

• Meta tests a proprietary chip for AI training.
• Collaboration with TSMC to manufacture the chip.
• Goal: Reduce cost and dependence on outside suppliers.
• Plan for large-scale use by 2026.

The social media giant has begun limited distribution of the chip and plans to ramp up production for wider use if testing is successful. The move is part of a long-term plan to reduce infrastructure costs as the company invests heavily in AI tools to drive growth. Meta has forecast total spending of $114 billion to $119 billion through 2025, with up to $65 billion earmarked for capital expenditures, mostly for AI infrastructure.

Meta’s new chip is a dedicated accelerator designed specifically to handle AI tasks, making it more power efficient than the integrated graphics processing units (GPUs) typically used for these workloads. Meta is partnering with Taiwanese semiconductor maker TSMC for production.

Testing began after the chip’s first “tape-out” was completed, a crucial step in silicon development that involves sending an initial design to a chip fab. A tape-out typically costs tens of millions of dollars and takes about three to six months to complete, with no guarantee of success. If it fails, it would need to diagnose the problem and repeat the process.

This chip is the latest development in the Meta Training and Inference Accelerator (MTIA) series. Despite a shaky start, with a chip at a similar stage of development being abandoned, Meta began using an MTIA chip for inference in the recommendation systems that determine what content appears in Facebook and Instagram news feeds last year.

Meta executives have said they plan to start using their chips for training by 2026, a process that involves feeding an AI system large amounts of data to “teach” it how to perform certain functions. The goal is to start with recommendation systems and then apply them to generative AI products, such as the Meta AI chatbot. Meta Chief Product Officer Chris Cox described the chip development efforts as a gradual progression, noting the success of the first-generation inference chip for recommendation systems.

Meta previously halted development of a custom inference chip after a failed small-scale test, opting instead to purchase Nvidia GPUs for billions of dollars in 2022. The company has since remained a major customer of Nvidia, amassing a wide array of GPUs to train its models, including recommendation systems, advertising systems, and the Llama model series. These units also run inference for the more than 3 billion users who use the company’s apps every day.

However, in 2025, the value of these GPUs has come into question, as AI researchers have expressed growing doubts about how much progress can be made by continuing to “scale” language models by adding more and more data and processing power. These concerns were reinforced by the launch in late January of new low-cost models by Chinese startup DeepSeek, which optimize computational efficiency by relying more heavily on inference than most existing models. This led to a global decline in AI stocks, with Nvidia shares losing as much as a fifth of their value at one point, before recovering most of the ground, although they have since fallen again due to broader business concerns.

Meta’s move to develop in-house AI chips reflects a broader trend among big tech companies to develop custom silicon solutions to meet their specific needs. For example, Google has developed its Tensor Processing Unit (TPU) to accelerate AI workloads, while Apple has introduced the Neural Engine into its chips to improve machine learning capabilities in consumer devices.

In addition, generative AI, which involves the ability of AI systems to create content such as text, images and music, has become a significant focus in recent years. Models like GPT.

Source: Reuters