Meta Says Its Own Chip Now Rivals the Best on the Market

Meta's homegrown silicon just hit a milestone. The company unveiled the MTIA 400, its first custom AI chip that it says delivers cost savings and raw performance competitive with the leading commercial products, meaning Nvidia's. It is a sign Meta's years of chip work are starting to pay off as it pours over 100 billion dollars into AI this year.

The claim is the important part. Custom chips built by big tech firms have often lagged Nvidia's on raw power, useful for saving money but not for the heaviest work. Meta saying the MTIA 400 is competitive on performance, not just cheaper, would mark a real step up, letting it run more of its AI on its own hardware instead of buying everything from Nvidia. Catching up on performance is the hard part.

It is part of a family. Meta laid out a lineup of chips, with the MTIA 450 adding faster memory and the MTIA 500 adding more memory at higher speeds, some already in use and others rolling out through 2026 and 2027. Building a whole range rather than a single chip shows Meta is committing to its own silicon for the long run, tuning different versions to different jobs. This is a program, not a one-off.

The stakes are enormous. Meta plans to spend between 115 and 135 billion dollars on AI-related capital this year, and a large chunk of that goes to chips, so even modest savings per chip translate into billions and less dependence on a single supplier. Every workload Meta can move onto its own silicon is money kept in-house. Owning the chip changes the math.

The honest caveat is that a company's own benchmarks deserve scrutiny. Competitive on performance is Meta's own claim, real-world results and independent testing matter more than a launch statement, and Nvidia keeps moving with new platforms of its own. Custom chips also tend to handle specific tasks rather than the whole range Nvidia covers. This narrows the gap, it does not close it.

So Meta says it has finally built a chip that can go toe to toe with the best, a meaningful moment in Big Tech's push to escape the Nvidia tax. An MTIA 400 competitive on performance, a full chip family behind it, and over 100 billion in AI spending around it. The buyers keep becoming builders. Watch independent benchmarks and how much of Meta's AI actually runs on its own chips.