Meta (NASDAQ:META) just made a bold move in the AI racetesting its first in-house training chip in a bid to cut reliance on Nvidia (NASDAQ:NVDA) and bring down sky-high AI infrastructure costs. The company has been burning cash on AI, projecting up to $65 billion in capital expenditures for 2025. Now, it’s shifting gears, aiming to power its recommendation engines and generative AI with custom-built silicon. The chip is currently in small-scale deployment, and if all goes well, Meta could ramp up production fast. Partnering with Taiwan Semiconductor Manufacturing (NYSE:TSM), the tech giant is looking to optimize efficiency and take greater control of its AI stack.

• Warning! GuruFocus has detected 2 Warning Sign with META.

Meta has played this game beforeits previous custom inference chip didn’t make the cut, forcing a multi-billion-dollar shopping spree on Nvidia GPUs. But things are different this time. The company’s MTIA chip for inference has already proven successful, running recommendation systems across Facebook and Instagram. Now, it’s tackling training, the real AI powerhouse. Meta’s Chief Product Officer, Chris Cox, calls this a walk, crawl, run processsuggesting they’re still in the early stages, but moving in the right direction. If this new chip delivers, it could shake up the AI hardware landscape, challenging Nvidia’s stranglehold on the market.

The stakes are high, and so is the skepticism. The AI industry is at a crossroadsscaling up large models with more GPUs isn’t the only game in town anymore. Meta’s custom chip strategy is a bet on efficiency, but will it pay off? Investors are watching closely. If Meta pulls this off, it could signal a seismic shift in AI computing. If not, it might just end up doubling down on Nvidia again.

This article first appeared on GuruFocus.