So, you thought the AI chip race was just for the cloud providers and semiconductor designers? Think again. Meta Platforms Inc. (META), the social media and metaverse behemoth, is quietly but determinedly building its own silicon brains. It's a move that speaks volumes about where the company sees its future—and how it plans to pay for it.
The Custom Silicon Playbook
Here's the thing about running the world's largest social networks: you have some very specific, very massive computing problems. Ranking your News Feed? That's a custom job. Serving you the perfect Reel? Another one. Meta has already deployed its own custom chips at scale for these ranking and recommendation tasks. Now, CFO Susan Li says the company wants to go further.
Speaking at a Morgan Stanley conference, Li revealed that Meta plans to expand its use of custom chips over time, with the ultimate goal of building processors capable of training its future AI models. It's a long-term bet. Meta isn't a cloud provider, but it operates some of the planet's largest data centers to train and run its AI. Owning the silicon that powers that could be a huge competitive—and financial—advantage.
The twist? This isn't happening in a vacuum. In recent weeks, Meta also signed those eye-popping supply deals with Nvidia (NVDA) and Advanced Micro Devices (AMD) for the chips that form the backbone of its current AI infrastructure. So it's a dual-track strategy: buy the best off-the-shelf hardware today while baking your own secret sauce for tomorrow. As Li put it, Meta evaluates different chips for different tasks, and custom silicon is a "key part" of its long-term plan for handling AI workloads.
Building the Whole AI Machine
Chips are just one piece of the puzzle. Meta is building out its entire AI ecosystem, and it's doing it on multiple fronts. On the product side, the company is testing a shopping research feature inside its Meta AI chatbot. Imagine asking for a recommendation for a new jacket and getting back images, prices, and links—a direct bridge from chat to commerce.
Behind the scenes, the organizational machinery is shifting. Meta is creating a new applied AI engineering organization, led by Reality Labs exec Maher Saba. This team will work on improving model training and development, operating alongside the company's ambitious Superintelligence Lab.
Then there's the fuel for all these AI engines: data. Meta is locking down licensing agreements with publishers to feed its models. The most notable is a deal with News Corp. (NWSA) that could be worth up to $50 million a year for access to news content and archives. It's a clear signal that quality, licensed data is a priority in the new AI landscape.












