
Meta's double-play
Meta is trying to do two things at once: build the chips that power its AI ambitions and sell more AI features to users. BofA stuck with a Buy rating and an $810 price target, arguing that Meta’s plan could make the company leaner, richer, and a lot less dependent on third-party silicon.
The chip math gets spicy
The big headline here is the custom silicon story. BofA says Meta wants to roll out its next-gen MTIA chips, with Arke expected in the first half of 2027 and Astrid coming later that year. The bank estimates that if Meta builds 5-6 gigawatts of owned AI capacity in 2027, and if custom chips save it around 40% versus buying off-the-shelf hardware, the company could pocket about $8.5 billion in savings.
That’s not pocket change. That’s “buy a small island and still have money left over” money.
It’s not just cost cuts
Meta also has a new revenue story brewing with Meta One, its subscription bundle for AI tools and creator features. The company says it already has 15 million subscribers and trials across existing offerings, which BofA sees as an early proof point that users may actually pay for the AI goodies instead of just tapping the free version forever.
BofA’s logic is simple:
- more paid users
- more recurring revenue
- less pressure to make every dollar come from ads
Why investors should care
Meta’s AI spend has been one of the market’s favorite anxiety buttons. If custom chips really lower the cost of inference at scale, that gives Meta more control over margins just as it pours money into the AI race. Add in a subscription business that could keep growing, and suddenly the story looks less like “expensive science project” and more like “big tech with a plan.”
Big picture: Meta is betting that owning the plumbing and selling the service is the best way to turn AI from a cost center into a profit machine.