Meta Is Building a Chip Empire to Chase Superintelligence

Meta Is Building a Chip Empire to Chase Superintelligence

Meta is not just building AI, it is building the silicon to run it. The company is working with Broadcom to roll out four generations of its own MTIA chips in two years, an unusually fast pace, all part of Mark Zuckerberg's pledge to deliver personal superintelligence to billions of people. The bet is backed by more than 60 billion dollars a year in infrastructure spending.

The chips are custom-built for Meta's needs. MTIA, short for Meta Training and Inference Accelerator, is designed specifically for the ranking, recommendation and generative-AI work that powers Meta's apps. The latest version is already in production, with three more on the way to handle heavier AI workloads into 2027. Building its own silicon lets Meta tune performance and cost to its exact use, instead of buying general chips from Nvidia.

The pace is the striking part. Four chip generations in two years is far faster than the usual multi-year cycle, a sign of how urgently Meta wants to control its own AI hardware. It has also lined up Qualcomm to supply server processors and partnered with AMD, spreading its bets across suppliers. Meta is assembling a whole stack rather than depending on any single vendor.

The goal Zuckerberg keeps naming is superintelligence. Meta has stood up a dedicated lab and hardware leadership focused on building toward artificial general intelligence, and Zuckerberg frames the enormous spending as the price of being first. Owning the chips, the data centers and the models is his theory of how Meta wins the AI race, by controlling every layer of it. Ambition at the largest scale.

The honest caveats are large. Spending tens of billions on custom silicon and AGI is a huge gamble, returns are far from proven, and chip projects often slip or underdeliver. Superintelligence itself is a contested, undefined goal that may be years away or never arrive as imagined. This is conviction spending on a vision, not a guaranteed payoff, and the market is right to question the bill.

So Meta is pouring money into becoming a chipmaker as much as an AI lab, betting that owning the hardware is the path to the most advanced AI. Four chip generations, 60 billion a year, and a superintelligence goal driving it all. Whether it is visionary or excessive is the open question. Watch whether the MTIA chips ship on schedule and what they actually power.