Shares of Meta Platforms surged more than 7% after Reuters revealed an internal memo showing the company will begin manufacturing its custom-designed AI chip in September, part of an infrastructure push that dwarfs the budgets of most nations. The question for shareholders: does pouring record capital into homegrown silicon accelerate Meta's AI ambitions, or does it merely accelerate its cash burn?
• The Chip Passed Its First Test — But the Spending Bill Is Staggering
Testing the chip took only six weeks and found no major issues — a relatively quick progress signal for an in-house effort that has floundered for more than half a decade. Yet Meta expects to spend as much as $145 billion on AI infrastructure this year, a significant portion of Big Tech's more than $700 billion projected outlay.
Morgan Stanley already models Meta's 2026 free cash flow as flat to negative. Investors are cheering a technical milestone while the balance sheet absorbs a historically large spending commitment.
• Building Your Own Chips Helps, but Doesn't Replace Nvidia
The chip is intended to supplement — not replace — the large volumes of GPUs Meta buys from Nvidia and AMD.
Meta has also struck a multiyear agreement with AMD to deploy up to six gigawatts of AMD GPUs , and its Broadcom partnership extends through 2029, with more than one gigawatt as an opening installment in a planned multi-gigawatt buildout. Custom silicon may cut per-unit costs over time, but near-term dependence on outside suppliers remains heavy.
• Doubling Capacity to 14 Gigawatts Raises an Uncomfortable Math Problem
Meta plans to deploy seven gigawatts of computing in 2026, having added one gigawatt in the first half and forecasting another 5.5 by year-end.
It plans to double capacity again next year to reach 14 gigawatts in 2027.
For context, 14 gigawatts exceeds the total electricity consumption of many small countries.
Conventional per-gigawatt cost estimates price the expansion near $350 billion, roughly two and a half times Meta's entire full-year capital spending guide.
• The Bigger Gamble Is Becoming an Infrastructure Company
Meta also plans to sell excess AI computing capacity to outside developers and businesses. That would transform a social-media advertising company into a cloud-computing landlord — a business model shift with entirely different margin profiles and competitors. A new chip roughly every six months through 2027 is a significantly more aggressive schedule than the annual or slower cycles common across the industry. Execution risk is high, but so is the potential payoff if Meta can lower the unit cost of powering AI across its 3.5 billion users.