MENLO PARK, Calif. — Inside Meta Platforms, the two numbers traveled together this month: capital expenditure for 2026 between $125 billion and $145 billion, the largest such commitment in the company’s history, and a plan to cut roughly 8,000 jobs from a workforce that, on the eve of the announcement, exceeded 75,000.
The numbers are linked, executives close to the company say, by a single accounting line. The capital is directed almost entirely at Nvidia graphics processing units, datacenter buildout, and the infrastructure deals — power, water, land — that the buildout requires. The job cuts are concentrated in layers of the organization where headcount substitutes for, rather than orchestrates, the new infrastructure. The implicit accounting is simple: a Llama-pipeline engineer or an infrastructure planner is, in Meta’s 2026 budgeting, a different kind of dollar than an organizational layer that the new infrastructure renders unnecessary.
The decision is the clearest articulation any large public technology company has yet offered of how it plans to translate the productivity claims surrounding artificial intelligence into the kind of operating leverage public-market investors will pay for. The story Meta is telling, in effect, is that the model is doing the work, the infrastructure is being bought, and the human cost of that bet is being absorbed up front rather than amortized across the next several quarters.
Whether the bet pays off is a question that will not be answered for at least eighteen months. The Llama line — Meta’s strategic counterweight to the closed-frontier laboratories — is now being funded at the closed-frontier capex envelope, a number that did not previously seem available to an open-weights program. The Ray-Ban Display, the $799 wearable that Meta has positioned as the consumer-facing application of its artificial-intelligence research, has yet to demonstrate the kind of unit economics that the headline capex would require for the company’s broader story to hold together at this run rate.
For those laid off this month, the question is more immediate. Most have been given separation packages that are, by the standards of the company’s past restructurings, generous. The internal reception, by accounts of several people inside the company who agreed to discuss it on the condition that they not be named, has nevertheless been bleak. The cuts are the second large-scale round in three years, and the message they communicate — that a particular layer of work no longer competes well, on the company’s own internal accounting, with a graphics-processing unit and a model checkpoint — is not one that lends itself to a comfortable severance meeting.
The next several quarters will tell whether Meta has built the right thing at the right time, or whether the company has bought the wrong decade’s infrastructure with the wrong decade’s confidence.