Meta this month raised its 2026 capital-expenditure guidance to between $125 billion and $145 billion, the largest single-year capex plan in the company’s history, and disclosed that it would cut roughly 8,000 jobs as part of a restructuring that aligns the org around that spend. The two numbers belong in the same paragraph: the layoffs are concentrated in the layers of the company the new spend will not save.
The capex itself is heavily directed at Nvidia GPUs, datacenter buildout, and the infrastructure deals that surround them. The math of the buildout has been visible in Meta’s quarterly disclosures for two years; the upward revision pushes the 2026 commitment well past what the most aggressive sell-side projections were tracking last quarter.
For the field’s compute story, three things matter about the raise. First, it sets a hard floor for the hyperscaler capex cycle through the end of the year — Meta has now committed to spend on GPU supply at a level that will pull on the Nvidia order book regardless of how the next two earnings prints land. Second, the open-weights pitch — Meta’s Llama strategy as a strategic counterweight to closed frontier labs — has not gotten cheaper; the company is now spending the closed-frontier capex envelope to keep its open program credible. Third, the layoffs ratify a particular theory of the operating leverage: the model is doing the work, the infrastructure is being bought, and the headcount the company is shedding is not in either of those columns.
The combined message is not subtle. Meta has decided that 2026 is the year the AI thesis either delivers operating leverage or does not, and the company is willing to take both the capital cost and the human cost of that bet up front.
The capex range will be revisited at the next earnings print, where investors will look for whether the spend is delivering Llama-pipeline progress or Ray-Ban Display-style consumer milestones — and which of those the market actually rewards.