Meta Plans Up to 20% Layoffs as AI Infrastructure Costs Balloon

Meta is planning layoffs of up to 20% of its workforce as the company's AI infrastructure spending climbs sharply, Reuters reported. Meta has projected $60–65 billion in capital expenditure for 2025 — a dramatic increase from prior years — and is restructuring costs to absorb the outlay. The cuts follow a run of compounding AI setbacks that have left the company without a clear flagship model heading into the back half of the year.

The Llama 4 model family drew public criticism over benchmark manipulation, with allegations that early published results were misleading. The largest planned variant in the family, codenamed "Behemoth," was cancelled ahead of its expected summer release. A subsequent internal model called "Avocado," developed by Meta's superintelligence team as a follow-up, has also reportedly underperformed expectations.

Much of the open-source agent tooling ecosystem — from LangChain integrations to locally-hosted agent frameworks — runs on Llama-class models. A sustained capability gap at Meta would further concentrate frontier model development among closed providers like OpenAI, Google DeepMind, and Anthropic, narrowing the options available to <a href="/news/2026-03-15-open-weights-vs-open-training-painful-reality-post-training-1t-parameter-model">developers building on open-weights infrastructure</a>. Meta has been the most consequential open-weights provider by volume of deployment; there is no obvious substitute if it falls behind.

Hacker News commenters responding to the Reuters story noted a telling cultural shift: Meta's 2022 layoff of 11,000 employees was among the most-discussed stories in the site's history, while the current round has attracted comparatively little alarm — a reflection of how normalized large-scale tech layoffs have become in the AI investment era. Internal morale has been described as "brutal." With model failures stacking up and a major capex pivot underway, Meta needs its superintelligence team to produce something competitive before the spending rationale gets harder to defend.