Meta internally projected that its annual spending on Anthropic's AI models could reach as much as $10 billion, according to a New York Times report aggregated on August 27, 20261. The projection was made even as Mark Zuckerberg publicly criticized Anthropic.
The figure, described as an internal estimate of future spending, reflects the scale of Meta's reliance on third-party large models for advancing its AI research and commercial applications2. Anthropic's Claude series models have established technical strength in areas including long-text processing, complex logical reasoning, and enterprise-level application ecosystems.
ANALYSIS The projected spend would represent one of the largest known commercial relationships between two frontier AI companies, and the gap between Zuckerberg's public posture toward Anthropic and Meta's internal spending projections adds a layer of strategic tension to the relationship.
The report lands during a turbulent period for Anthropic. A federal judge on August 28, 2026, struck down the Pentagon's designation of Anthropic as a national security supply-chain risk, ruling the action "illegal and baseless" ctx. Separately, a threat intelligence report published on August 27, 2026, documented a ransomware operator using Anthropic's Claude Sonnet 4.5 model, running inside SpaceX's Cursor AI coding agent, to conduct exploitation across ten enterprise networks ctx.
ANALYSIS The $10 billion projection, if realized, would position Anthropic as a major infrastructure vendor to one of the world's largest technology companies, even one that operates its own family of open-weight Llama models. That dynamic underscores how even companies with substantial in-house model development capacity see value in procuring external frontier models at scale.