Meta Shelves Plan to Lease AI Compute Power to Competitor Anthropic
Meta has reportedly decided not to move forward with a plan to lease AI computing infrastructure to Anthropic. The decision comes after Anthropic CEO Dario Amodei directly approached Meta earlier this year to negotiate access to Meta's AI hardware resources. The move highlights severe compute hardware constraints across the AI industry and underscores growing strategic rivalries among tech giants. As Meta accelerates its own AI efforts with models like Muse, hoarding internal compute capacity takes precedence over generating revenue from direct competitors. Anthropic was previously rumored to be seeking a two-year, $10 billion lease agreement for Meta's computing resources. However, Meta's internal demand for compute surged following the launch of its native multimodal Muse model family, leading executives to reject the deal to avoid strengthening a rival.
## BACKGROUND
AI compute infrastructure relies on vast clusters of specialized hardware to train and run large neural network models based on architectures like the Transformer. As companies build increasingly capable multimodal foundation models and autonomous AI agents, securing sufficient compute power has become a critical bottleneck affecting both product velocity and talent retention.