Meta Nears $10 Billion Deal to Lease AI Compute to Anthropic

Meta is in early talks to lease some of its AI computing power to Anthropic, in a deal that may be worth up to $10 billion over two years. The agreement could help Anthropic get more GPUs, which it says it needs for its Claude models, while giving Meta extra income. The deal appears to be Meta's possible first move toward acting like a cloud provider, though talks are still in the early stages. Some experts suggest that the deal could affect competition in the market and raise questions among regulators, but nothing has been finalized yet.

Meta is in early talks to potentially lease computing power to Anthropic in a deal worth up to $10 billion over two years, but no agreement has been signed and the deal may not happen. The NYT reports Meta is in talks for a potential agreement that would provide Anthropic with GPU capacity and create a revenue stream for Meta, but the deal is not confirmed.

This potential agreement signals Meta's strategic pivot towards operating as a cloud provider, positioning it to compete with established hyperscalers like AWS and Azure in the lucrative market for AI infrastructure services.

Why Anthropic Needs Meta's AI Compute

Anthropic requires massive-scale GPU capacity to train its advanced Claude AI models and meet surging user demand. Despite existing partnerships, the company faces a significant compute deficit, which this deal with Meta aims to alleviate by providing immediate access to a large fleet of AI-ready servers.

Despite multi-billion-dollar deals with Amazon, Google, and others, Anthropic executives state the company has only a fraction of the computing power required for its Claude models (CNBC). NVIDIA will supply Meta with millions of Blackwell and Rubin GPUs for Meta's internal AI training and execution, not lease them to Meta.

Meta's Pivot to a "Compute" Business

To support this new direction, Meta is reportedly establishing an internal "Meta Compute" division to lease raw GPU time and hosted AI models to external clients (Bloomberg). This move helps justify Meta's soaring 2026 capital expenditure forecast of $115 - $135 billion, allowing it to monetize surplus capacity and offset massive infrastructure investments.

Key Deal Terms Under Discussion

Term: A multi-year lease, with initial drafts pointing to a two-year period.
Value: Up to $10 billion over two years, though terms remain unconfirmed.
Services: Access to GPU clusters for AI model training and inference, potentially including managed services from Meta's proprietary stack.
Timing: Negotiations are in preliminary stages and depend on Meta establishing its commercial billing and service infrastructure.

Competitive Ripple Effects

Meta's entry into the compute rental market would challenge specialized providers like CoreWeave, which already serves both Meta and Anthropic. While this could intensify competition, analysts predict that severe, industry-wide GPU scarcity will likely persist due to the 12-24 month lead times required for new data center construction.

This arrangement would create a dynamic of "coopetition," where rival model developers - Meta with Llama and Anthropic with Claude - collaborate on infrastructure. This highlights a key industry trend: access to scarce AI chip supply has become a more critical bottleneck than protecting intellectual property, forcing even direct competitors to partner for compute resources.

Strategic Outlook and Regulatory Scrutiny

Investors and regulators will closely watch this development, scrutinizing its potential impact on Meta's profit margins and whether it raises antitrust concerns. For Anthropic, a successful deal promises tangible benefits, including increased capacity for its users and an accelerated timeline for launching more powerful Claude models. As talks are still fluid, neither company has announced a projected closing date.