The frontier AI race is shifting from chasing a single model’s capability to winning on computing scale, business models, and capital access, with Meta aiming to master its own models, operate competitors’ models, and sell computing power externally to secure a leading position.

Meta is aggressively expanding talent and data infrastructure, reassigning thousands of engineers to an internal reinforcement learning environment factory and investing billions into Scale AI to attract top researchers from OpenAI and Anthropic.

Meta plans to deploy and operate Anthropic’s Claude model directly within Meta data centers as a Private Instance, bypassing API access and cloud bottlenecks to offer AI computing as a service.

Anthropic’s Claude Code is strengthening its moat, with Claude’s strong GitHub impact and over $60 billion in ARR, signaling high profitability and robust net expansion that bolster its competitive stance against OpenAI.

Meta’s strategy could extend to integrating OpenAI models into the same private-instance framework in the future, placing Meta among AI infrastructure incumbents like Microsoft AI Foundry, AWS Bedrock, and Google Vertex AI.

Meta’s broader AI push includes massive hardware and software investments, targeting up to $145 billion in AI infra capex with 7 GW of computing power in 2026 and 14 GW in 2027, plus five Titan data-center clusters and a mass-produced Iris AI chip with Broadcom and TSMC.

SemiAnalysis estimates Meta could generate more than $10 billion annually by leasing substantial computing power, viewing this as building a hyperscaler business rather than a short-term resource sale.

The drive comes amid ongoing shortages of AI computing capacity and token quotas; Meta aims to lift traffic restrictions and ensure constant access to top models for internal and premium services.

The plan includes negotiating a Private Instance of Claude to provide isolated, high-performance computing with unrestricted model access within Meta’s own infrastructure.