Meta plans capital expenditures of $66-72 billion in 2025 and expects similarly large increases in 2026 to bring additional AI compute capacity online. Zuckerberg revealed plans to build several multi-gigawatt datacenter clusters, with the first (Prometheus) coming online in 2026, and a second (Hyperion) scaling up to 5GW over several years. Meta is targeting deployment of more than one million GPUs to support next-generation AI models.

However, Meta signed a $14.2 billion contract with CoreWeave to supply compute services for AI through 2031, and Meta also signed a six-year agreement with Google Cloud valued at more than $10 billion for AI model training and deployment. This indicates Meta is currently a buyer of external compute rather than a seller.

Meta CFO Susan Li stated the company is exploring ways to work with financial partners to co-develop data centers, noting there could be models that attract external financing while providing Meta flexibility if infrastructure requirements change. This suggests Meta may eventually monetize excess capacity, but no concrete plans for external compute sales have been announced. Analysts note that "Meta wants to be an AI company" and "doesn't really want to figure out how to be a hyperscaler for GPUs," indicating infrastructure-as-a-service may not align with Meta's core business strategy.

Resolves YES if by 11:59 PM PT on December 31, 2026, Meta offers AI compute services (training or inference) to external customers through ANY of the following:

Resolves NO if no external compute offerings exist by deadline, or if Meta only uses infrastructure internally for its own AI/products.