Background

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.

Considerations

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.

Resolution Criteria

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:

Public cloud service offering:

• Publicly accessible platform where customers can purchase/rent compute

• Website or portal allowing sign-ups for compute access

• Published pricing for compute hours/tokens

Enterprise/Partnership deals:

• Announced deals to provide compute to named companies

• SEC filings or earnings reports mentioning compute service revenue

• Customer testimonials or case studies on Meta channels

Official confirmation via:

• Meta website, blog, or press releases

• Mark Zuckerberg on Facebook/X explicitly confirming external compute sales

• Earnings call statements from Meta executives

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

Important clarifications:

• ✅ DOES count: Any external customer (not Meta/Facebook/Instagram/WhatsApp) purchasing compute; compute for training, fine-tuning, or inference; beta programs with paying customers

• ❌ Does NOT count: Free research partnerships; internal use by Meta products; only licensing models without compute infrastructure; traditional software/API licensing without compute component