By December 31, 2028, which category will best describe the realized scale of Meta Platforms' external AI-compute monetization business (i.e., AI compute/data-center capacity that Meta has sold or leased to third parties outside Meta, whether via raw GPU rental, hosted model API access, or similar)?

rationale:

As of late 2026, Meta’s massive AI infrastructure buildout—targeting tens of gigawatts this decade with a narrowed 2026 capex guidance of $130–145B 2 sources—creates a fundamental tension between internal demand and external monetization. While Zuckerberg announced the "Meta Compute" initiative in January 2026 to potentially sell excess capacity, the Q2 2026 posture was notably cautious. Management emphasized that compute remains a "scarce strategic asset," heavily prioritized for Meta's internal superintelligence labs and core ad-ranking, and that it would be "foolish" to sell it all for short-term profit 2 sources. Nonetheless, external monetization is "definitely on the table," with companies making unsolicited premium offers almost weekly 2 sources.

The defining dynamic for this forecast is the sheer scale of current AI-compute pricing, which makes the revenue thresholds significantly easier to cross than the capacity (GW) thresholds. The resolution criteria explicitly dictate using the higher implied category if revenue and capacity conflict. At current premium market rates of roughly $10B to $50B per GW-year—echoed by precedents like the Anthropic–SpaceX and Nscale deals —leasing just 200 to 500 megawatts could generate between $3B and $10B in annualized revenue 2 sources. Consequently, even a modest external allocation could easily push the outcome into the Moderate or Major revenue tiers, bypassing the need for multi-gigawatt external commitments.

Evaluating estimated timelines for near-term revenue generation centers the probability mass heavily around the middle categories, Pilot (33%) and Moderate (35%), acknowledging that high current AI-compute pricing makes moderate revenue tiers achievable even without multi-gigawatt capacity leases. A single wholesale lease, such as the preliminary Anthropic proposal to rent up to $10B of compute over two years 2 sources, would secure a $5B annual run-rate and immediately trigger a Moderate resolution. However, Meta is currently still a net buyer of compute—relying on massive leases from CoreWeave and Crusoe 2 sources—and internal AI workloads are voracious. Furthermore, Zuckerberg has clearly stated a preference for the higher margins of "selling intelligence rather than selling compute directly" 2 sources. This suggests Meta is most likely to execute deliberate, capped allocations of external capacity rather than wholesale dumping, anchoring the most probable outcomes in the Pilot to Moderate range.

The extremes of the distribution face structural headwinds. A Major outcome (19%) is constrained by a strict conjunctive hurdle: even if revenue exceeds $10B, Meta must function as a "recognized infrastructure/cloud provider directly competing with AWS, Azure, and Google Cloud." Building the enterprise sales, compliance, and B2B operational muscle required for that status by 2028 is a steep organizational climb, even with the recent poaching of AWS veteran Dave Brown 2 sources. Conversely, a complete lack of external business (None at 13%) is highly unlikely. Meta has already established a commercial monetization channel by launching the paid Muse Spark 1.1 developer API in July 2026 2 sources, which explicitly qualifies under the question's criteria, virtually ensuring at least a baseline level of third-party revenue by the end of 2028.