rationale:

Final operational capacity estimates for December 31, 2027: p10: 4.8 GW, p25: 6.8 GW, p50: 8.9 GW, p75: 11.4 GW, p90: 14.0 GW.

OpenAI's available compute scaled rapidly through 2025, reaching 1.9 GW across all providers datacenterdynamics.com. Against an internal target of reaching the "low-double-digit" range by 2027 2 sources, the fundamental constraint shifts from capital commitments to the physical realization of secured contracts. While 10 GW in contracts were reported signed by mid-2026 2 sources, transitioning these to operational status inherently encounters multi-year data center construction and interconnection friction.

In-house deployment timelines necessitate a heavy reliance on cloud and neocloud partners for near-term operational scale. With the Stargate Abilene site reaching roughly 0.84 GW of IT power by late 2026 epoch.ai and broader campuses like the Michigan site not expected online until late 2027 or 2028 shortform.com, immediate capacity bridges must run through Microsoft, AWS, CoreWeave, and Cerebras 2 sources. Expectations for massive 2026 capital expenditures and enterprise growth make these committed baseline builds more credible, particularly on the downside, even as their physical execution remains subject to construction schedules.

Applying a physical execution discount to the aggressive "low-double-digits" target yields the final baseline operational capacity of approximately 8.9 GW by the end of 2027. This represents a structurally sound expansion path supported by committed builds and partner capacity, but constrained by utility, construction, and supply chain realities.

The distribution tails reflect the remaining binary uncertainties. Downside risks (p10 of 4.8 GW) account for severe financing limits or stalled partner negotiations driven by operating losses, though committed baseline builds raise the lower percentiles. The right tail (p90 of 14.0 GW) accommodates the potential for near-flawless execution across all partner builds and generous pre-IPO self-reporting metrics, while recognizing that supply chain constraints limit the most aggressive outcomes.