Question
How many gigawatts of total AI data center capacity will OpenAI have operational by December 31, 2027?
Summary OpenAI's total operational data center capacity by the end of 2027 is projected to be around 2.2 GW. As of May 2026, the company has roughly 0.3 to 1 GW operational at its flagship Abilene, Texas campus epoch.ai. This initial campus is highly likely to reach its full planned capacity of 1.2 GW by the end of 2026 epoch.ai. Beyond Abilene, capacity growth relies on additional sites within the massive $400B+ Stargate initiative and various international partnerships. While five new US sites were announced in May 2026, data centers typically require 18 to 36 months from ground-breaking to become fully operational. Given the December 2027 deadline, there is only a 19-month window, placing these new sites at the absolute earliest edge of standard construction cycles. Consequently, only initial modules or phases of the most fast-tracked sites—such as Shackelford County and Milam County in Texas—are expected to be energized in time, each potentially contributing 0.2 to 0.5 GW epoch.ai. International projects, particularly in the UAE (initially 200 MW in 2026, scaling toward 1 GW) and Argentina (100 MW by 2027), will also provide marginal additions intuitionlabs.ai. However, widespread industry headwinds, including supply chain bottlenecks and labor shortages, make it highly probable that actual operational capacity will land closer to the 2.2 GW mark, representing full Abilene completion plus limited initial phases from other early-stage sites.
Strongest Arguments for Higher Values
- Unprecedented Capital and Political Support: With over $400 billion committed to Stargate and potential political tailwinds easing permitting processes, OpenAI and its partners have massive incentives to accelerate construction timelines.
- Modular Energization: Sites do not need to be fully completed to contribute to operational capacity. Rapid, module-by-module commissioning could allow newly announced sites to bring hundreds of megawatts online within the 18-month minimum timeframe 2 sources.
- Rapid Progress in Texas and the Middle East: Sites like Shackelford and Milam have already begun structural or foundation work epoch.ai. Additionally, the UAE partnership's initial 200 MW phase could rapidly scale toward its 1 GW target if early milestones are hit efficiently intuitionlabs.ai.
Strongest Arguments for Lower Values
- Severe Industry-Wide Headwinds: Approximately 40-50% of US data centers planned for 2026 are facing delays 2 sources. Equipment shortages, tariffs on electrical equipment, and labor deficits (e.g., electricians and pipe fitters) are significant bottlenecks arstechnica.com.
- Extreme Supply Chain Lead Times: Transformer lead times have stretched up to 5 years in some markets, and grid interconnection waits can span 4 to 7 years in major data center hubs tech-insider.org.
- Firm Mathematical Constraints: The 19-month window from the May 2026 site announcements barely satisfies the 18-36 month historical construction cycle. Any minor logistical delay will push the operational dates for these new multi-gigawatt sites into 2028 datacenterfrontier.com.
- Project Cancellations and Frictions: OpenAI has already cancelled or paused international sites like Stargate UK and Norway, and local opposition has stalled progress at US sites like Lordstown, Ohio 2 sources.
Key Uncertainties
- Partnership and Co-Location Details: It is slightly ambiguous whether "all OpenAI partnerships" will ultimately include other CoreWeave or Microsoft-hosted capacity beyond the core Stargate-branded sites.
- UAE and International Delivery: The actual execution speed of the UAE and Argentina projects remains uncertain. If international partners bypass the US-centric supply chain issues, they could deliver capacity faster than domestic sites intuitionlabs.ai.
- Supply Chain Resolution: Whether recent massive capital influxes can effectively bypass or solve the current transformer and electrical equipment shortages before the end of 2027.