OpenAI
2026-07-24
OpenAI Launches Project Camellia: $20B Self-Built AI Data Center, Shift from Cloud Renting to Ownership
Summary
Key Takeaways
On July 22, 2026, OpenAI announced Project Camellia, its first self-designed and developed large-scale AI data center in Effingham County, Georgia, spanning 2,600 acres with a capital investment of $20 billion and a 3.2GW power supply agreement with Georgia Power (phased delivery 2028-2032). This marks OpenAI's shift from cloud renting to self-build and control. It simultaneously raised its 2030 compute spending forecast from $600 billion to $750 billion (+25%).
On the personnel front, OpenAI hired two core members of xAI's Colossus supercomputer: Uday Ruddarraju (promoted to Compute CTO, reporting to Greg Brockman) and Brent Mayo (Head of Data Center Construction and Delivery), forming its first in-house data center core team. Known compute contracts include Oracle 6GW, Amazon AWS 8-year $138 billion (including 2GW Trainium), and Microsoft Azure $250 billion.
The project uses a closed-loop water system, akin to a car radiator, and commits $80 million in community benefits. Strategic implications include AI capex as national infrastructure, volatile compute budgets, and accelerated regionalization of AI data centers.
Why It Matters
OpenAI's self-built data center is ostensibly a tech upgrade but essentially a defense against cloud provider control and a encirclement of competitors like Anthropic. By self-building, OpenAI shifts control from cloud providers to itself, but implicitly locks into specific hardware supply chains (e.g., NVIDIA GPUs) and power infrastructure. Delays in power delivery or GPU upgrades could create compute bottlenecks.
The announcement downplays the physical limitation of phased 3.2GW power delivery (2028-2032); until 2028, OpenAI still relies on cloud providers who may raise prices. The closed-loop water system may be insufficient for high-density AI training, hinting at cooling constraints. The Colossus team's rapid deployment experience may not scale to this massive project, posing replication risks. The $20B capex could strain finances if revenue growth disappoints.
PRO Decision
For competitors like Anthropic, they should accelerate self-built data centers or secure long-term exclusive cloud deals to capitalize on OpenAI's construction window (2028-2032). Microsoft should reduce dependency on OpenAI by boosting its own AI models and investing in other AI startups, hedging against OpenAI's potential reduction in Azure usage.
For enterprises, CIOs should be wary of service lock-in from OpenAI's self-built infrastructure; OpenAI may prioritize its own training over external API reliability. Adopt multi-cloud AI strategies to avoid single-vendor dependency and monitor OpenAI's financial stability amid volatile compute spending.
For investors, see through the raised compute spending forecast ($600B→$750B) as a potential financial risk if revenue growth lags. Construction delays and cost overruns are likely understated. Compare capital efficiency across AI vendors to assess long-term competitiveness.
Get 3-5 key AI infrastructure signals weekly →
💬 Comments (0)