The Scale of 10GW
OpenAI announced it has signed contracts for 10 gigawatts of AI compute capacity in the United States. To put this in perspective: 10GW is roughly equivalent to the entire electricity consumption of Denmark. It's enough power to run approximately 2 million H100 GPUs simultaneously.
This milestone was originally planned for 2029 but was achieved three years early, driven by partnerships with Microsoft, Oracle, and several new data center operators. The capacity is distributed across Texas (4GW), Virginia (3GW), Nevada (2GW), and Ohio (1GW).
The Energy Question
The 10GW milestone raises serious questions about AI's environmental footprint:
- AI compute now consumes approximately 3% of US electricity generation
- Data center water usage for cooling has increased 40% year-over-year
- Only 35% of OpenAI's contracted capacity comes from renewable sources
- Utilities are delaying coal plant closures to meet AI data center demand
Competitive Landscape
OpenAI isn't alone in the compute arms race:
- Google: Estimated 7GW across global data centers
- Meta: 6GW with plans for 12GW by 2028
- xAI: 4GW in Memphis with expansion to 8GW planned
- Amazon: 5GW through AWS, serving third-party AI workloads
When AI compute uses more electricity than entire countries, we can't pretend this is just a tech story. The good news: OpenAI's aggressive compute buildout is accelerating investment in nuclear and solar. The bad news: in the short term, it's burning fossil fuels. The AI industry needs to treat energy sustainability as a first-class engineering problem, not an afterthought.