Executive Summary

The U.S. Department of Energy has selected Amentum to develop and construct a 1 GW AI data center campus supported by approximately 2 GW of on-site energy generation. The project will initially be powered by natural gas generation with a planned transition to advanced nuclear energy. While deal size was not disclosed, a 2 GW co-located generation and data center project of this scale typically represents billions in total capital deployment when accounting for both power generation assets and data center infrastructure.

The Players

Amentum is a government services contractor with deep experience in DOE site management and large-scale infrastructure projects. The company will serve as the primary developer and construction manager. The DOE’s direct procurement signals a strategic federal priority: building sovereign AI compute capacity with dedicated, on-site dispatchable power that doesn’t strain regional grids or depend on utility interconnection queues.

The Numbers

2 GW of generation capacity is enough to power a Tier 1 AI training facility or a distributed inference network. At current data center power densities (20-40 kW per rack for AI workloads), this could support 50,000 to 100,000 GPU-dense racks. Natural gas generation at this scale would likely involve 4-6 large-frame turbines in combined-cycle configuration, delivering capacity factors above 85%.

The gas-to-nuclear transition timeline is critical but unspecified. Realistically, CCGTs can be online in 24-36 months. Advanced nuclear, even on an expedited DOE site with streamlined NRC review, is 2030+ for first power.

So What?

This is the federal government validating the behind-the-meter, gas-bridge-to-nuclear model that private developers have been pitching to hyperscalers for two years. It removes any remaining doubt that AI data centers will require dedicated, dispatchable generation — not grid power, not renewables-plus-storage, but firm, controllable baseload.

For power developers: DOE just set the standard. If you’re developing AI-ready generation, your pitch needs to include a nuclear offramp, even if you’re starting with gas. We should still expect a decarbonization pathway, and gas alone won’t satisfy that 2030-2035 horizon.

For nuclear developers: this is your proof of concept. If DOE can de-risk this gas-to-nuclear AI data center, the private market will follow. The question is who gets the backend contract and whether they can deliver on time.

What to Do:

If you’re a gas-fired developer, start building relationships with advanced nuclear vendors now. If you’re an EPC, get fluent in hybrid generation site design. If you’re a capital provider, the behind-the-meter AI power market just got a federal stamp of approval — and it’s moving faster than the grid ever will.

Source: Business Wire, July 20, 2026