Why Savannah River’s AI Nuclear Deal Is Not Yet a Construction Start

The United States government is preparing to pair a 1-gigawatt artificial intelligence data center with roughly 2 gigawatts of dedicated power generation at a Cold War nuclear weapons site in South Carolina. It is one of the most ambitious energy commitments the federal government has attached to AI infrastructure so far.

Yet not a single small modular reactor technology has been named. No lease has been signed. No construction date has been set. The distance between the ambition and the current reality is where this story actually sits.

On 20 July 2026, the Department of Energy’s National Nuclear Security Administration (NNSA) selected Amentum to enter negotiations, not to build, for a phased lease at the Savannah River Site. That distinction matters, and it is easy to miss beneath the headline capacity figures.

Savannah River is one of four federal sites the current administration has advanced under its AI and energy executive orders, which places this deal inside a deliberate national strategy rather than treating it as a one-off. This piece separates what the project has actually committed from what remains speculative, explains why the federal site model is strategically sound even before commercial reactors exist, and shows what the gas-to-nuclear pathway means for anyone tracking the DOE AI data center nuclear project as an investment signal.

What the NNSA-Amentum deal actually commits to, and what it does not

Start with the verified facts. On 20 July 2026, the NNSA selected Amentum to negotiate a phased lease for a 1-GW AI data center paired with approximately 2 GW of on-site power generation at the Savannah River Site. DC BLOX has been confirmed as the digital infrastructure partner.

That is the headline. The fine print underneath it changes how the headline should be read.

The DOE’s own announcement is explicit that this selection does not amount to a lease award. Any agreement remains contingent on permitting, safety and security evaluations, and other federal approvals.

Federal nuclear energy oversight shapes every stage of what Savannah River must clear before a lease is executed, from NNSA security evaluations to NRC environmental reviews, and understanding the regulatory architecture explains why the contingency language in the DOE announcement is substantive rather than boilerplate.

“Selection does not constitute a final lease award.”

An 7 August 2026 analysis by the energy-project publication mGrid put it plainly: NNSA selected Amentum to negotiate, not to construct, and permitting and security reviews still lie ahead. A 31 July 2026 summary from the American Nuclear Society (ANS) reiterated the same contingency language directly from the DOE.

So what is genuinely locked in, and what is still open? The gap is wider than the announcement suggests.

The absence of a vendor, a tenant, and a signed lease tells you this project sits at its most important decision-making point, not past it. The signals Amentum and the DOE send over the next 12-18 months will matter more to the nuclear-AI investment thesis than the July announcement itself. A federal selection is meaningful validation, but treating it as a construction commitment would produce a mispriced risk assessment.

Why the federal site model is strategically rational even before SMRs exist

An incomplete project can still rest on sound logic. The federal reservation model solves a specific set of problems that private nuclear-AI campuses struggle to solve on their own, and understanding why changes how every comparable deal should be read.

Start with the grid. The 2 GW of behind-the-meter generation means the regional grid does not have to absorb the new load. No major transmission upgrades are required, and no ratepayers are cross-subsidising private infrastructure.

That last point is the one the DOE emphasises most. Federal statements frame the on-site generation model as a way to meet very large electricity demand without shifting costs to existing utility customers. Surplus generation is intended to increase local power availability while keeping the cost of new capacity off household bills.

There is a second advantage buried in the mechanics: behind-the-meter siting bypasses the grid interconnection queue. New generation that plugs directly into a dedicated load sidesteps the 8-10 year interconnection timelines that private developers routinely face.

Savannah River is not being used in isolation. It is one of four DOE sites advanced for private-sector AI and energy partnerships under the 2025 executive orders on data center permitting and advanced nuclear deployment.

1. Savannah River Site (South Carolina): the Amentum data center and generation project.
2. Paducah Gaseous Diffusion Plant (Kentucky): a legacy uranium enrichment site being repurposed for a large AI campus.
3. Idaho National Laboratory: a federal research site hosting proposed nuclear-powered compute.
4. Oak Ridge (Tennessee): a legacy nuclear reservation advanced for private partnerships.

For investors, the federal designation functions as a form of regulatory pre-clearance. It meaningfully de-risks the permitting phase compared with a private greenfield SMR project, and that compression of permitting risk is one of the clearest investable signals in this deal.

The security and siting advantages private developers cannot replicate

Federal reservations come with things money struggles to buy quickly. Savannah River already has a nuclear security perimeter, established regulatory relationships, and transmission infrastructure in place. A private developer starting on open land has none of these and would spend years building them.

Fuel supply operations at Savannah River extend beyond the data center project; the site’s H Canyon facility has recently begun recovering uranium for advanced reactor programmes, adding another layer of strategic significance to the DOE’s decision to advance this particular reservation for public-private nuclear partnerships.

There is a policy dimension too. Federal site participation lets the DOE and NNSA test public-private leasing models that could later be replicated at Paducah, Idaho, or Oak Ridge. That makes Savannah River a template as much as a project, which is why its execution carries weight beyond its own fence line.

What the energy transition in this deal actually looks like

The project’s energy plan is sequential, and the risks emerge from that sequence rather than sitting beside it as caveats. Initial power comes from natural gas. A transition to advanced nuclear follows. No specific SMR design has been named for that second phase.

That ordering is a pragmatic response to where the technology actually stands. No SMRs operate commercially in the United States today. Every such project remains in licensing or permitting.

Independent technical reviews target first commercial SMR deployments around 2030, with significant market penetration not expected until roughly 2035. NRC licensing, environmental review, and site-specific safety evaluations can each take years. Layer them, and the nuclear phase realistically arrives in the early 2030s at the soonest.

The economics carry their own gap. A 2025 analysis of SMRs for AI workloads put levelised cost of electricity at USD 89-102/MWh, well above the USD 55-58/MWh that vendors have claimed.

That USD 30-45/MWh gap between vendor claims and independent analysis is the single most important number to hold when evaluating whether the nuclear phase delivers on its economic rationale. It is where optimism and reality diverge most sharply.

SMR economics for industrial loads follow a different cost curve than utility-scale deployments because dedicated behind-the-meter installations avoid transmission costs while bearing the full capital burden of a smaller fleet, a dynamic that partly explains the gap between vendor LCOE claims and independent analysis figures.

The gas bridge carries the sequence’s central risk. If SMR schedules slip, the natural gas plants may run far longer than planned, potentially locking in emissions or leaving stranded gas assets once reactors eventually come online.

Amentum’s own framing acknowledges the complexity of the second phase.

Deploying nuclear capacity at sufficient speed requires partners versed in licensing, permitting, engineering, project delivery, and ongoing operations, not just innovative reactor designs, according to Gareth Vaughan, Vice President of Amentum North America Nuclear.

The gas-to-nuclear pathway is not a design flaw. It is a candid acknowledgment of where SMR commercialisation actually stands. But the distance between the bridge and the destination is exactly where the investment risk concentrates.

The broader signal: how federal site deals are reshaping SMR commercialisation

Lift the view above Savannah River and a pattern comes into focus. AI data center demand has changed nuclear’s commercialisation calculus, and that shift is the more durable story.

For years, nuclear developers waited on grid decarbonisation policy to drive orders. Now they have a class of buyers with firm, 24/7 power requirements that align precisely with nuclear’s output profile. AI compute runs continuously, and continuous baseload is exactly what reactors provide.

High-profile federal anchor projects accelerate two things when they proceed: investor confidence and regulatory learning. If Savannah River advances smoothly, it compresses uncertainty for the reactor design it eventually selects. Visible delays or cost overruns would do the opposite, reinforcing scepticism about nuclear’s role in digital infrastructure.

Savannah River is one of several parallel tracks, not the only one. A cluster of comparable deals is emerging across federal and commercial sites.

The ANS cited Paducah as the most direct analogue to Savannah River, framing both as case studies in leasing federal nuclear reservations for AI campuses. Deep Atomic’s 9 July 2026 proposal for an integrated MK60 campus at Idaho National Laboratory and the AWS-Dominion exploration near North Anna show the same logic playing out on both federal and commercial ground.

The structural case for nuclear over renewables for these specific loads rests on several points:

• Baseload match: continuous output aligns with continuous AI workloads.
• Asset life: reactors are designed for 60-80 years of operation.
• Land efficiency: far smaller footprint than equivalent solar capacity.
• Carbon-free output: supports corporate clean-energy pledges.
• Behind-the-meter siting: bypasses multi-year grid interconnection queues.

The convergence of federal reservations, commercial utility deals, and private developer agreements tells you that nuclear-AI co-location is becoming a structural market phenomenon rather than a scatter of speculative one-offs. For investors, that proliferation is the more durable signal than any single announcement. AI energy demand is now functioning as a genuine commercialisation catalyst for nuclear, not merely a narrative backdrop.

What remains unresolved, and why those gaps matter more than the headline

Before treating Savannah River as a committed infrastructure project, look at what is still missing. Three variables remain open, and each one carries analytical weight.

1. SMR vendor and technology selection. No reactor design has been named as of August 2026. Whichever design wins receives a high-profile federal anchor load that could reshape its commercialisation trajectory and pull in follow-on investment.
2. Lease finalisation and the permitting-security sequence. No lease is signed. Any agreement remains contingent on permitting, safety, security, and federal approvals, per the DOE and NNSA language from 20 July 2026.
3. End-user identification. No data center operator has been publicly named. The project’s commercial viability depends on a long-term power purchase or lease commitment from a tenant, and no such party has surfaced.

Advanced reactor commercialisation is tracking along a broader global curve, with several non-US programmes moving faster through licensing than domestic equivalents, which matters for the Savannah River vendor selection because a design with international operating experience carries meaningfully lower regulatory risk than a first-of-kind US deployment.

Vendor selection matters well beyond this one site. A federal anchor load is the kind of validation that can move a reactor design from the drawing board toward a fundable commercial future, which is why the eventual announcement will be read as a sector signal, not just a project update.

The missing tenant is a genuine information gap rather than a formality. Amentum has stressed that successful deployment requires partners across licensing, permitting, engineering, project delivery, and operations. A committed end-user with firm, long-term power demand is the commercial spine the whole project hangs on.

Until a vendor is named, a tenant is confirmed, and a lease is executed, Savannah River should be read as a strong federal policy signal and a legitimate proof-of-concept for the federal site model. It is not yet a committed infrastructure project that moves SMR deployment timelines.

What the information gaps tell investors right now

Treat the absence of a vendor, tenant, and lease as a diagnostic tool rather than a red flag. These gaps are not unique to Savannah River. They reflect the standard pipeline stage for nuclear-AI deals in 2026.

That framing has a practical consequence. Investors who read each federal announcement as a construction commitment will systematically misprice the sector. The next 12-18 months give a concrete watch-list: vendor announcement, lease execution, and end-user identification are the three events that would convert this from a policy signal into a fundable infrastructure thesis.

This article is for informational purposes only and should not be considered financial advice. Investors should conduct their own research and consult with financial professionals before making investment decisions. Past performance does not guarantee future results. Financial projections are subject to market conditions and various risk factors, and forward-looking statements are speculative and subject to change based on market and regulatory developments.

Making sense of Savannah River in a sector still finding its footing

Here is the verdict, stated plainly. Savannah River is the most consequential federal proof-of-concept for the nuclear-AI co-location model currently in motion, and its value to the SMR investment thesis lies in the federal site precedent it establishes, not in any near-term construction.

The project embodies a structural tension worth naming directly. The government is moving faster on public-private nuclear-AI partnerships than the SMR industry can commercially deliver. Commercial reactor deployments sit around 2030, with significant penetration near 2035, while data center construction cycles run just 2-4 years. The gas bridge is the honest acknowledgment of that mismatch.

What should not be understated is the durability of the policy signal. The willingness to designate federal nuclear sites for private-sector AI infrastructure, now stretched across four DOE locations, is not a commitment that walks back easily. For investors, that institutional shift is the takeaway from 2026 regardless of whether this specific project executes on schedule.

The mGrid analysis from 7 August 2026 framed it well: selection is not construction, and the gaps ahead are both procedural and substantive. Watch these four indicators over the next 12-18 months:

• An SMR vendor and reactor technology are named.
• The phased lease is executed with permitting and security clearances.
• An end-user data center operator is publicly confirmed.
• Formal NRC licensing engagement is initiated.

Treat federal site designations as durable policy signals, vendor and tenant announcements as the fundable milestones, and an actual SMR construction commitment as the event that finally moves this thesis from analytical to investable.