Last week, Google announced a new partnership — it’s going nuclear (again). The tech giant signed a strategic agreement with Elementl Power, a nuclear project developer, to co-develop three advanced nuclear power projects. Each one is targeting 600MW of capacity. If completed, that’s 1.8GW of clean, always-on electricity, enough to power millions of homes or a whole lot of AI model training runs.
What happened
Elementl, a tech-agnostic nuclear developer founded in 2022, is working to bring more than 10GW of new nuclear online in the US by 2035. This deal with Google marks its first public customer, and its biggest vote of confidence yet.
Under the agreement, Google will provide early-stage capital to pre-position three project sites. That means funding upfront work like permitting, interconnection rights, and site selection — taking on some development risk directly. Then, Google has the option to purchase power from the reactors once online.
This isn’t Google's first nuclear rodeo. The company already has a deal with Kairos Power, an SMR startup, with plans for commercial generation by 2030. But the scale and urgency of this latest agreement signal a new level of commitment — and a clearer line between AI and energy.
Why it matters
Google and other hyperscalers like Microsoft and Amazon are racing to build out massive new data center capacity — each betting that AI will define the next decade of business. That investment is as existential as it is exponential. Google alone plans to spend $75bn this year on infrastructure to train and deploy advanced models.
The AI race isn’t just about better models anymore — it’s about who can lock in power first. Solar and wind are cheap and abundant, but intermittent, so tough for always-on AI workloads. Advanced nuclear like SMRs offer firm, carbon-free energy in compact, factory-built units that can plug-and-play into existing grid sites.
But SMRs have long been stuck in the promise phase. As of mid-2025, none have been built outside of China and Russia. NuScale, once the US frontrunner, saw its flagship project canceled in 2023 after costs ballooned. But momentum is picking up. The sector is getting more creative, with new technologies, new buyers, and now, new development models.
Just last week, the US Nuclear Regulatory Commission began reviewing the Tennessee Valley Authority’s application to build a 300MW GE Vernova Hitachi SMR at the Clinch River site in Oak Ridge, TN, kicking off with public meetings to discuss the permitting process. And across the border, Ontario Power Generation got the green light to begin construction on its grid-scale SMR, also a GE Vernova Hitachi, at its Darlington site, the first SMR to do so in North America.
Sightline clients can access our full dataset of SMR projects, as well as deployment charts and tech profiles, on the platform.
Elementl hasn’t picked a specific reactor technology yet but says it will select the most commercially ready option when the time comes. Contenders include different nuclear technology types like:
- Light-water reactors: GE Vernova Hitachi, Holtec, NuScale, and Last Energy all benefit from regulator familiarity and existing fuel supply chains, which can shorten timelines and reduce costs.
- Gen IV SMRs: Kairos Power, Terrapower, Oklo, and X-energy are pursuing alternative approaches like helium or liquid metal coolants. Though less mature than LWRs many can still draw on decades of development. These offer potential safety and performance advantages but may face longer licensing timelines and issues sourcing their specialized fuel.
Based on Elementl’s timeline (site permitting/early development from 2025-26; construction permitting from 2026-28; break ground in 2029, and begin operations 2032-33), they’ll likely opt for a small light-water design. Of the list, GE Vernova Hitachi looks best positioned to meet that schedule.
Overall, three signs of SMR market maturity stand out in this deal:
- Development risk sharing. Google isn’t just signing a PPA — it’s funding early-stage project development. Elementl’s bet is that hyperscalers — with huge demand and deep pockets — are better suited than traditional regulated utilities to take development risk in a sector where early-stage costs can reach hundreds of millions.
- Grid-native siting. These reactors will be grid-connected — not behind-the-meter — unlocking access for utilities and public markets. Transmission access is a gating factor, and Elementl is specifically choosing sites with existing infrastructure (e.g. old coal or industrial sites) to avoid long interconnection timelines. That’s a key sweetener for grid operators increasingly wary about grid reliability and new large data center loads.
- Capital agility. Elementl’s strategy is to secure land and interconnection first, then select tech. It’s a solar-style playbook — de-risk early, move fast later — but adapted for gigawatt-scale clean firm power.
What’s next
- Policy make-or-break. In the past, three policy levers have mattered most: (1) the IRA tax credits, which dramatically improve project economics, (2) DOE Loan Programs Office support, essential for large-scale builds, and (3) licensing speed, which is still a bottleneck despite aims to speed up. According to the DOE's Liftoff report, combining IRA + LPO funds can halve the cost of new nuclear. But all three are under pressure from the Trump admin, which has cut staff, proposed lower budgets and removing tax credits, and possibly politicized the NRC process.
- Timing is everything. Data center developers want megawatts now, not in 2033. But with clean firm options scarce, SMRs are in the running again. The bigger picture: this isn’t just a tech bake-off. SMRs are competing not just with each other, but with gas + CCS, geothermal, and renewables + storage in the emerging clean firm stack. Unlike in the 2000s — when the shale boom killed off US nuclear momentum — natural gas prices are no longer the same blocker.
- But can they deliver? Elementl CEO told Catalyst that the goal is <$10,000/kW capex and <6-year deployment cycles — faster than China, and faster than the NRC has ever licensed a project. That would be a sea change. But first, they have to build the SMR — three times. Still, iteration and lessons learned can drive down costs and timelines.
.png)