Amazon Web Services (AWS) is doubling down on its AI ambitions with a $20-billion expansion plan to build two new data center campuses in Pennsylvania, including one directly adjacent to a major nuclear power plant, Reuters reports.
AWS is targeting the deployment of multiple data centers over the next 10 years, and the buildout will be fueled by carbon-free nuclear power, making it one of the largest private-sector nuclear-backed energy deals in the U.S. to date, according to OilPrice.
The first site, slated for Salem Township near the 2.5 GW Susquehanna Steam Electric Station, leverages a standing engineering framework based on the campus’s 960 MW design capacity.
Amazon is partnering with Talen Energy, a former power utility-turned-nuclear innovator, which will supply the cloud giant with electricity from its Susquehanna nuclear power station, located in Luzerne County. Talen previously spun off its nuclear arm into Cumulus Data, which is developing a 475 MW data center campus adjacent to the power plant. That infrastructure will now be part of Amazon’s AI backbone.
That project is currently under FERC review after regulators capped its supply to 300 MW, citing grid reliability concerns. Still, AWS is pushing ahead, eyeing renewable-like stability without the typical grid bottlenecks.
Analysts say the move could accelerate the return of baseload nuclear as a strategic energy asset in the U.S. data economy. Pennsylvania Governor Josh Shapiro called the deal the largest in the state’s history, with construction expected to generate over 1,250 union jobs in the near term.
"Pennsylvania is competing again—and I'm proud to announce that with Amazon's commitment of at least $20 billion to build new state-of-the-art data center campuses across our Commonwealth, we have secured the largest private sector investment in the history of Pennsylvania," said Shapiro.
In the broader energy context, Amazon’s bet aligns with a rising wave of private-sector clean energy procurement that hopes to successfully sell a different story about AI’s energy use: That hyperscalers can reframe this as ESG-possible.