DOE's $1.9B nuclear loan turns Google's PPA into infrastructure

NextEra Energy has secured a $1.9 billion federal loan to restart the Duane Arnold nuclear plant in Linn County, Iowa, with the revived station's economics tied to a 25-year power purchase agreement with Google, according to Data Center Dynamics. The loan comes from the Department of Energy's Office of Energy Dominance Financing, its third nuclear-restart commitment after $1 billion for the Crane Clean Energy Center at Three Mile Island Unit 1 and $1.52 billion for Michigan's Palisades plant.

Duane Arnold is a 1975 machine being asked to answer a 2029 problem. It entered service in February 1975 as Iowa's only nuclear plant, then shut after the 2020 Midwest derecho damaged its cooling towers and NextEra judged repairs uneconomical. Reports that NextEra might reverse that decision surfaced in 2024, after Microsoft agreed to take 100 percent of the output of a revived Three Mile Island unit in Pennsylvania; in October, NextEra announced the Google contract and gave the restart a public timetable: deliver power by 2029.

Deputy Secretary of Energy James P. Danly cast the loan as part of the administration's nuclear strategy, saying the plant would return 615MW of reliable baseload generation, drive down electricity costs, and support "thousands of American jobs." The politics are visible in that framing.

The interesting layer is not the reactor but the debt. A restarting nuclear plant is a construction project in a 1970s shell: it has spent down its original cost base, gone cold, and must now be repowered on a schedule measured in years. Lenders would normally price that long re-commissioning corridor as merchant risk; Google's PPA replaces merchant risk with a fixed buyer, so the DOE loan becomes a bet on a counterparty with a 25-year load requirement. The underwriting logic is the same as a lease-backed data center debt deal, where the revenue contract is what supports the financing.

The loan office exists to move restart projects off the drawing board. With three loans now on the books, it is effectively telling the private market that a shuttered reactor, when paired with a long-term PPA, has a route to construction capital. That signal may matter more than the $1.9 billion itself.

That is the baseload-scarcity trade this publication has tracked in Google's geothermal pact with Fervo and in the queue politics of data-center siting. Duane Arnold makes the trade explicit: a hyperscaler buys years of guaranteed power, and Washington lends against the purchase.

Duane Arnold is not alone in its buyer class: Microsoft is attached to a revived Three Mile Island unit, Meta has agreed to buy the entire output of Constellation Energy's 1.1GW Clinton Clean Energy Center in Illinois for 20 years, and AWS signed a 1.92GW PPA with Talen Energy for power from the 2.5GW Susquehanna plant in Pennsylvania. The federal loans to Crane and Palisades show the same pattern from the supply side.

None of the four is buying a merchant plant's spot output. Each contract runs roughly two decades or longer, and each is large enough to support a full capital decision. That converts baseload units from grid cost centers into long-dated infrastructure assets, and it makes the PPA do the work an interconnection queue position used to do: establishing that the power will exist when the servers turn on.

Together, the four buyers are creating a private market for firm, dispatchable, low-carbon power, where the buyer is a cloud company with a load curve that never sleeps, rather than a regulated utility.

The three DOE restart loans total roughly $4.4 billion, small against hyperscaler capital expenditure but outsized for a category where private lenders have historically been reluctant to go first. The federal government has moved from lender of last resort to lender of first resort for these projects.

The disclosed terms stop at the headline amount, so the cost to taxpayers cannot be assessed; the structure can. The DOE is lending into projects whose revenue layer has already been placed with some of the strongest corporate balance sheets in the economy.

Duane Arnold shut because a storm made its economics difficult. What changed was the buyer: Google signed for the output, and that signature was sufficient to bring a $1.9 billion federal check behind a 50-year-old plant. Microsoft's earlier commitment did the same at Three Mile Island, making the hyperscaler power purchase agreement the instrument that turns decommissioned infrastructure back into financeable assets.

The date that remains unresolved is 2029. Google's contract has been signed, the loan has been secured, and the plant has a target for first power. Between now and then, the project must clear the ordinary risks of nuclear restart: re-licensing, component replacement, and the slow process of proving that a plant idle since 2020 can run safely again. The DOE loan cannot retire those risks, only provide the patient capital to live through them.

The federal government has decided that contracted baseload power is worth financing at a time when the grid is short of firm capacity and the data center buildout is waiting on electrons. Duane Arnold will test that decision on a hard deadline in 2029, and the result will set the price of the next restart.