The dollars behind a very specific comeback
Nuclear power spent years in retreat after the Fukushima disaster reshaped public and regulatory appetite for it worldwide. Data Center Knowledge reports that AI's electricity demand is now reversing that trend in the United States specifically, driven by hyperscalers that have run out of patience with grid interconnection queues. Industrial Info Resources puts new nuclear project spending at roughly 20 billion dollars in 2026, rising to nearly 50 billion dollars by 2029, with another 68.3 billion dollars committed for new plants between late 2026 and 2029 and a further 30 billion dollars going toward maintaining and upgrading existing facilities.
IIR tracks more than 100 North American nuclear projects in total, including 61 small modular reactor projects and 11 new conventional reactor builds, alongside a wave of life extensions, uprates, relicensing efforts, and restarts of plants that had been written off as retired. Ontario and Alberta combined account for roughly 40 billion dollars of this activity, with Texas at about 15 billion dollars and Tennessee around 10 billion dollars. IIR analyst Britt Burt frames it plainly: investment is rising for new and existing nuclear assets amid what he calls a sector revival, driven by reliability, zero emissions, and insulation from fuel price swings.
Washington puts a loan behind the oldest reactors
The federal government has moved beyond rhetoric. The Department of Energy offered Constellation Energy a loan of up to 5 billion dollars to fund uprates and modernization across eight existing reactors spanning Illinois, Maryland, and Pennsylvania, a package that could add more than 1 gigawatt of capacity to the grid without building a single new reactor from scratch. Earlier in 2026, the DOE also launched a separate program targeting 2.5 gigawatts of additional nuclear capacity by 2027 and 5 gigawatts by 2029.
The strategic logic is straightforward: existing reactors are the fastest nuclear capacity to bring online, faster than any new build, because the licensing, the site, and most of the physical plant already exist. Boosting output from operating reactors, reopening dormant ones, and extending the lives of aging plants is a far shorter path to new megawatts than greenlighting a reactor that has never generated a watt of power. It is also, not coincidentally, the part of this story closest to actually helping anyone's 2027 capacity plan.
Each hyperscaler picked a different nuclear bet
The specific deals are worth knowing by name, because they are not interchangeable. Amazon is working with American Electric Power and X-energy on a 1 gigawatt small modular reactor project in Indiana, betting on next-generation reactor technology that has not yet been proven at commercial scale. Microsoft went the opposite direction, signing a 20-year power purchase agreement with Constellation Energy to draw from the refurbished Three Mile Island plant in Pennsylvania, now rebranded as the Crane Clean Energy Center, a bet on restarting proven, already-licensed infrastructure rather than waiting on new technology.
Google split the difference again, partnering with NextEra Energy Resources to restart the Duane Arnold Energy Center, a plant that had been shut down since 2020. Three hyperscalers, three distinct nuclear strategies: new SMR technology, a refurbished historical asset, and a straightforward restart. That spread itself is a signal. If any one path were clearly the fastest or cheapest route to reliable nuclear capacity, you would expect more convergence rather than three different companies placing three different bets.
Fusion is a long-dated option, not a current supply
Fusion gets attention at every energy conference adjacent to AI, and the companies working on it are honest, when asked directly, about the actual timeline. Type One Energy, based in Knoxville, Tennessee, plans to begin building a fusion power plant in 2027 at the Tennessee Valley Authority's Bull Run Energy Complex, with design elements already running at the University of Wisconsin-Madison and at Germany's Wendelstein 7-X project. Its own business development director has described the planned prototype as meant to validate assumptions and train a workforce, not to meet any data center's near-term power needs.
General Atomics Magnetic Fusion Energy puts the realistic commercial timeline even further out, saying grid-connected demonstration plants and power purchase agreements may not arrive until the mid-2030s, with gigawatt-scale deployment likely waiting until the early 2040s. Treat every fusion headline attached to a data center announcement accordingly: it is a research and workforce investment with a decade-plus horizon, not a line item that changes anyone's power procurement plan for this decade.
The regulatory clock is finally moving faster than the demand clock expected
Nuclear projects have historically been slowed as much by permitting and review timelines as by construction itself, and that is the piece showing the most near-term movement. Cameron Hughes, an associate at law firm Orrick, says recent reforms have cut key regulatory reviews to roughly 18 months, down meaningfully from the multi-year reviews that used to be standard, alongside gains from modularization, factory fabrication, and a maturing supply chain for reactor components.
Hughes argues nuclear needs an all-of-the-above approach, with the market ultimately deciding which reactor types and ownership structures prove most viable over time, rather than any single technology winning outright. That is a reasonable read of where things stand today. Small modular reactors and plant restarts are the parts of this buildout likely to matter on a timeline relevant to current data center planning, with the first new SMRs potentially online within a couple of years, well ahead of fusion and most new conventional builds.
What this means for your power procurement timeline
If your organization is counting on nuclear-backed cloud capacity to ease a power crunch in 2026 or 2027, the honest read of this buildout is that very little of it lands that soon. The fastest wins are uprates and restarts of plants that already exist, which is exactly where the Constellation loan and the Microsoft and Google deals are pointed. New SMR technology, where Amazon has placed its bet, carries more technical risk and a longer runway before it contributes real megawatts.
For enterprise planning purposes, nuclear-backed power purchase agreements are a multi-year supply diversification story worth tracking in your vendor's power mix disclosures, not a lever that resolves this year's capacity constraints. Ask your cloud and colocation providers specifically which of their committed nuclear capacity is contracted for delivery within your planning horizon, and which is aspirational beyond 2030, because those two categories get conflated in press releases far more often than they should be.
