AI Drives the Nuclear Comeback – A Sustainable Comeback

(Photo by Mick Truyts on Unsplash) Governments, utilities, and big tech push nuclear back into the center of power planning. You see it in the decisions taken worldwide: life-extensions of older nuclear installations, restarts of previously retired units, and fresh construction on paused sites. Not all that surprising, in the past I already wrote about the return of nuclear energy as a sustainable energy source.

But the renwed attention for nuclear energy – as a way to cut emissions, offer energy security, and produce a round-the-clock supply for data-hungry economies – is increasing. At COP28, no less then 22 countries pledged to triple global nuclear capacity by 2050, a coalition that kept growing through 2024–2025. Also major corporations like Amazon, Google, and Meta publicly backed the goal.

In this article I will explain why restarts are happening now, and it has indeed a lot to do with AI, but not only that.

Where restarts of nuclear reactors and extensions are happening

Not every country moves in the same direction. Some phased out completely (e.g., Germany), while others add capacity (China, Finland, UAE) or extend lifetimes. The global picture is mixed but nevertheless trending toward using existing nuclear assets longer while adding new units where policy and financing align.

Here are the countries that will restart their retired plants or that go for a life extension of their existing nuclear plants.

United States: the first restart of a retired plant

Michigan’s Palisades shut in 2022, but now the plant won NRC approvals to load fuel and move from decommissioning back to operational status – a U.S. first. DOE is disbursing a federal loan guarantee as work advances toward returning the 800-MW reactor to the grid.

Belgium: reversing a phase-out with targeted extensions

Belgium abandoned its disastrous energy policy led by the Groen and Ecolo, and approved the restart and 10-year life extension of Tihange 3 with plans to operate Doel 4 and Tihange 3 to the mid-2030s (with regulatory processes continuing at EU level). On the grid, the tangible outcome is quite clear: Belgium keeps 2 GW of firm, low-carbon supply while rolling out more renewables and CRM capacity. Prudent or a mess? It depends on how you weigh timing, risk transfer, and consumer cost.

Japan: steady restarts after Fukushima safety overhauls

Japan now counts a growing fleet back in service – with high-profile restarts like Onagawa-2 and additional units in the queue (e.g., Shimane-2). These moves will cut LNG imports and support new chip and data-center builds.

South Korea: policy reset and construction resumes

The Yoon administration ended the prior phase-out, set a ≥30% nuclear share by 2030, and restarted construction on Shin Hanul 3 & 4.

Canada: keep running now, refurbish for decades

CNSC authorized Pickering B (Units 5–8) to operate until end-2026 while Ontario advances refurbishment plans. The province ties this to electrification and rapidly growing load.

Why does AI so much need nuclear power?

AI doesn’t take a nap. Training runs and inference farms pull an extreme amount of power every hour, week after week. You must feed that load with clean, steady megawatts which are not too pleased with weather swings. That’s where nuclear fits. It delivers 24/7 energy with high capacity factors, compact land use, and price stability that keeps total cost of ownership in check. Pair it with wind, solar, and batteries, and you get a grid that stays green during calm, dark spells when demand spikes and renewables dip.

Let’s check each element:

• 24/7 load that never ends: AI training clusters and inference farms draw power day and night. A single hyperscale AI data center can pull 100–300 MW; multi-site campuses cross 1 GW. Wind and solar swing with weather and daylight. Nuclear runs at high capacity factors (~90%), so you get steady megawatts every hour of the year.
• Carbon targets that measure every hour: Big tech now chases 24/7 carbon-free energy (hour-matched). Gaps in wind/solar output force fossil peakers unless you have firm clean power. Nuclear fills those gaps without CO₂, so your hourly scorecard stays perfectly green.
• Storage isn’t built out for weeks of coverage: Four-hour batteries help with daily peaks. They don’t carry a region through multi-day low-wind, low-sun periods. Long-duration storage is growing, but not at the terawatt-hour scale AI loads demand this decade. Nuclear provides the backbone so storage and demand response can play their roles.
• Grid constraints and siting reality: You can’t always run new high-voltage lines fast enough to stitch together distant renewables. Nuclear lets you co-locate firm supply near AI campuses, reuse existing grid interconnects, and reduce the need for sprawling new corridors.
• Price stability for capex-heavy compute: AI costs hinge on electricity. Gas-linked prices are volatile. Nuclear PPAs can lock in long-term, predictable tariffs, derisking multi-billion compute investments and keeping total cost of ownership under control.
• Low lifecycle emissions, tiny land footprint: Lifecycle carbon for nuclear clusters around single-digit to low-teens gCO₂e/kWh, comparable to wind and below most solar medians when you include manufacturing. Land use is compact. You avoid the acreage and transmission build that slow other clean builds.
• Thermal by-products you can use: AI racks create heat; so do reactors. Pairing the two unlocks district heating, absorption cooling, or direct-use process heat. That boosts overall system efficiency and local community value.
• System resilience when it counts: Cold, windless weeks are when grids get stressed. AI demand won’t pause. Nuclear keeps factories, hospitals, and your clusters online in those hours, cutting blackouts and price spikes.

Nuclear provides firm, low-carbon megawatts that keep AI running through every hour of the year while stabilizing costs and the grid. Not surprisingly several major corporations backed the goal to triple global nuclear capacity by 2050.

In March 2025, Amazon, Google, and Meta joined a cross-industry pledge announced during CERAWeek in Houston. They then followed up with real procurement moves – PPAs, joint-buyer initiatives, and early agreements with advanced-nuclear developers – that channel money and demand toward nuclear capacity through 2030–2050.

Let’s see what each company actually did.

Amazon

• Locked in nuclear power for data centers. AWS bought a 960-MW data-center campus adjacent to the Susquehanna nuclear plant and later entered a 1,920-MW PPA with Talen to supply Pennsylvania data centers.

Google

• Aggregated demand for advanced nuclear. Google, with Microsoft and Nucor, launched a joint buying initiative to de-risk first-of-a-kind clean power, explicitly including advanced nuclear.
• Pursued next-gen reactors. Google also announced a clean-energy agreement with Kairos Power aimed at bringing its first SMR online around 2030, scaling toward up to 500 MW of 24/7 carbon-free power by 2035.

Meta

• Closed a 20-year nuclear deal with Constellation for the Clinton Clean Energy Center in Illinois. Meta takes the plant’s output (1,121 MW) via a long-term PPA – an anchor commitment that helps keep the reactor operating after state ZEC support expires in 2027.

Why restart nuclear plants now besides the need coming from AI?

After being demonized for decades, nuclear is back on the table because the grid needs firm, low-carbon power now, not in ten years. Gas shocks exposed the fragility of the energy supply causing blackouts and price spikes. Electrification – from heat pumps to EV fleets and AI data centers – demands round-the-clock supply that wind and solar alone simply can’t cover during calm, dark weeks.

Life-extending existing reactors delivers faster and cheaper than greenfield builds, while reusing sites, staff, and grid hookups. Safety cases tightened post-Fukushima, and real waste solutions – think Finland’s deep repository – have completely shifted the credibility debate.

Policymakers also chase climate targets; keeping proven gigawatts online cuts CO₂ today and reduces the storage and transmission overbuild you’d otherwise need.

In short, the renewed interest in nuclear is based on 4 elements:

• Firm low-carbon supply for a peaky, electrified economy: You need 24/7 power for heat pumps, EV fleets, electrolyzers, AI data centers, and fabs. Nuclear provides dispatchable, high-capacity-factor generation without CO₂ at the point of use. The IEA’s net-zero roadmap still relies on nuclear roughly doubling by 2050.
• Climate math that works: On a life-cycle basis, nuclear’s median emissions sit near 12 gCO₂e/kWh, on par with wind and below solar’s medians in major reviews. That’s because most emissions come from materials and construction, not operation.
• Energy security after the gas price shock: The 2021–2023 gas crisis exposed dependence risks. Countries extended or restarted units to curb gas burn and stabilize markets. Even critics concede nuclear’s value during fuel shocks, though they debate costs.
• Grid integration with renewables: High wind-and-solar systems face dunkelflaute and seasonal gaps. Firm nuclear cuts curtailment and reduces storage needs per MWh of clean energy delivered. It shows that nuclear acts as a system complement to variable renewables and definitely not a rival.

What is ‘dunkelflaute’? A dunkelflaute is a period of time in which little or no energy can be generated with wind and solar power, because there is neither wind nor sunlight. In meteorology, this is known as anticyclonic gloom.

Is reviving shuttered nuclear plants sustainable? Yes, and there’s evidence

Keeping existing reactors running is quite logical, and should be also for those who oppose it. The life-extension of nuclear plants turns these old assets into low-carbon workhorses with a tiny land footprint. You reuse sites and grids, cut the carbon per kilowatt-hour, and shore up winter reliability when wind and solar dip. Meanwhile, real waste solutions – like the earlier mentioned Finland’s Onkalo repository – offer a real storage solution.

The points below break this down in cost, land use, waste, and system resilience.

• Lower carbon per kilowatt-hour: When you keep a reactor running longer, you spread the one-time build emissions over many more megawatt-hours. That lowers the carbon footprint of each unit of electricity. Life-extension projects often deliver some of the cheapest, cleanest power on mature grids.
• Small land footprint, big reuse: Restarting or extending a plant uses the same site, switchyard, and trained crews. You avoid new land take, new transmission corridors, and years of siting fights that slow other clean-energy builds.
• Real progress on waste: Finland’s Onkalo deep geological repository has completed full-system trials and is entering final commissioning, with first disposal expected as licensing wraps. It provides a practical end-state for spent fuel.
• Stronger system resilience: Nuclear power doesn’t depend on weather. It keeps the lights on during cold, windless weeks when demand jumps and variable renewables dip. IEA models use this 24/7 output to hit net-zero targets while holding system costs in check.

Even if you are not convinced, restarting these nuclear plants also means we buy time for grids to build transmission. Keeping reactors running avoids backsliding to gas while you permit lines, storage, and demand-response at scale. On top they anchor clean industrial load. Electrolyzers, e-fuels, and heat-intensive processes need predictable megawatt-hours; nuclear provides that floor. And as for their financial impact, they also stabilize wholesale prices in scarcity hours. Firm, zero-carbon baseload damps volatility, and that is basically one of the biggest cost drivers for end users.

However, despite the above positive points for reviving shuttered nuclear plants, starting new plants remains a not so easy endeavor:

• Cost and delivery risk for new builds. Flagship projects like Hinkley Point C and Vogtle faced overruns and delays. This fuels arguments that renewables plus storage win on price in many regions. Restarts and LTOs, however, avoid greenfield risk and deliver sooner.
• Industry pace vs. targets. The World Nuclear Industry Status Report 2025 warns output could plateau without faster build rates, noting renewables attract much larger annual investment. That’s a real constraint.
• Safety in conflict zones. The Zaporizhzhia crisis shows the risk of running reactors in war. The IAEA calls the situation “deeply concerning,” with repeated grid outages. Restarts must sit within stable, well-regulated contexts.

Nuclear’s role in a 24/7 clean-power economy

Life-extension lowers the carbon per kilowatt-hour, avoids new land take, and reuses grid assets. Deep repositories like Onkalo turn waste from a talking point into an engineered end-state. Meanwhile, AI data centers, electrified heat, and transit raise a simple demand: clean megawatts every hour, and this not just on sunny afternoons.

We should treat nuclear as the backbone and layer wind, solar, storage, and demand flexibility on top. For that we have to lock in long-term PPAs that match power by the hour. Sitse need to be near existing interconnects to move fast. This way we will cut CO₂, hold prices steady, and keep the lights on through calm, dark spells.

That’s basically how you turn pledges into uptime and how you build a net-zero grid that actually works for AI, industry, and households.

So, keep the reactors that already work. Restart the ones that still make sense. Build the next wave where industry and grids need firm, carbon-free power. That is the core message of this comeback.

---

Quick FAQ – The Nuclear Comeback

Does nuclear crowd out renewables?

No. The IEA’s net-zero modeling uses both. Nuclear reduces system costs by limiting storage and grid overbuild for resilience hours.

Isn’t new nuclear too expensive?

Large new builds can overrun. But restarting and life-extending existing units are different projects with faster timelines and lower system cost impact than many think.

What about safety in conflict zones?

Running reactors in war is an unacceptable risk. The Zaporizhzhia situation proves that, which is why restarts must occur only under stable governance and regulator control.

How many reactors has Japan actually restarted?

Post-Fukushima, around 14 have restarted, with more in the approval pipeline, including recent high-profile units.

Are major companies really betting on nuclear?

Yes. Big tech and industrials signed public pledges in 2025 to support tripling capacity by 2050, citing data-center demand and clean-power procurement.

---

---