Why Big Tech Is Betting Over 10 GW on Nuclear Power to Feed the AI Data Center Boom

Direct answer: Big tech has collectively committed more than 10 GW of new nuclear capacity to power AI data centers, and 2026 is the year small modular reactors (SMRs) moved from slide-deck promise to physical reality: China's Linglong One is on track to become the world's first land-based commercial SMR, the BWRX-300 became the first Western SMR under construction, and TerraPower secured a US Nuclear Regulatory Commission construction permit for its Natrium plant in Wyoming. It matters right now because AI data centers need firm, round-the-clock power at a scale and reliability that intermittent renewables alone can't guarantee, and nuclear — long stalled by cost overruns and slow regulatory cycles — has suddenly found a buyer willing to pay for exactly what it offers.

The Nuclear Pivot Nobody Expected a Decade Ago

For most of the 2010s and into the early 2020s, nuclear power's dominant storyline in most of the world outside China and France was one of quiet decline: aging plants running past their original design lives, new-build projects running years late and billions over budget, and a public conversation that treated nuclear as a technology from the past rather than a technology with a future. Small modular reactors existed mostly as an engineering concept — a promising one, but one that had been "five years away" for long enough that skepticism was the reasonable default position for anyone who had watched the sector's track record.

2026 is the year that default position became genuinely harder to hold. IDTechEx's own framing — "Data Centers Go Nuclear: Why AI Giants are Investing in SMRs" — captures a shift that would have sounded implausible even three years earlier: the companies driving the shift back toward nuclear aren't utilities or governments primarily, they're technology companies that need enormous, reliable, always-on electricity to run AI training and inference at scale, and have concluded that nuclear is the most credible way to get it without betting their entire power strategy on grid capacity they don't control.

The scale of the collective commitment is the first thing worth sitting with. Big tech has committed to more than 10 GW of new nuclear capacity specifically to feed AI data centers — a number large enough to represent a meaningful share of an entire country's electricity generation capacity, committed by companies whose primary business, a decade ago, had nothing to do with power generation. That's not a research grant or a speculative pilot; it's the kind of capital commitment that only makes sense if the companies making it believe nuclear power is genuinely going to be online, reliable, and contracted to them specifically, within a timeframe that matters to their AI infrastructure roadmaps.

Three specific 2026 milestones are doing most of the work to make that belief look justified rather than aspirational. China's Linglong One, a 125 MWe reactor developed by CNNC, is on track to become the world's first land-based commercial SMR, with commercial operation expected in the first half of 2026 — the first time a purpose-built small modular design has actually reached commercial operation on land, rather than existing only as a demonstration unit or a floating/marine application. The BWRX-300 becoming the first Western SMR under construction is a second, separately significant milestone, because it marks the point at which the US and allied nuclear industry moved from licensing paperwork and engineering studies into an actual, physical build — the step that has historically been where ambitious nuclear projects either prove themselves or start racking up the delays and overruns that have plagued the industry for decades. And TerraPower securing a construction permit from the Nuclear Regulatory Commission for its Natrium plant in Kemmerer, Wyoming, is the clearest US-specific signal that the regulatory system itself — often cited as the single biggest obstacle to nuclear's revival — is actually working through applications and issuing approvals rather than sitting in indefinite review.

None of this means SMRs have "arrived" in the sense of being a mature, widely deployed technology the way solar or wind are today. What it means is that the gap between "promising concept" and "physically under construction or operating" has closed meaningfully in a single year, across multiple countries and multiple reactor designs simultaneously — and that convergence, more than any single project, is what's making 2026 feel like a genuine inflection point rather than another round of optimistic nuclear-revival talk that fails to materialize.

It's worth being precise about why this particular convergence is different from the nuclear-revival narratives that have come and gone roughly once a decade since the 1970s. Previous revival moments were typically driven by a single factor at a time — a climate-policy push, a specific country's energy-security scare, a single company's ambitious but ultimately underfunded pilot program — and tended to fade once that single driving factor lost political or financial momentum. What's happening in 2026 stacks multiple independent drivers simultaneously: a genuinely new category of buyer (AI hyperscalers) with genuinely urgent demand and genuinely deep capital reserves, a federal policy apparatus that has already committed more than $10 billion and built a purpose-designed regulatory pathway rather than merely issuing supportive statements, and multiple countries pursuing parallel but independently motivated nuclear strategies (China's SMR commercialization, the UK's dual-design regulatory review, France's large-reactor recommitment) rather than one country's program depending on the others succeeding. That kind of multi-factor, multi-country convergence is structurally harder to unwind than a single-driver revival wave, which is part of why serious industry analysts are treating 2026 differently than they treated earlier nuclear-optimism cycles.

It's also worth naming who isn't yet fully part of this story, because the gaps are informative in their own right. Grid-scale storage and transmission investment — the pieces of infrastructure that let any new generation source, nuclear included, actually reach the load it's meant to serve — haven't received anywhere near the same volume of 2026 headline attention as the reactor deals themselves, even though a reactor sitting behind an insufficient transmission connection or an under-built local grid can't deliver its promised value regardless of how well the reactor itself performs. Anyone building a long-term power strategy around this trend needs to look past the reactor announcements themselves and ask the less glamorous but equally important question of whether the surrounding grid infrastructure is being built out on a matching timeline.

Why 2026 Is the Inflection Point

The core driver behind this entire story is straightforward once you look at what AI data centers actually need from a power supply, and why that need is different in kind from most electricity demand growth the grid has absorbed in recent decades. AI training and inference workloads run continuously, at enormous and rapidly growing scale, and the companies operating them need power that's available 24 hours a day regardless of weather, time of day, or grid congestion elsewhere. Solar and wind are excellent tools for decarbonizing a grid's overall generation mix, but neither delivers the kind of firm, always-on baseload output that a data center running trillion-parameter model training jobs around the clock actually requires without extensive (and expensive) battery storage or backup generation layered on top.

Nuclear power, whatever its historical cost and schedule problems, has always had exactly the characteristic AI data centers need most: it runs continuously, independent of weather, at a predictable output level, for years at a stretch between refueling. That characteristic didn't change in 2026 — what changed is that a set of buyers with enormous capital reserves, an urgent timeline, and a strong preference for a specific technical property (firm, clean, 24/7 power) finally showed up with the willingness to pay for nuclear's advantages directly, rather than leaving nuclear development to utilities operating under the traditional regulated-rate-of-return model that has made new nuclear construction so financially difficult to justify for decades.

That shift in who's paying, and how, is arguably more important to the 2026 story than any single technical breakthrough in reactor design. Oklo's master agreement with Switch for up to 12 GW, its 500 MW deal with Equinix backed by over $10 billion in investment, and its 1.2 GW project in Ohio all represent a fundamentally different financing model than the one that has struggled to fund new nuclear for decades: a direct corporate buyer, with its own balance sheet and its own urgent demand, contracting for power on terms that make the economics work in a way traditional utility rate-basing often couldn't. Multiply that pattern across the more than 10 GW of collective big-tech nuclear commitments, and you get a nuclear financing environment that looks meaningfully different from anything the industry has had access to in a generation.

The federal policy backdrop in the US has reinforced rather than fought this shift. Federal SMR investment has exceeded $10 billion since 2020, split between more than $4 billion through the Advanced Reactor Demonstration Program and more than $2.7 billion specifically for HALEU enrichment — the specialized nuclear fuel most next-generation SMR designs require and that the existing global supply chain, dominated for decades by Russian enrichment capacity, hasn't been built to supply at the volumes newly ambitious Western SMR programs now need. The NRC's finalization of its new Part 53 licensing framework in March 2026 is the regulatory half of that same story: a purpose-built licensing pathway for advanced reactors, replacing a framework originally designed around traditional large light-water reactors, that's specifically meant to make the review process more predictable and less punishing for genuinely novel SMR designs.

The capital markets have registered this shift clearly and quickly. Publicly traded SMR-adjacent companies — Oklo at roughly $12.9 billion market cap, NuScale at roughly $5.3 billion, Nano Nuclear Energy at roughly $1.2 billion, and Centrus Energy at roughly $3 billion — represent a genuinely new investable category that essentially didn't exist in this form a few years ago, built almost entirely around the thesis that AI-driven power demand makes nuclear's long-standing economic and technical case newly urgent rather than merely theoretically interesting. Holtec and X-energy separately filing IPO paperwork in February and March 2026 extends that same signal: private capital markets are betting that public investors want direct exposure to this specific growth story, not just exposure to it indirectly through the tech companies signing the power-purchase agreements.

There's also a competitive-urgency dimension that's easy to underweight if you only look at the technical and financial pieces of this story. Every hyperscaler racing to build out AI training and inference capacity is competing for the same scarce inputs — advanced chips, cooling capacity, construction timelines, and, increasingly, firm power. A company that locks in a multi-gigawatt nuclear power-purchase agreement today isn't just solving a power problem; it's securing a scarce input ahead of competitors who haven't locked in their own supply yet, in exactly the way companies compete for chip allocation or construction contractor capacity. That competitive dynamic is part of why these commitments have scaled up so quickly and so publicly in 2026 rather than staying quiet, incremental pilot programs — being seen to have secured firm power at scale has become its own kind of competitive signal in the AI infrastructure race.

The Businesses and Stakeholders in the Crosshairs

It's worth pausing to map out just how many distinct groups this single shift touches, because the breadth is part of what separates 2026's nuclear-for-AI story from a narrower industry development that would only matter to utilities and reactor engineers. Reactor developers, hyperscalers, traditional utilities, regulators, capital markets, local communities, and an emerging layer of specialized suppliers (HALEU enrichment, compliance software, grid-interconnection engineering) are all being pulled into this story at once, each with a genuinely different stake in how it plays out. Understanding those distinct positions matters for any business trying to figure out where it fits, or should fit, in this rapidly forming ecosystem.

The most obvious beneficiaries of this shift are the SMR developers themselves, and it's worth being honest that the exposure within that group is highly uneven. Companies with signed, binding power-purchase agreements and active construction permits — Oklo with its Switch and Equinix deals, TerraPower with its NRC-approved Natrium plant — are in a fundamentally different position than companies that are earlier in their development cycle, still working through licensing, engineering validation, or first-of-a-kind construction risk without a signed offtake agreement backing the balance sheet. ExoSwan's own risk framing is directly relevant here: pure-play SMR developers are largely pre- or early-revenue, and "not all will survive," which is a useful reminder that this remains a genuinely high-variance sector even amid the current wave of enthusiasm, closer in risk profile to an early-stage growth technology bet than to a mature utility investment.

Hyperscalers and large data-center operators are the buyers driving this entire story, and their stakes are strategic rather than purely financial. A company like Equinix or Switch signing a multi-hundred-megawatt or multi-gigawatt nuclear deal isn't simply diversifying its energy mix — it's attempting to solve what may be the single largest structural constraint on AI infrastructure growth over the next decade: whether enough firm, clean power actually exists, in the right locations, on the right timeline, to support the scale of compute buildout the industry is planning. A hyperscaler that secures nuclear capacity ahead of competitors gains a form of infrastructure advantage that's much harder for a rival to quickly replicate than, say, ordering more GPUs, because a nuclear plant's development timeline is measured in years, not procurement cycles.

Traditional utilities occupy a more complicated position, straddling both opportunity and disruption. On one hand, the renewed interest and capital flowing into nuclear benefits the broader sector utilities operate in, and some utilities are directly involved as partners or co-investors in specific SMR projects. On the other hand, a data-center operator that signs its own direct power-purchase agreement with an SMR developer is, to some degree, disintermediating the traditional utility relationship — sourcing firm power directly rather than purely through the regulated utility and grid-connection model that has historically been the only game in town for large industrial power buyers. How that tension resolves — utilities as essential grid partners versus utilities as an increasingly optional intermediary for the largest power buyers — is likely to be one of the more consequential structural questions in the power sector over the next several years.

Regulators and policymakers carry a different kind of stake: credibility. The NRC's Part 53 framework, and equivalent regulatory modernization efforts in the UK's Generic Design Assessment process and elsewhere, are being tested in real time by the current wave of applications. A regulatory system that processes these applications with genuine rigor but without the multi-decade delays that have historically plagued nuclear licensing will have demonstrated something genuinely valuable for the industry's long-term credibility; a system that either rubber-stamps applications too quickly (risking a safety failure that could set the entire sector back for a generation) or reverts to historical multi-year bottlenecks will undermine the current wave of investment regardless of how promising the underlying reactor designs are.

Local communities and grid operators near proposed SMR sites carry stakes that are easy to overlook in coverage focused on corporate deals and stock prices. A community hosting a new reactor — whether in Kemmerer, Wyoming, at a UK site like Wylfa, or elsewhere — is weighing genuine local economic benefits (construction jobs, long-term operating employment, tax base) against legitimate questions about safety, water use, and long-term waste handling that don't disappear just because a reactor is smaller and factory-built rather than a traditional large plant. The "walk-away safe" passive-safety design claims common to many SMR designs are a genuine engineering advance over older reactor generations, but they're a claim that has to earn public trust through operating history, not simply through design specification.

Nuclear Power's Global Map

United States. The most active and best-documented market in this story. Oklo's pipeline includes a 1.2 GW project in Ohio, a master agreement with Switch for up to 12 GW, and a 500 MW deal with Equinix backed by more than $10 billion in investment. TerraPower's Natrium plant in Kemmerer, Wyoming secured its NRC construction permit. Federal SMR investment has exceeded $10 billion since 2020 — more than $4 billion through the Advanced Reactor Demonstration Program and more than $2.7 billion for HALEU enrichment. Publicly traded players include Oklo ($12.9 billion market cap), NuScale ($5.3 billion), Nano Nuclear Energy ($1.2 billion), and Centrus Energy ($3 billion), with Holtec and X-energy filing IPO paperwork in February and March 2026. The NRC's new Part 53 licensing framework, finalized in March 2026, is meant to give advanced-reactor applicants a more predictable regulatory pathway than the traditional large-reactor framework.

United Kingdom. Rolls-Royce SMR, a 470 MWe design, is in Step 3 of the UK's Generic Design Assessment process, with Wylfa selected as a site and first concrete targeted as early as 2027; GDA completion was expected around August 2026. Holtec's SMR-300 is separately under UK review at GDA Step 2, giving the UK two distinct advanced-reactor designs moving through formal regulatory review simultaneously.

UAE and Dubai. The signal here is real but thin: Nano Nuclear Energy holds a memorandum of understanding with the UAE for potential Gulf deployment of its microreactor technology. No large-scale UAE SMR program has been confirmed in this research pass, so it's fair to describe this as an early-stage, exploratory relationship rather than a committed build-out — worth watching, not yet a data point to build a forecast around.

Australia. No distinct regional-specific reporting on SMRs for Australia turned up in this research pass. Given Australia's long-standing domestic moratorium debates around nuclear power generally, the absence of specific SMR project reporting here is itself a data point worth noting honestly rather than working around with speculation.

Germany. Similarly, no distinct regional-specific reporting on SMR or broader nuclear revival activity for Germany turned up in this research pass — consistent with Germany's post-2011 policy trajectory away from nuclear power generation more broadly, though this research doesn't confirm that explanation directly, so it's presented here as an honest gap rather than an assumed cause.

Europe and France. France stands apart from the SMR-focused story unfolding elsewhere by pursuing large-scale conventional nuclear as its primary path rather than prioritizing small modular designs. The PPE3 energy programme, published on 12 February 2026, commits to six new EPR2 reactors at Penly, Gravelines, and Bugey, with an option for eight more, targeting 380-420 TWh per year of nuclear output between 2030 and 2035. The construction cost is estimated at €72.8 billion, with roughly 60% covered through a subsidized state loan. EDF's board is expected to reach a final investment decision by the end of 2026, targeting initial commissioning by 2038, and the utility projects 2026 nuclear output of 350-370 TWh following a six-year output high in 2025. France's approach is a genuinely different bet than the SMR-led strategy elsewhere: doubling down on proven large-reactor economics and an existing deep bench of nuclear engineering expertise, rather than betting on the factory-built, smaller-scale model other markets are pursuing.

China. Home to the technology's most advanced commercial milestone globally: Linglong One, a 125 MWe SMR developed by CNNC, is on track to become the world's first land-based commercial SMR, with commercial operation expected in the first half of 2026. China's HTR-PM, a 210 MWe design, has already been operating commercially since 2023, giving China a multi-year head start in actual operating experience with commercial-scale SMR technology that most other countries' programs are still working to match.

What Comes Next for Companies Betting on Nuclear-Powered Compute

The most important reframe for any business trying to understand where this story goes next is recognizing that "SMRs" isn't a single technology moving at a single pace — it's several genuinely different reactor designs, regulatory pathways, and national strategies advancing in parallel, at different speeds, with different risk profiles. China's Linglong One achieving commercial operation is a different kind of milestone than the BWRX-300 breaking ground as the first Western SMR under construction, which is different again from TerraPower's Natrium plant clearing a US regulatory hurdle, which is different again from France's decision to skip the SMR wave entirely in favor of large-scale EPR2 reactors. A business assuming "nuclear is coming" as a single, uniform trend is likely to misjudge both the timeline and the specific opportunities available in its own market and sector.

For hyperscalers and large enterprise power buyers, the practical implication is that securing firm power capacity — whatever the specific technology — has become a genuine competitive input worth treating with the same urgency as chip supply or construction capacity. The companies that moved early on multi-gigawatt nuclear commitments (Oklo's agreements with Switch and Equinix being the clearest examples) are positioning themselves for a power-constrained future that increasingly looks likely to arrive before enough new generation capacity, of any type, comes fully online to meet AI-driven demand growth. Businesses evaluating their own long-term compute and power strategy should treat 2026's nuclear deal-making as a signal about how seriously the largest, best-capitalized players in the industry are taking the power-constraint risk, not as an isolated curiosity confined to a handful of hyperscalers.

For businesses in the broader technology and infrastructure ecosystem — not just the hyperscalers signing the headline deals — there's a genuine opportunity in the tooling, monitoring, and operational software layer this build-out will need. Nuclear plants, even factory-built SMRs designed for simpler operations than traditional large reactors, still require sophisticated monitoring, compliance reporting, and operational-data systems, and the pace at which multiple designs are moving from construction permit to commercial operation simultaneously in 2026 creates real demand for software and systems that can be stood up quickly and reliably. Companies building any part of that operational technology stack — from compliance dashboards to predictive-maintenance systems to the enterprise software layer connecting a plant's output data to a data-center operator's power-management systems — are working in a genuinely growing niche, and teams that need help building that kind of specialized, mission-critical software quickly are exactly the kind of problem our custom software development and AI agents and automation teams are built to help move on.

For businesses evaluating whether to participate in this ecosystem at all — as a vendor, a partner, or simply a company deciding whether nuclear-adjacent infrastructure belongs anywhere in its own long-term planning — the practical starting point is separating genuine near-term signal from longer-horizon speculation. A construction permit already granted, a reactor already under construction, or a signed power-purchase agreement with a named counterparty is a meaningfully different category of signal than a memorandum of understanding, an IPO filing, or a national energy plan targeting a 2038 commissioning date. Both categories matter for understanding where this trend is headed, but treating them with the same confidence level is a mistake worth avoiding when the difference between "under construction" and "under discussion" can mean years of difference in when any given project actually delivers power.

The honest caveat, and the one worth ending on, is that nuclear power's history is full of ambitious announcements that didn't survive contact with construction reality, cost overruns, or regulatory delay — and 2026's wave of announcements, however genuinely more advanced than prior nuclear-revival cycles, hasn't fully escaped that risk. ExoSwan's blunt framing that "not all will survive" among pure-play SMR developers is a useful check against overconfidence: some of today's headline projects will hit delays, cost overruns, or outright cancellation before reaching commercial operation, the way earlier nuclear-revival waves did. What's genuinely different this time is the specificity and financial weight of the demand side — real, signed power-purchase agreements from companies with the capital and the urgent need to see these projects succeed — which gives 2026's wave a more credible foundation than prior nuclear-revival cycles had. Whether that foundation is strong enough to get a meaningful share of these projects across the finish line on anything close to their announced timelines is the question the next two to three years will answer.

Straight Answers on the SMR and Nuclear Data Center Boom

How many small modular reactors are operating in 2026?

The research available for this piece points to two specific commercially operating designs rather than a comprehensive global count: China's HTR-PM (210 MWe) has been operating commercially since 2023, and China's Linglong One (125 MWe) is on track to become the world's first land-based commercial SMR, with commercial operation expected in the first half of 2026. Beyond these two, most other prominent SMR projects referenced in this piece — the BWRX-300, TerraPower's Natrium plant, Rolls-Royce SMR — are at earlier construction or regulatory-review stages rather than already commercially operating, which is consistent with 2026 being described as a genuine inflection point rather than a year in which SMRs became widely operational across multiple countries simultaneously.

What is the total investment in small modular reactors?

The clearest figure available is US-specific: federal SMR investment has exceeded $10 billion since 2020, split between more than $4 billion through the Advanced Reactor Demonstration Program and more than $2.7 billion specifically for HALEU enrichment. On top of that public investment, private capital is flowing in through both direct corporate deals — Oklo's Equinix agreement alone is backed by over $10 billion in investment — and public markets, where Oklo, NuScale, Nano Nuclear Energy, and Centrus Energy carry a combined market capitalization well over $20 billion. A comprehensive global total across all countries and both public and private capital isn't available from this research, but the US-specific figures alone make clear this is now a multi-billion-dollar-a-year commitment rather than a niche research program.

Which SMR companies are publicly traded?

Publicly traded SMR-adjacent companies referenced in 2026 coverage include Oklo (market capitalization of roughly $12.9 billion), NuScale (roughly $5.3 billion), Nano Nuclear Energy (roughly $1.2 billion), and Centrus Energy (roughly $3 billion, primarily a HALEU and nuclear-fuel supplier rather than a reactor developer itself). Holtec and X-energy both filed IPO paperwork in February and March 2026 respectively, meaning they weren't yet publicly traded as of those filings but were moving toward public listings targeting valuations above $10 billion. Investors evaluating this space should note that these companies span meaningfully different risk profiles — from fuel-supply infrastructure like Centrus to pre-revenue reactor developers — and shouldn't be treated as a single homogenous investment category.

What is HALEU and why is it important for SMRs?

HALEU stands for high-assay low-enriched uranium, the specialized nuclear fuel that most next-generation SMR designs require to achieve the compact core sizes and longer refueling intervals that are central to their design advantages over traditional large reactors. It matters because the existing global enrichment supply chain, historically dominated by Russian enrichment capacity, wasn't built to supply HALEU at the volumes the current wave of Western SMR programs now needs, creating a genuine bottleneck risk that sits underneath the more visible reactor-construction milestones. The US federal government's more than $2.7 billion investment in HALEU enrichment specifically reflects how seriously policymakers are treating this fuel-supply constraint as a make-or-break input for the broader SMR rollout, independent of how well any individual reactor design performs technically.

Which big tech companies are buying nuclear power for data centers?

The clearest documented examples in this research involve Oklo, which has a master agreement with Switch for up to 12 GW of capacity and a 500 MW deal with Equinix backed by over $10 billion in investment, alongside a 1.2 GW project in Ohio. These specific deals sit within the broader collective big-tech commitment of more than 10 GW of new nuclear capacity for AI data centers referenced throughout 2026 coverage, indicating that Switch and Equinix are far from the only companies involved even though this research surfaced their deals in the most specific detail. Readers wanting a comprehensive list of every hyperscaler's nuclear commitments should track each company's own public power-purchase agreement announcements directly, since new deals in this space are being announced at a fast pace.

Is investing in SMR stocks like Oklo or NuScale a good idea in 2026?

This is a question this piece can frame honestly but shouldn't answer as personalized investment advice. What the research does support is a genuine risk framing worth weighing: ExoSwan's own assessment characterizes pure-play SMR developers as largely pre- or early-revenue, explicitly noting that "not all will survive" even amid the current wave of enthusiasm and capital commitment. That's a meaningfully different risk profile than an established utility stock, closer to an early-stage growth-technology bet where the underlying demand thesis (AI data centers needing firm nuclear power) is strong, but where individual company execution, construction timelines, and regulatory outcomes remain genuinely uncertain. Anyone considering this sector should treat it with the same due diligence they'd apply to any pre-revenue or early-revenue technology investment, and consult a licensed financial advisor for guidance specific to their own situation.

What is the difference between an SMR and a traditional large nuclear reactor?

Based on the framing used throughout 2026 coverage of this sector, small modular reactors are generally smaller-output, factory-built designs — the ones referenced in this piece range from 125 MWe (Linglong One) to 470 MWe (Rolls-Royce SMR) — intended to be manufactured with more standardized components and assembled faster on-site than traditional large reactors, which are typically built as bespoke, site-specific projects generating well over 1,000 MWe, like the EPR2 reactors France is building under its PPE3 programme. SMRs are also frequently designed around passive, "walk-away safe" safety systems that rely on physics (like gravity-fed cooling) rather than active mechanical or human intervention during an emergency, though the specific safety architecture varies by design and shouldn't be assumed uniform across every SMR on the market.

How much does it cost to build a small modular reactor?

The research available for this piece doesn't include a specific per-unit construction cost figure for any individual SMR design, so providing one here would go beyond what's actually grounded in the source material. What can be stated with confidence by comparison is France's large-reactor cost figure: its six-EPR2 programme is estimated at €72.8 billion total, with roughly 60% covered by a subsidized state loan — useful context for the scale of large conventional nuclear costs, even though it doesn't directly answer the SMR-specific cost question. Readers needing SMR-specific cost figures should consult individual developer disclosures, since per-unit costs vary substantially by design, output size, and site.

Why are tech companies signing nuclear deals instead of just building more solar or wind?

The core reason, as this piece's earlier sections describe, is that AI data centers need continuous, round-the-clock power output, and neither solar nor wind delivers that reliably on its own without substantial (and expensive) battery storage or backup generation layered on top. Nuclear power runs at a predictable, weather-independent output level continuously for years between refueling, which matches the "firm power" requirement of an AI training or inference workload far more directly than intermittent renewable generation does. This doesn't mean tech companies are abandoning renewable investment — most large data-center operators continue substantial solar and wind commitments as part of a broader clean-energy strategy — but for the specific problem of guaranteeing always-on power at data-center scale, nuclear offers a technical property renewables alone don't provide.

What is the HALEU fuel bottleneck and how does it affect SMR rollout?

Centrus Energy's current HALEU output is reported at roughly 900 kilograms per year, a volume that falls well short of the multi-ton annual demand that a fully scaled-out fleet of next-generation SMRs would require to operate. That gap between current enrichment capacity and projected future demand is exactly the bottleneck risk underlying the federal government's more than $2.7 billion investment in HALEU enrichment capacity — a recognition that reactor design and construction progress alone won't matter if there isn't enough qualified fuel available to actually load into the reactors once they're built. Any assessment of how fast the broader SMR rollout can scale needs to weight this fuel-supply constraint alongside the more visible construction and regulatory milestones.

When will the first SMR actually power an AI data center?

Based on the timelines referenced in this piece, China's Linglong One is expected to reach commercial operation in the first half of 2026, and Oklo's deals with Equinix and Switch are structured around 2026-era timelines for at least initial capacity — though this research doesn't specify an exact date for when SMR-generated electricity will first flow directly into an operating AI data center's power supply. Given how new most of these agreements and construction projects still are, readers should treat any specific "first live SMR-powered data center" date with some caution and verify against each individual project's own updated public timeline, since these schedules are exactly the kind of detail most likely to shift as construction and licensing proceed.

Is nuclear power safe to co-locate directly next to a data center?

Modern SMR designs are frequently built around "walk-away safe" passive-safety principles, meaning the reactor is engineered to shut down and cool safely using physics (like gravity-fed coolant flow) rather than requiring active mechanical systems or human intervention during an emergency — a genuine design advance over many older reactor generations. That said, "walk-away safe" is a design claim that needs to earn public trust through actual operating history over time, not simply through engineering specification, and this research doesn't include specific co-location safety studies or regulatory findings for any particular data-center site. Businesses or communities evaluating a specific proposed co-location project should look at that project's own safety case and regulatory review directly, since safety architecture and review rigor vary by design and jurisdiction.

What are the pros and cons of small modular reactors versus large nuclear plants?

Based on the comparative framing in this piece, SMRs generally offer faster, more standardized factory-based construction, smaller upfront capital outlay per unit, and modern passive-safety design, making them attractive for buyers like data-center operators who need power on a faster timeline than a traditional mega-project allows. Large reactors like France's EPR2 design offer greater total output per site and can draw on a much longer track record of large-scale nuclear engineering and operating experience, but come with the higher upfront capital requirements and longer construction timelines that have made large-scale nuclear projects historically difficult to finance and schedule reliably — exactly the trade-off reflected in France choosing the large-reactor path while the US, UK, and China pursue SMR-led strategies in parallel.

Why did China bring the world's first commercial SMR online before the US?

This research doesn't detail the specific institutional or policy reasons behind China's lead with Linglong One, so it would be speculative to name a definitive cause. What can be said with confidence is that China has a longer continuous track record with advanced reactor commercialization generally — its HTR-PM design has been operating commercially since 2023, giving Chinese developers multiple years of real operating experience with commercial-scale advanced reactor technology ahead of most Western SMR programs, which are largely still at the construction-permit or early-construction stage as of 2026.

What is the Rolls-Royce SMR and when will it be built in the UK?

The Rolls-Royce SMR is a 470 MWe small modular reactor design currently in Step 3 of the UK's Generic Design Assessment (GDA) regulatory review process, with Wylfa selected as a site and first concrete targeted as early as 2027. GDA completion itself was expected around August 2026, meaning the design still needs to clear final regulatory approval before construction can proceed on the announced timeline. The UK is reviewing a second advanced-reactor design in parallel — Holtec's SMR-300, currently at GDA Step 2 — giving the UK two distinct SMR programs moving through formal regulatory review at different stages simultaneously.

Why is France building large EPR2 reactors instead of SMRs?

France's PPE3 energy programme, published 12 February 2026, commits specifically to six new large-scale EPR2 reactors at Penly, Gravelines, and Bugey, with an option for eight more, rather than prioritizing small modular designs the way the US, UK, and China have. This research doesn't detail the specific policy rationale behind that choice, but it's consistent with France's deep, decades-long institutional expertise in large-scale conventional nuclear engineering and its existing large-reactor fleet, which gives large-reactor construction a more proven cost and engineering track record domestically than an unproven SMR program would offer at this stage.

How much will France's new nuclear reactors cost and who pays for it?

France's six-EPR2 programme under the PPE3 energy plan is estimated to cost €72.8 billion, with roughly 60% of that financing covered through a subsidized state loan — meaning the French government is directly underwriting the large majority of the project's financing rather than relying primarily on private capital or a pure utility rate-base model. EDF's board is expected to reach a final investment decision by the end of 2026, targeting initial commissioning by 2038, which gives a sense of both the scale of the financial commitment and the long timeline over which it will play out.

What regulatory hurdles do SMR developers face in the US?

The US Nuclear Regulatory Commission finalized a new Part 53 licensing framework in March 2026, specifically designed to give advanced-reactor developers — including SMR designs — a more tailored regulatory pathway than the traditional framework built around large light-water reactors. TerraPower's securing of an NRC construction permit for its Natrium plant in Wyoming is direct evidence that the regulatory system is processing applications and issuing real approvals under the current framework, rather than leaving advanced-reactor applications stuck in indefinite review, though this research doesn't detail every specific procedural hurdle developers still navigate under Part 53.

Are any Middle Eastern countries pursuing small modular reactors?

Nano Nuclear Energy holds a memorandum of understanding with the UAE for potential Gulf deployment of its microreactor technology, which is the clearest Middle East-specific signal in this research. No large-scale, committed UAE SMR construction program was confirmed in this research pass, so it's accurate to describe this as an early-stage, exploratory relationship rather than a committed build-out timeline — a signal worth watching for further development rather than a confirmed regional trend yet.

How many gigawatts of nuclear capacity have tech companies committed to overall?

Big tech has collectively committed to more than 10 GW of new nuclear capacity specifically to feed AI data centers, based on 2026 reporting. Individual deals contributing to that total include Oklo's master agreement with Switch for up to 12 GW (which on its own approaches or exceeds the collective figure, suggesting the 10 GW+ figure may reflect a different measurement point in time or a more conservative accounting than the sum of all announced deals), its 500 MW deal with Equinix, and its 1.2 GW Ohio project. Readers should treat the collective 10 GW+ figure as a meaningful order-of-magnitude signal of the scale of this shift rather than a precisely reconciled sum of every individual announced deal.

What happened to Holtec and X-energy's IPO plans in 2026?

Holtec and X-energy both filed IPO paperwork in February and March 2026 respectively, targeting valuations above $10 billion each. This research doesn't include the outcome of those filings — whether the IPOs completed, at what final valuation, or on what timeline — so it would be inaccurate to state a completed outcome here. Readers tracking the current status of either IPO should consult recent financial news directly, since IPO timelines and outcomes are exactly the kind of detail that can shift quickly after an initial filing.

Is nuclear energy considered 'clean energy' on the same footing as solar and wind?

Nuclear power is widely framed in 2026 coverage of the AI data-center power story as "clean baseload power," emphasizing its lack of direct carbon emissions during operation and its ability to provide continuous output the way solar and wind, as intermittent sources, cannot on their own. Whether nuclear is classified as "clean energy" under specific policy or investment frameworks varies by jurisdiction and program — some clean-energy tax credit and investment frameworks include nuclear explicitly, others don't — and this research doesn't detail every specific regulatory classification, so businesses relying on a specific clean-energy classification for tax, reporting, or investment purposes should verify the applicable definition in their specific jurisdiction and program directly.

How long does it take to get regulatory approval for a new reactor design?

Timelines vary meaningfully by country and design based on the examples in this research. In the UK, Rolls-Royce's SMR design reached Step 3 of the Generic Design Assessment process with GDA completion expected around August 2026, while Holtec's separate SMR-300 design is only at GDA Step 2 — suggesting a multi-year, multi-step process even for designs moving relatively efficiently through review. In the US, the NRC's newly finalized Part 53 framework (March 2026) is specifically intended to make advanced-reactor review more predictable than the traditional large-reactor framework, and TerraPower's securing of a construction permit for its Natrium plant demonstrates that approvals are being granted under the current system, though this research doesn't specify a standard total timeline from initial application to final approval.

Which SMR designs are closest to commercial deployment in 2026?

Based on the milestones described throughout this piece, China's Linglong One is the furthest along, on track to become the world's first land-based commercial SMR with operation expected in the first half of 2026, following China's HTR-PM, which has already been commercially operating since 2023. The BWRX-300 becoming the first Western SMR under construction and TerraPower's Natrium plant securing its NRC construction permit both represent meaningful progress in the US and allied nuclear industry, but both are still at the construction rather than commercial-operation stage, putting them a step behind China's most advanced projects on the specific dimension of reaching commercial operation.

What role does Constellation Energy play in the SMR/data center nuclear trend?

Constellation Energy is referenced in 2026 coverage among the broader set of publicly traded companies adjacent to the SMR and nuclear-for-AI-data-centers trend, reflecting the reality that established nuclear utilities — not just pure-play SMR developers — are positioned to benefit from renewed interest in nuclear power as a solution for AI data-center demand. This research doesn't detail Constellation's specific project agreements or deal terms, so businesses wanting Constellation-specific details on power-purchase agreements or capacity commitments should consult the company's own public disclosures directly.

Are utilities or tech companies footing the bill for new nuclear plants?

Based on the deal structures referenced in this research, it's increasingly tech companies directly footing much of the bill for new nuclear capacity aimed at AI data centers — Oklo's Equinix deal alone is backed by over $10 billion in investment, and its Switch master agreement and Ohio project reflect the same direct-buyer financing pattern. That's a meaningfully different model than the traditional utility rate-base approach, where a regulated utility finances a plant and recovers costs from ratepayers over time. France's large-reactor approach shows the traditional model still very much in use elsewhere: its PPE3 programme is financed roughly 60% through a subsidized state loan, with EDF (a state-linked utility) as the direct builder and operator — a reminder that both financing models are operating in parallel globally, just applied differently depending on the market and the specific buyer's needs.

Building the Operational Layer Behind the Nuclear-AI Buildout

Every one of these projects — from a data-center operator managing a new power-purchase agreement to a reactor developer standing up compliance and monitoring systems for a first-of-a-kind build — needs software that can be built quickly, reliably, and to a genuinely high standard, because mistakes in this kind of mission-critical infrastructure carry real consequences. If your organization is building any part of the operational technology stack around this shift, our custom software development and AI agents and automation teams work directly with companies operating in exactly this kind of high-stakes, fast-moving infrastructure environment.