In partnership with

Zhangzhou Unit 2 connected to the grid in November 2025. By January 2026, it was in full commercial service. Two months from grid connection to generating revenue. In the United States, that transition alone takes years of regulatory review. The UAE's Barakah complex brought 5.6 GW online across four units, now powering 25% of national electricity. France has 56 reactors and EDF has identified four industrial sites for datacenter development. The capacity exists globally. The question is why China deploys it faster. The answer reveals something important about the relationship between energy, grid, and compute. The lesson is not that state planning works. It is that reducing the decisions between generation and demand compresses timelines everywhere.

59 reactors, 28 under construction, one design philosophy

China now operates 59 nuclear reactors generating 62 GW. A further 28 reactors are under construction, adding 32 GW. The Hualong One design has entered batch-scale expansion, with multiple units progressing simultaneously across provinces.

The numbers are staggering, but not new. What deserves attention is the timeline compression. Zhangzhou Unit 2 went from grid connection to commercial service in two months. Taipingling Unit 1, a 1,116 MW Hualong One, targets commercial operation in the first half of 2026. The Linglong One SMR at Changjiang targets commissioning in 2026. It aims to become the world's first commercial onshore Small Modular Reactor (SMR). Cold testing, containment integrity, and steam turbine tests are complete.

China's draft 15th Five-Year Plan targets 110 GW of nuclear capacity by 2030. That is nearly double the current fleet in under five years.

Three markets, three regulators, one decade

In the United States, the interconnection queue exceeds 2,600 GW. A nuclear operator, a grid company, and a datacenter developer are three separate entities. Each has its own regulator, capital structure, and timeline. Connecting a reactor to a datacenter means negotiating across all three. Projects stall not because the engineering fails but because the coordination does.

The UK faces similar fragmentation. National Grid ESO manages transmission. A separate planning authority governs land use. The datacenter operator negotiates power purchase agreements independently. Each layer adds years.

Europe's fragmented model creates a specific outcome. Even where nuclear capacity exists, deploying it for compute takes seven to ten years. The constraint is not physics. It is governance architecture.

One decision chain from reactor to rack

China's advantage is structural. The China National Nuclear Corporation (CNNC) controls reactor design, construction, and operation. The State Grid Corporation of China (SGCC) invested £70bn ($89bn, €84bn) in grid infrastructure in 2025. That is CNY 650 billion in a single year. Ultra-high voltage transmission across 38 dedicated lines moves power from generation to demand at continental scale. No Western grid operator invests at this rate.

The integrated energy zone model places nuclear generation, grid connection, and industrial demand, including compute, under coordinated planning authority. CNNC's nuclear sites at Qinshan, Yangjiang, and Tianwan collectively represent gigawatts of baseload capacity positioned within established grid corridors. China's "East Data, West Compute" initiative creates a national architecture that routes computing demand to where power exists.

One entity controls generation. Another controls transmission. A national strategy coordinates demand. Three decisions that take the West a decade happen in China within a single planning cycle. The compression between power generation and operational compute is the real differentiator.

Reactor counts miss the point

Western analysts focus on reactor counts. The actual advantage is vertical integration across energy, grid, and compute within a single decision chain. China does not just build reactors faster. It connects them to demand faster.

Consider the comparison. France has 56 reactors and world-class nuclear expertise. Yet EDF's datacenter site programme is still in early stages, with two additional sites expected by 2026. The capacity exists. The governance model slows deployment.

The UAE offers a contrasting approach. Barakah's 5.6 GW already powers a quarter of national electricity. The government's AI infrastructure strategy targets 5 GW of compute capacity, drawing directly on nuclear baseload. A single sovereign authority coordinates energy, infrastructure, and compute investment. The timeline compresses accordingly.

Culham, Macron, and the governance experiments

The UK's response reveals awareness of the problem. The Culham AI Growth Zone sits at the UK Atomic Energy Authority site near Oxford. It repurposes grid capacity freed by JET fusion reactor decommissioning. Starting at 100 MW and scaling to 500 MW, it bypasses years of grid connection delay. The model works because it uses existing infrastructure. No new transmission lines required. The grid engineer validates the headroom and the datacenter follows.

France's third Multiannual Energy Programme (2026 to 2035) calls for six new EPR2 reactors. President Macron's message at the 2026 World Nuclear Energy Summit was direct. Nuclear gives France the ability to open datacenters and build computing capacity. "At the heart of the artificial intelligence challenge," he said. The ambition is clear. The governance model will determine the speed.

Three principles that compress timelines

Three principles separate fast deployers from slow ones. First, reduce the number of decision-makers between generation and demand. China uses coordinated state planning. The UAE uses sovereign investment authority. Both compress timelines.

Second, use existing grid capacity before building new connections. Culham demonstrates this in the UK. Nuclear sites with grid headroom offer immediate compute deployment. No queue delays. No decade of permitting. The operator and the grid engineer make that assessment together.

Third, treat energy, grid, and compute as a single infrastructure stack. Not three separate markets. Regions that coordinate across all three deploy faster. Those that optimise each layer independently add years to every project.

GTC 2026: the digital twin response

At NVIDIA GTC this week, AtkinsRealis announced a collaboration with NVIDIA to build digital twins of nuclear-powered AI factories using Omniverse before construction begins. Bechtel, Jacobs, Cadence, Siemens, and GE Vernova are part of the same Omniverse DSX Blueprint ecosystem. It is a reference architecture for designing gigawatt-scale AI factories as simulated systems: power, cooling, electrical, and compute modelled together in a single environment.

This is the Western answer to China's integrated planning model. Instead of one state authority compressing the decision chain, a federated digital twin compresses the design-build timeline virtually. Simulate first. Build once. Commission faster.

Separately, Idaho National Laboratory's PROMETHEUS project with NVIDIA aims to cut reactor development times in half using AI-accelerated simulation. The UK Atomic Energy Authority is using Omniverse for fusion digital twins. The momentum is unmistakable.

It is a significant step. But simulation solves the design gap. It does not solve the handover gap. A 100 MW facility still generates 30,000+ documents across 15 to 25 standards during construction. Those documents must be structured, harmonised, and validated before the facility can operate. The digital twin shows you what the building looks like. The standards architecture tells you whether it is compliant. Until both exist in the same pipeline, the 12 to 18 month commissioning delay persists.

The physics is solved. The governance is not.

Where the capital is moving

For infrastructure investors, the signal is unambiguous. China's nuclear fleet will reach 110 GW by 2030. The associated compute infrastructure is being planned alongside it. The £70bn ($89bn, €84bn) annual grid investment creates transmission corridors that serve both industrial and digital demand.

The UAE presents a second governance archetype. Barakah's output feeds directly into national AI strategy. A sovereign authority coordinates energy and compute. For operators in fragmented markets, the Culham model offers the fastest path. Existing grid. Existing site. Existing capacity. The engineer validates the headroom. The datacenter follows.

South Korea adds another dimension. KEPCO's export programme built the UAE's Barakah reactors. Seoul now explores domestic nuclear-datacenter corridors. Japan's reactor restart programme, with 33 operable units, creates further opportunity. Each region's deployment speed maps directly to its governance architecture.

Signatures, not physics, determine speed

China's speed advantage is not bureaucratic efficiency. It is vertical integration across energy, grid, and compute within a single decision chain. CNNC builds the reactor. State Grid connects it. National strategy directs the demand. The gap between generation and operational compute collapses to two years. The West stretches it to ten.

This model has limitations. Centralised planning creates opacity around safety data. Intellectual property transfer remains contested. Environmental review processes differ from international norms. No governance model is without trade-offs.

Yet the engineering observation stands. Every region faces the same physics. The difference is how many signatures stand between a megawatt generated and a megawatt consumed.

Next week: Nuclear carbon credits do not exist. We examine what operators and investors should do instead.

The Tech newsletter for Engineers who want to stay ahead

Tech moves fast, but you're still playing catch-up?

That's exactly why 200K+ engineers working at Google, Meta, and Apple read The Code twice a week.

Here's what you get:

  • Curated tech news that shapes your career - Filtered from thousands of sources so you know what's coming 6 months early.

  • Practical resources you can use immediately - Real tutorials and tools that solve actual engineering problems.

  • Research papers and insights decoded - We break down complex tech so you understand what matters.

All delivered twice a week in just 2 short emails.

The GPU

The GPU

Independent insights on AI infrastructure. Trusted by decision makers from Nvidia, AMD, CoreWeave, Vultr, Macquarie, Brookfield, CBRE, JLL, Bloomberg, DCD, and many more.

Keep Reading