AI Data Centres Drive Nuclear Revival Across the US and Japan, Uranium ETFs Trail the Buildout
The uranium fund tracking one of 2026's more talked-about investment themes slipped on Monday (2026-07-06). Global X Uranium ETF (URA) fell 0.37% to $43.23 as enthusiasm for nuclear's role in artificial intelligence power supply met the everyday reality of reactor development timelines.6
Wood Mackenzie estimated in May 2026 that Japan's data centres would consume electricity equivalent to 15 million to 18 million households by 2034, driven by a 60% share of the country's total power demand growth. Hyperscalers have committed $28 billion following the Japanese government's selection of Oracle, Google, and Microsoft as official cloud partners.4
The US scale is larger and further advanced. Microsoft signed a 20-year, 835-megawatt power purchase agreement with Constellation Energy in September 2024 to restart Three Mile Island Unit 1, a $1.6 billion project targeting a 2027 startup.6 US power generation from data centres is projected to climb from roughly 5% of total output to about 15% over five years, a step-change for a grid that has barely expanded since 2000.
The US government's long-range ambition adds a further dimension. Federal energy policy as of mid-2026 called for quadrupling nuclear capacity from roughly 100 gigawatts to 400 GW by 2050, a pace that would exceed anything since the original 1970s buildout.2
Investors have positioned across three main vehicles. Global X Uranium ETF (URA), with $6.86 billion in assets, provides the most liquid access to uranium spot prices and major miners including Cameco.6 NUKZ, concentrated in nuclear utilities and enabling technology, returned roughly 52% in the year through late May (2026-05-31), with shares near $71. Uranium & Nuclear ETF (URAN), at roughly $841 million in assets, tilts toward Japanese and South Korean reactor builders at lower cost than larger competitors.6
Japan's energy vulnerability sharpens the regional case. Roughly 90% of Japan's crude imports transit through the Strait of Hormuz, a route that has faced effective closure pressure from the conflict in Iran.3 That exposure, combined with the data centre electricity outlook, has reinforced the argument for domestic reactor restarts. South Korea signalled its own nuclear confidence in late May 2026 (2026-05-26) when Defence Minister Ahn Gyu-back unveiled a roadmap for nuclear-powered submarines — but commercial power applications remain a separate, longer-horizon proposition.7
On the technology side, Idaho National Laboratory partnered with NVIDIA in May 2026 (2026-05-18) on a project called Prometheus that aims to cut reactor development times by up to 50% and reduce operating costs by a similar margin using AI-assisted engineering.5 Still, Three Mile Island refurbishes an existing structure on a relatively accommodating regulatory path and targets 2027. Greenfield projects in most US jurisdictions run a decade or more even under favourable conditions.
The equity market's sensitivity to AI procurement news was illustrated on May 8 (2026-05-08), when Fluence Energy closed at $24.16, up 98.2% in a single week after disclosing supply agreements with two hyperscalers and a record $5.6 billion backlog.1 Shares then fell back, closing down roughly 39% year to date by late May. The whipsaw reflected genuine grid stress but also the difficulty of pricing long-duration power infrastructure in a market trained on quarterly results cycles.
A nuclear reactor running above 90% capacity on a fraction of the land area of equivalent solar generation represents a physical advantage that grid planners in both Washington and Tokyo understand clearly.6 The constraint is capital, permitting, and time. Announced reactor timelines in the US have routinely shifted right, and whether the current commercial pressure from hyperscalers holds commitments to schedule is what separates the investment thesis from the investment outcome.