1. The AI Energy Crisis: When Data Centers Broke the Grid

The IEA estimates global data center electricity consumption reached 415 TWh in 2024 and projects it could approach 945 TWh by 2030. In the United States alone, data center demand may reach 260 TWh, roughly 6 per cent of national electricity consumption. The primary driver of this acceleration is the computational intensity of Large Language Models. A traditional Google search query consumes approximately 0.3 watt-hours of electricity. A single ChatGPT query consumes approximately 2.9 watt-hours, nearly ten times more energy. Reasoning models can consume 10 to 40 watt-hours per exchange. Training GPT-4 required an estimated 1,750 MWh.

The hardware evolution tells the story with clarity. Traditional CPUs operated at 150 to 200 watts. AI GPUs consumed 400 watts in 2022, 700 watts in 2023, and are expected to reach 1,200 watts by 2026. Server racks have grown from 36 kW in 2023 to a projected 50 kW per rack by 2027. A single modern data center can require more than 100 MW, the equivalent of a city of 200,000 inhabitants, and the largest facilities now planned reach 1 to 4 GW per site.

The consequences are already here. In the PJM Interconnection region, capacity prices exploded from $28.92 per megawatt-day for 2024 to 2025 to $329.17 for 2026 to 2027. Electricity costs near large data center developments have risen by as much as 267 per cent compared with 2020 levels. In Amsterdam, plans for 30,000 new homes are at risk because grid capacity is being consumed by data centers. In Europe, electricity demand from data centers in Ireland reached 17 per cent of the country’s total consumption in 2022, and the IEA forecasts 32 per cent by 2026.

2. Policy Responses: Moratoriums, Ratepayer Protection and the BYOP Model

Faced with grid saturation and soaring electricity prices, governments worldwide are taking extraordinary measures. The Netherlands has become the most prominent case study. In January 2024, the Dutch national government banned all new hyperscale data centers across the entire country. In April 2025, Amsterdam announced it would not permit any new data center developments until at least 2030.

The most consequential policy response came from the United States. On 24 February 2026, President Trump announced a Ratepayer Protection Pledge requiring major technology companies to fund their own electricity needs. Amazon, Google, Meta, Microsoft, xAI, Oracle and OpenAI signed the pledge at the White House on 4 March. The Pledge codifies the Bring Your Own Power model. Microsoft pledged to ensure its data center electricity costs are not passed to residential customers. Anthropic stated it would cover 100 per cent of electricity price increases that consumers face from its data centers. Google told utility regulators it pays for 100 per cent of the incremental infrastructure needed to power its facilities.

3. Finding the Right Energy Source

3.1 Solar: cheap but insufficient

Solar photovoltaic energy presents the lowest headline cost, with a 2025 LCOE of $29 to $92 per MWh. However, its capacity factor is only 20 to 30 per cent. Solar does not generate electricity at 3 AM on a cloudy winter night, precisely when a hyperscale data center running AI training workloads still requires 100 per cent of its rated capacity. The land footprint is immense: approximately 60 square miles per 1,000 MW.

3.2 Batteries: not as clean as they seem

Hybrid solar plus storage achieves an average LCOE of $57 per MWh. However, four hours of storage cannot sustain a data center through a 14-hour winter night. The environmental calculus also warrants closer scrutiny: lithium extraction carries significant ecological consequences in Chile, Argentina and the Democratic Republic of Congo; cobalt supply chains remain ethically compromised; and battery disposal creates a hazardous waste stream.

3.3 Natural gas: carbon-intensive and geopolitically fragile

Combined-cycle gas turbines have historically served as the default dispatchable baseload. But in 2025, CCGT LCOE rose 16 per cent to $102 per MWh. Carbon emissions are approximately 350 grams of CO₂ per kWh, inconsistent with hyperscaler net-zero commitments.

Table 1. 2025 Unsubsidised LCOE Comparison by Technology

Source: IEA (2025); IAEA (2024); UNECE (2025). Author compilation, VerdAbility.

3.4 Nuclear: the 24/7 answer

What remains is a technology that operates 24 hours a day, 365 days a year, emits zero carbon during operation, and occupies a fraction of the land required by any alternative. Nuclear delivers a capacity factor of 90 to 93 per cent, roughly 3.5 times the annual energy output of equivalent-rated solar. Its energy density is unmatched: approximately one square mile per 1,000 MW versus 60 for solar and 300 for wind. Uranium fuel represents only 10 to 15 per cent of operating expenses, insulating operators from commodity price volatility. Nuclear plants operate for 40 to 60 plus years, aligning with multi-decade investment horizons.

4. Hyperscalers Go Nuclear

In 2025, Amazon, Microsoft, Google and Meta collectively represented 49 per cent of all corporate power purchase agreements worldwide. Microsoft is financing the restart of Three Mile Island Unit 1, now renamed the Christopher M. Crane Clean Energy Center, through Constellation Energy to secure 837 MW at approximately $1,900 per kW. Amazon has invested $500 million in X-energy. Google’s partnership with Kairos Power pursues molten-salt technology. Meta has signed nuclear agreements exceeding 6 GW.

Table 2. Hyperscaler Nuclear Strategies (2025)

Source: Company announcements; IEA (2025); IAEA (2024).

5. Three Pathways to Scalable Nuclear

5.1 Generation III plus SMRs: the proven path

Generation III plus SMRs leverage proven technology, benefiting from existing licensing frameworks, established supply chains and decades of operational experience. The NuScale VOYGR was the first SMR to receive full design certification from the U.S. NRC. The GE Hitachi BWRX-300 is committed for construction at Darlington, Canada, with operations targeted by 2029. The Rolls-Royce SMR is a 470 MWe modular PWR. The Westinghouse AP300 is derived directly from the operational AP1000 platform. China’s CNNC Linglong One is scheduled to begin commercial operations in the first half of 2026, becoming the world’s first commercial onshore SMR.

5.2 Generation IV SMRs: the next-generation path

Generation IV SMRs use alternative coolants (sodium, helium, molten salt or lead) to achieve operating temperatures of 500°C to 900°C. China’s HTR-PM has been in full commercial operation since late 2023, the world’s first Generation IV reactor at commercial scale. X-energy’s Xe-100 uses TRISO pebble-bed fuel and is the technology behind Amazon’s nuclear strategy. TerraPower’s Natrium, backed by Bill Gates, includes integrated molten-salt energy storage. Kairos Power’s KP-FHR employs molten fluoride salt coolant with TRISO fuel.

5.3 Microreactors and remote settings

Microreactors produce 1 to 5 MWe, small enough to be transported by truck, barge or aircraft. The Westinghouse eVinci is the most advanced example: a 5 MWe heat-pipe cooled reactor, less than 3 metres in diameter, fully factory-assembled. It uses TRISO fuel, has no moving parts during normal operation, requires no water for cooling, and runs for 8 plus years on a single fuel load. Its footprint is less than 2 acres.

6. Case Study: Project Matador

Fermi America, co-founded by former U.S. Energy Secretary Rick Perry, has launched Project Matador in Amarillo, Texas, a $90 billion, 11 GW vertically integrated campus designed to power 18 million square feet of hyperscale AI facilities entirely independently of the public grid. Four Westinghouse AP1000 reactors will provide 4.4 GW of constant nuclear baseload. An additional 4.5 GW of natural-gas generation leverages the site’s strategic location over one of the nation’s largest gas fields.

Project Matador leverages Private Use Network regulations under ERCOT to bypass interconnection queues that can last 5 to 10 years, expecting first gigawatt by late 2026. The Trump administration has issued an executive order to fast-track reactor construction to an 18-month deadline.

7. Scalability to Small Island Developing States

Over 90 per cent of Eastern Caribbean and Pacific island power generation depends on imported fossil fuels, with energy costs often exceeding $0.30 to $0.50 per kWh. International adaptation finance to SIDS represents only 0.2 per cent of global climate flows, and 44 per cent of that is delivered as debt. A nuclear-hybrid microgrid, combining a Generation 3 plus SMR scaled to 50 to 300 MWe with solar and battery storage, delivers efficient, sovereign energy. Nuclear fuel is extraordinarily compact: a single load provides years of generation versus daily fossil-fuel tanker shipments. The European Commission has endorsed SMRs precisely because they allow nuclear capacity to be installed where a large conventional plant would not be viable. In February 2026, USTDA announced funding for the Philippines to evaluate U.S. SMR designs. More than 40 countries now include nuclear energy in their national strategic plans.

8. VerdAbility’s Advisory Framework

8.1 Developed markets

VerdAbility brings the capability to structure the financial architecture connecting data centre developments with institutional capital, through ICMA and LMA-aligned green finance frameworks, tax-credit-optimised capital stacks, and energy-efficiency gains embedded from inception. With power exceeding 50 per cent of operating expenditures, every 0.1 improvement in PUE can translate into approximately USD 1.2 million in annual savings for a 10 MW facility

8.2 SIDS and emerging markets

VerdAbility’s blended-finance experience across IFC, World Bank, KfW, AFD, EBRD, USAID, MCC and FCDO provides the institutional capacity to support the transition of nuclear-hybrid microgrid concepts from strategic feasibility into bankable, implementation-ready project structures.

9. Conclusions: The Hybrid Nuclear Solution

The energy demands of artificial intelligence have moved beyond forecast into present-tense reality. Solar alone cannot provide 24/7 reliability. Battery storage carries environmental externalities. Natural gas ties operators to volatile commodity markets. The foundation of any credible long-term solution for powering data centers must include nuclear energy.

However, nuclear alone is not the answer either. The optimal architecture is a hybrid generation mix: solar PV for daytime peak shaving, onshore wind where geography permits, battery storage for short-duration bridging, and nuclear baseload, whether from restarted conventional plants, Gen 3 plus SMRs like the AP300 and Linglong One, next-generation designs like the Xe-100 and Kairos KP-FHR, or microreactors like the eVinci.

The IAEA projects global nuclear capacity will increase 2.5 times by 2050. The first purpose-built SMR for a data center will mark the inflection point at which strategic intent becomes operational reality. VerdAbility calls on operators, developers, investors and policymakers to accelerate this transition, not only in the world’s deepest capital markets, but in the island nations and remote territories where the need is greatest and the alternatives are fewest.

About the Author

América Hernández Calonge, MBA, SMC

Digital Solutions and Agile Operations

América Hernández Calonge is a Digital Solutions and Agile Operations specialist at VerdAbility, with expertise in AI-driven digital transformation, energy efficiency optimisation, and ESG-aligned implementation frameworks. Within the Green Data Center Series, her research examines hybrid generation architectures combining renewable energy, storage, and nuclear solutions across hyperscale markets and Small Island Developing States.

a.calonge@verdability.com | linkedin.com/in/americahernandezcalonge