Key highlights
- US AI data centre investment: The US leads in AI data centre investment, with President Trump’s Stargate Project aiming to fund AI infrastructure by 2028, making the US a critical hub for global energy transition risks
- Nuclear power limitations: Nuclear power, despite renewed interest, faces challenges with no new reactor construction underway, and only limited small modular reactor (SMR) deployment expected by 2030
- Solar and battery storage: Solar energy and battery storage are favoured by hyperscalers for AI data centres due to cost efficiency and quick deployment, though solar’s intermittency, potential tariffs and grid infrastructure pose challenges
US AI data centres: a power conundrum by 2030
The US is the undisputed global leader in AI data centre investment. President Trump’s Stargate Project announcement, a USD500bn initiative to fund AI infrastructure by 2028, is the latest in a series of large investments in AI and data centres by the US tech industry.
As the largest AI market, the US is not only a hub for innovation but a testing ground for global energy transition risks. A setback in the US therefore would have far-reaching implications, signaling that even one of the most advanced markets in the world could struggle to reconcile its AI ambitions with infrastructure reality.
The latest open-source large-language model developed by Chinese startup DeepSeek adds a new layer of complexity to the AI landscape. The DeepSeek R1 model suggests that AI technology may be less computationally intensive to develop than previously thought.
While the latest development may seem bearish for AI power demand, it could accelerate the pace of AI and could even lead to faster deployment, a process we have seen with other technologies.
Nuclear power: no panacea for 2030
According to the International Atomic Energy Agency, nuclear power currently accounts for about 18.5% of US electricity production. As of end-2024, there were 94 reactors in operation with a combined capacity of 97GW. Capacity of about 20GW (41 reactors) has been permanently shut down. Despite renewed interest in nuclear energy to meet growing AI-related power demand, particularly small modular reactors (SMRs), concrete action has yet to occur. Notably, no reactor construction is taking place in the US.
SMRs can be built in two to three years, significantly faster than large reactors, which take a minimum of five to eight years. The World Nuclear Association has tracked 20 SMR projects proposed in the US as of January 2025. However, only 1.5GW is expected to be deployed before 2030.
What are small modular reactors (SMRs)?
Small Modular Reactors (SMRs) represent an innovative approach to nuclear power generation, offering a more flexible and scalable solution compared to traditional large-scale reactors. These compact nuclear reactors are designed to address the growing power demands of US AI data centres while providing enhanced safety and environmental benefits. Here’s a breakdown of their key features:
- Compact design: SMRs are smaller, more compact nuclear reactors compared to traditional large-scale reactors
- Scalability: Designed to be built quickly and efficiently, SMRs can be added incrementally to meet growing power demands
- Flexibility: SMRs can be deployed in various locations, including remote areas, providing a reliable power source where traditional grids may not reach
- Safety features: Incorporate advanced safety features, such as passive cooling systems, which reduce the risk of meltdowns and enhance overall safety
- Lower capital costs: Due to their smaller size and modular construction, SMRs require lower upfront capital investment compared to large nuclear plants
- Reduced construction time: The modular design allows for faster construction and deployment, addressing the urgency of meeting power demands for US AI data centres
- Environmental benefits: Produce low-carbon energy, contributing to the reduction of greenhouse gas emissions and supporting the transition to cleaner energy sources
Solar and battery storage: silver linings for hyperscaler growth
Solar energy has seen remarkable expansion across the US in recent years. New utility-scale solar capacity rose by 32GW last year, according to an analysis by research organisation Cleanview. Texas, known for its oil and gas industry, led the US in solar installations, with annual installed capacity more than doubling y/y and totalling 8.9GW.
In addition to SMRs, hyperscalers have increasingly favored solar energy and battery storage to power AI data centres. Solar seems like a rational choice from a cost perspective, the team considers. Additionally, lithium-ion battery prices have fallen 20% y/y in 2024, which should further enhance energy storage project economics going forward. Solar projects are also easier and faster to site and build, which aligns better with hyperscalers’ urgent timelines for data centre deployment. However, solar’s inherent intermittency requires it to be coupled with either a baseload power generation source that can run when solar doesn’t, or a battery storage solution that fills the gap. The analysts think tariffs, particularly aimed at southeast Asian countries, present a risk that could push solar panel prices higher, making them less competitive.
Power grids: vulnerable to the AI power shock wave
US generation capacity buildout is set to remain robust over the next five years, boosted by strong renewable buildout. However, Markets 360 analysts note that the US grid is facing significant integration challenges, with nearly 1,570GW of generator capacity and 1,030GW of storage capacity queued for interconnection as of the end of 2023, according to an analysis by the Lawrence Berkeley National Laboratory.
Building grid infrastructure in the US can take more than a decade. While construction itself is relatively efficient, typically taking 2-4 years, the planning and permitting process can be a significant obstacle. In the analysts’ view, while President Trump’s order to fast-track federal permitting could accelerate the build-out of transmission grids, new policies may cause further confusion amid the changes in the federal workforce. Tariff hikes may also make grid investment in the US even more challenging.
Baseload and transmission capacity constrains demand growth
“In our view, an increase in AI data centre demand on the grid requires increases in baseload (thermal) capacity and transmission capability between regions so that the margin remains sufficient to absorb peak load demand. Given that upgrading and reinforcing the existing power infrastructure is unlikely to keep pace with the unprecedented AI-driven power demand in the near term, we think microgrids could offer a viable solution,” commented Jason Ying, Commodities Desk Strategist at BNP Paribas.
“We see demand-side flexibility as another way to mitigate the surging power demand and optimise energy use. Data centres could shift data processing workloads to times or regions where the grid is less constrained or renewable energy sources are more abundant. This approach could also maximise the utilisation of excess renewable energy that would otherwise be curtailed,” added Jinyi Yue, Sustainability Research Analyst at BNP Paribas Markets 360.
How high could it grow?
In Markets 360’s view, thermal capacity, the need for new nuclear and renewable capacity, and transfer capability constraints will limit AI power demand growth, exacerbated by long lead times and policy uncertainty. However, they think microgrids may offer a solution that could satisfy rapidly increasing demand growth, leveraging on SMRs and renewable technologies, while bypassing issues related to grid interconnection. The team expects different demand scenarios to play out depending mainly on infrastructure, technologies and project lead times.