The power demand brought about by large-scale model training and inference is becoming the most realistic constraint in the global cloud computing industry. In the past two years, multiple tech giants have coincidentally turned their attention to nuclear power, and this transformation is not accidental.
The power consumption characteristics of AI load are different from traditional cloud computing: the training cluster has high power density and long running time, while the inference service requires 7 × 24 hours of uninterrupted response. For computing power providers who pursue stability and predictability, the intermittency of wind and solar power is not friendly, while nuclear power provides zero carbon, stable, and high-capacity factor base load electricity.
Several iconic public actions are worth noting: Microsoft has reached an agreement with Constellation Energy to restart Unit 1 of the Three Mile Island nuclear power plant in Pennsylvania, specifically for powering data centers; Google has signed a power purchase agreement with nuclear power developer Kairos Power for a small modular reactor (SMR), with the first batch of units expected to be put into operation around 2030; Amazon also lays out nuclear power by investing in SMR developers such as X-energy and signing power purchase agreements. These trends indicate that nuclear power is no longer just an issue in the energy industry, but a part of the competition for AI infrastructure.
Of course, nuclear power projects have long cycles and complex regulatory approvals, making it difficult to solve the entire power gap caused by the expansion of AI computing power in the short term. Therefore, technology giants are also simultaneously deploying geothermal, energy storage, and grid side energy efficiency optimization. For industry observers, the truly trackable signal is that electricity costs and availability are shifting from backend cost items to pre variables that determine data center location and even computing power pricing.
It can be foreseen that the intersection of AI and energy will give rise to new industrial opportunities, ranging from SMR supply chain and grid digitization to liquid cooling and energy efficiency management in data centers. For readers who are concerned about the implementation of AI, understanding where electricity comes from is as important as paying attention to model parameters.
[Reference source] Comprehensive compilation of industry information released publicly (Microsoft Constellation Energy Three Mile Island restart agreement, Google Kellogg Power SMR power purchase agreement, Amazon's investment in X-energy, and other company public statements and media reports).