The rapid growth of Artificial Intelligence is causing a surge in power supply demand. AI giants are switching to sustainable options to match them.
The usage of Artificial Intelligence has skyrocketed in the last two years. With increasing operational demand for high-end AI servers, the grid supply is lagging behind in provisioning adequate electricity to AI companies. As a result, top AI giants are investing in alternate power supply options like renewable energy and AI Nuclear Integration.
The demand for nuclear power for commercial industries has risen in the first quarter of 2024 as AI companies struggle with grid shortages. Multiple incidents of power outages in the last month have raised concerns over the high energy utilization of AI chatbots. A Goldman Sachs report highlights a steep jump of 160% in the power demand for AI applications in overall data center power needs. This simply means that ChatGPT queries consume 10 times the energy required to perform one Google search. Pair this lack of power with the near-peak capacity of existing fossil-fuel plants, and an AI energy crisis is cooking to perfection.
NuScale Power, a nuclear startup, has a potential solution to the high power demand of AI servers. Each of its Small Modular Nuclear Reactors (SMRs) is capable enough to generate 77 megawatts of carbon-free energy continuously. This power supply can detach AI servers entirely from the dependency on the grid supply. The SMRs can be easily transported from the manufacturing plant to the centers through roadways or air transport. However, the Nuclear Regulatory Commission conducts lengthy review processes, and the company expects deliveries to remain pending approval till the end of this decade.
Taking a real on-ground example, one Nvidia H100 Chip, loaded with eight GPUs and two CPUs, is one of the most in-demand AI chatbot servers for data centers. Each one utilizes nearly 10 kilowatts of energy. The older servers used only a fraction of that energy but also lacked the capabilities to perform intelligent operations.
Microreactors and sodium batteries are becoming viable alternatives to reducing grid dependency. Microreactors developed by Radiant Industries can efficiently replace diesel gen-sets as a powerful and carbon-free power backup source. The only thing stopping them from being deployed is the safety concern regarding the disposal of used radioactive fuel rods and other dangerous materials.
Sodium batteries are a lifesaver for data centers using AI GPU chips, as they can withstand a sudden surge of power. These sodium-ion batteries offer a unique solution of rapid charging and disbursement on demand. Unlike traditional lithium-ion batteries, which are known for storing ions for longer periods, sodium batteries come with a short storage timeline. One can think of them as a race car that provides immense power at the cost of a limited fuel storage capacity. These solutions can offer top AI companies a better solution than grid supply, but only if compliance and safety regulations are met.
