Meta’s energy deal highlights AI’s rising demands

    0
    1

    Meta’s recent partnership with Constellation Energy signifies the Facebook and Instagram parent company’s commitment to a future powered by artificial intelligence. This 20-year agreement aims to breathe new life into an Illinois nuclear power plant, following in the footsteps of similar deals made by other tech giants like Amazon, Google, and Microsoft. Nevertheless, it’s expected to take several years before nuclear energy can fully satisfy the ever-growing power needs of AI technologies.

    The energy demands of AI are immense, largely due to the reliance on fossil fuels, which contribute to climate change. The surprising surge in the popularity of generative AI products has caught many tech firms off guard, challenging their plans to transition to environmentally friendly energy sources. While Meta is looking toward a nuclear future, their short-term strategies still depend heavily on natural gas. Entergy, a major utility provider, is fast-tracking the development of gas-fired power plants in Louisiana to support a new Meta data center complex.

    A question arises: Is the U.S. ready to power AI with nuclear energy? In contrast, France stands out with its significant reliance on nuclear power—accounting for about 75% of the country’s electricity, the highest globally. This has been a central part of its strategy to become a leader in AI. Earlier this year, at an AI summit in Paris, French President Emmanuel Macron emphasized the ease of leveraging nuclear power for AI, compared to the reliance on fossil fuels.

    In the U.S., data centers predominantly depend on fossil fuels like natural gas and sometimes coal, according to a report by the International Energy Agency in April. With AI demand increasing, the primary source of new power is projected to be gas-fired plants, which, while affordable and reliable, also emit greenhouse gases. Currently, renewable energy sources such as solar and wind supply approximately 24% of electricity to U.S. data centers, with nuclear energy contributing around 15%. It will likely take years before cleaner energy sources, including nuclear, can significantly reduce the dependency on fossil fuels.

    A report from the U.S. Department of Energy estimated that the electricity demand from data centers in the U.S. has tripled in the last decade and is expected to double or even triple again by 2028. At that point, data centers could consume as much as 12% of the nation’s electricity.

    Why does artificial intelligence require so much power? Creating AI chatbots and their foundational systems, such as Meta’s Llama, demands significant computational power. This begins with a training or pretraining process, during which AI systems learn from large datasets. This requires specialized computer chips, often graphics processors or GPUs, that can handle numerous simultaneous calculations across a network of interconnected devices.

    Even after the initial training, generative AI tools require electricity to perform tasks, like creating a document or generating an image upon request. This is known as inferencing, wherein a trained AI model processes new data to produce responses based on its learned knowledge. All these computations consume vast amounts of electricity and produce significant heat, necessitating cooling systems such as air conditioning in data centers. These systems can increase electricity consumption, prompting operators to seek alternative cooling techniques, often involving water circulation.