Competition in the field of artificial intelligence (AI) enters a new stage of "energy efficiency": Former Facebook privacy chief warns of a hidden power grid crisis in the midst of a trillion-dollar infrastructure boom.

Former Facebook privacy affairs chief Chris Kelly said that the next stage of the current development boom in the artificial intelligence (AI) industry will focus on enhancing the efficiency of technology applications. Kelly stated on Monday that top AI companies globally are competing to build infrastructure to support AI computing power needs, and the AI industry must optimize this high-energy-consuming infrastructure construction model. He stated, "The human brain operates on only 20 watts of energy. We can deduce and think without the need for gigawatt-level energy centers to support us. I believe that exploring efficiency potential will be one of the core areas of exploration for top AI companies." He added that companies able to achieve breakthroughs in reducing data center costs will ultimately become winners in the AI field. According to statistics from S&P Global, Inc., by 2025, major hyperscale cloud computing service providers have sparked a global data center construction boom, with the total infrastructure-related transaction volume exceeding $61 billion that year. Only the company OpenAI has planned investments of over $1.4 trillion in the AI industry in the coming years, including a series of heavy collaborations with graphics card giant NVIDIA Corporation, infrastructure industry leader Oracle Corporation, and CoreWeave. However, this booming data center construction wave has raised increasing concernshow will the already stressed power grid system provide sufficient electricity support for these computing power infrastructure projects? NVIDIA Corporation and OpenAI jointly announced a partnership plan in September this year, and the scale of computing power infrastructure involved requires at least 10 gigawatts of power supply, which is approximately equal to the annual total electricity consumption of 8 million American households. Data from the New York Independent System Operator shows that this power consumption of 10 gigawatts is roughly equivalent to the total demand for electricity during the peak summer period in New York City in 2024. After DeepSeek launched a free and open-source large language model in December 2024, the AI industry's concerns and worries about costs have further intensified. DeepSeek stated that the development cost of this model was less than $6 million, much lower than its American competitors. Kelly said he expects to see "several Chinese companies emerge," especially after U.S. President Trump recently approved the sale of NVIDIA Corporation's H200 chips to China. Kelly added that open-source modelsespecially those from Chinawill provide people with "basic-level computing power" and the ability to use generative and agentic AI.