“We need the energy in all forms, renewable, non-renewable, whatever. It needs to be there, and it needs to be there quickly,” testified Eric Schmidt, ex-Google CEO, in a recent U.S. House Committee on Energy and Commerce testimony. His comments were discordant, both for their import but also for the magnitude of the projection that he laid out: the AI industry has the potential to call for 99 percent of global electricity generation.
Schmidt’s testimony created a vision of a future in which the energy demands of AI, and specifically data centers, may be larger than those of entire industries. He projected an additional 29 gigawatts of energy would be needed by 2027, climbing to 67 gigawatts by 2030. To put this into perspective, the International Energy Agency (IEA) estimates that by 2030, global electricity consumption from data centers alone could exceed 945 terawatt-hours, roughly equivalent to the entire electricity usage of Japan today. The driving force? AI’s insatiable appetite for computational power.
The implications of Schmidt’s words stretch far beyond energy usage. His threat that a Chinese AI superintelligence advance would change the dynamic of power globally, in ways that we have no way of understanding or predicting, bodes the geopolitical dimensions. As discussed in a recent analysis of U.S.-China AI competition, competition in AI is as much about acquiring economic and military hegemony as it is about technological progress.
The majority of AI researchers, however, remain unconvinced by Schmidt’s apocalypse. Near-term superintelligence is regarded as ubiquitous speculation. These gloomy predictions are criticized as being used as a political instrument of “corporate capture,” with risk-industry businesses churning out exaggerated expectations of danger as a means of shaping policymakers. Schmidt himself came under fire for investing in AI start-ups as chairman of the National Security Commission on Artificial Intelligence—a legally possible conflict-of-interest move, others thought, that was questionable on moral grounds. What no one can dispute, though, is that AI is putting pressure on energy infrastructure.
The information centers that make up the bulk of the AI operations now account for 1-2 percent of the entire power consumption, which is equivalent to the aviation industry. The MIT researchers have estimated it to be roughly 21 percent in 2030, when transportation of the AI to the consumers is accounted for. Its impact on the environment is something that should be worrying us. A single huge-sized AI model emits as much carbon as taking twenty to five miles in a gas-powered vehicle. AI power requirements are not only an ecological problem, but a geopolitical concern as well. Consumption of essential minerals in data center hardware, for example in processors and in cooling systems, exposes supply chain risk. These minerals tend to be mined from geopolitically unstable areas, again adding to the risk.
Besides all these challenges, AI also has the potential to revolutionize energy efficiency.
The IEA states that AI makes power systems more flexible so as to accept the use of sources like wind and solar power. Through operating optimisation and pre-emptive maintenance via AI, organisations are able to avoid waste and emissions. To cite an instance, AI-based solutions are increasingly being utilised now for public transportation systems and industry energy-consuming process control through better design. All these benefits, however, will only come true if proactive governance is practiced.
Uncontrolled development of AI will worsen the inequalities in the access to energy and environmental sustainability. The optimistic visions of the potential of AI drown out its potential risk, Luxembourg’s energy minister Claude Turmes said. Turmes accused the IEA of not giving practical recommendations to governments on how to regulate and thus minimize the huge negative impact of AI and new mega data centers on the energy system. There are solutions in the pipeline.
Researchers at MIT’s Lincoln Lab have shown that small fixes like “power capping” and adjusting AI model training can lead to significant advances in energy efficiency. By power-supply limiting to processors and employing smaller, less voracious models, they saved huge quantities of energy with no compromise in performance. Clover, which was developed with Northeastern University, also enables AI systems to tap into real-time carbon intensity, reducing emissions by as much as 90 percent in some situations. The way forward demands a fine balancing act. Policymakers will have to steer the conflicting demands of promoting AI innovation, achieving energy security, and reducing environmental footprints. Investments in renewable energy and grid modernization are critical, as is the development of international frameworks to govern AI’s ethical and environmental implications.
As IEA executive director Fatih Birol so rightly put it, “AI is a tool, potentially an incredibly powerful one, but it is up to us – our societies, governments and companies – how we use it.” Whether AI will be a force for good or a squandering of the world’s resources is up to us through choices made today.
It is a dangerous gamble, and time is running out. AI’s energy requirements are not a ghost of things to come but a here-and-now reality. What we do or don’t do will determine not only the future of technology but the future of the world.
