The comparison with the energy industry, celebrated for its immense investment outlays, is apt not only because of the amounts involved. Artificial intelligence requires enormous computing power. And this computing power requires huge amounts of electricity. On May 2, Bob Blue, chief executive of Dominion Energy, one of the largest U.S. utilities, said data center developers now regularly ask him for “several gigawatts” (GW). Dominion’s total installed capacity is 34 GW.
See the full image
JPMorgan Chase calculates that Microsoft, Amazon’s cloud division (AWS), Alphabet, Meta and Microsoft consumed 90 terawatt hours (TWh) of electricity in 2022, or as much as Colombia. And that was mostly before ChatGPT launched its AI revolution in November this year. The resulting confusion led the International Energy Agency (IEA), the official forecaster, to predict that data centers (including those dedicated to artificial intelligence and equally energy-intensive cryptocurrencies) will consume more than 800 TWh globally in 2026, more than double more in 2022 (see chart). Consulting firm BCG believes data processing could triple its share of US energy consumption by 2030, to 7.5%.
And not just any power will be enough. Technologists want their devices to be tidy. In April, their industry association warned Georgia Power, which managed to accelerate the approval of 1.4 GW of next-generation fossil-fuel powered generation, pointing to growing demand from data centers, that its members would build fewer such facilities in the South American state if the utility issued extra carbon. Combined with the growing demand from increasingly electrified transport, heating and parts of hefty industry, the energy demand of digital technologies places a huge burden on companies that generate and distribute electricity.
BloombergNEF, a news firm, estimates that annual grid investments needed to fully decarbonize global electricity by 2050 will need to augment from about $300 billion in 2022 to $600 billion in 2030 and well over $800 billion in 2050. Risk-averse companies that would normally undertake network expansion projects under the watchful eye of cost-conscious regulators have neither the money nor the appetite to do so.
Enter substantial tech itself. Wealthy giants were already the biggest force behind green “power purchase agreements” that helped usher in America’s renewable energy boom by persuading utilities and other investors to build wind and solar farms. Now they are getting involved in the green energy campaign more directly.
On May 1, Microsoft and Brookfield, one of the largest infrastructure investors in the world, announced an agreement to build 10.5 GW of renewable energy capacity in America and Europe by 2030. The agreement is intended to enable the software giant to fulfill its commitment to own 100% of the shares in By 2030, 100% of electricity will be consumed by sources with zero carbon dioxide emissions. Microsoft and Brookfield did not disclose pricing, but adding a gigawatt of wind or solar capacity could cost about $1 billion.
One problem is that data centers tend to consume power at a constant rate, even when the sun isn’t shining or the wind isn’t blowing. Therefore, technology companies are also considering ways to make data processing more adaptable. In March, Sidewalk Infrastructure Partners, an infrastructure technology company co-founded by Alphabet, unveiled a detailed plan to achieve this goal. It involves a combination of microgrids (which can operate independently but also exchange power with others nearby), batteries and advanced software to enable less time-sensitive tasks, such as training artificial intelligence models, to be moved to fallow periods. Jonathan Winer, one of Sidewalk’s founders, expects these types of data centers to appear first in places with narrow energy employ, such as Arizona, California and Massachusetts.
Renewable energy sources are not the only area of interest for substantial technologies. In March, AWS paid $650 million for a 960-MW data center in Pennsylvania, powered by a nuclear reactor next door. Microsoft has struck a deal with Constellation Energy, the largest U.S. nuclear operator, to supply nuclear power to its data center in Virginia as a backup in the event of a lack of wind and solar power. The two companies are also looking at “diminutive modular reactors,” a promising but unproven nuclear technology.
Meanwhile, Google is looking into geothermal energy. The mining giant has signed its first-ever corporate deal to develop “enhanced” geothermal energy with Fervo, a startup that has raised $430 million in venture capital. Inspired by the shale industry, hot-rocks hotshot drilled horizontal drilling monitored by fiber-optic cables. Its Nevada facility produces 24/7, emission-free power for a local grid that is then purchased by Google. Tim Latimer, head of Fervo, says every drilling rig operated by his company can augment its capacity by 100 MW. The company is building a 400-MW commercial power plant in Utah that will begin feeding the grid in 2026. The Department of Energy believes innovations like Fervo’s could augment America’s geothermal energy production by about 20-fold, to more than 90 GW, by 2050 r.
Google and Microsoft have also partnered with Nucor, a giant American operator of mini-steel mills that employ a lot of electricity. In March, the trio announced that they would aggregate demand and jointly offer contracts for tidy energy projects, both early-stage commercial and completely groundbreaking, “first-of-a-kind” ventures. The idea is to guarantee developers promising technologies such as long-term energy storage, tidy hydrogen, geothermal energy and next-generation nuclear energy.
The most exotic power games in the AI industry are made possible by courtesy Sam Altman, the techno-optimistic head of OpenAI, creator of ChatGPT and Microsoft’s main model building partner. Looking to power the AI revolution, it has invested in Helion, a nuclear fusion startup, and Exowatt, a startup developing solar modules that can act as both electricity generators and thermal batteries. Altman now wants to raise $500 million for Oklo, which is working on micro nuclear reactors powered by spent fuel from larger ones that could power individual factories, corporate campuses and, of course, AI server farms. These bets may seem fancy. On the other hand, 18 months ago, the idea emerged that artificial intelligence could write essays or paint like a human.
To stay up to date with the most significant stories in business and technology, sign up to Bottom lineour weekly newsletter available only to subscribers.
© 2024, The Economist Newspaper Constrained. All rights reserved. From The Economist, published under license. Original content can be found at www.economist.com
