Countries In The AI Race Are Realizing They Are Also In An Energy Race

From Washington to Beijing, leaders are pouring billions into AI research, model development, and chip manufacturing. But these efforts rest on a fundamental and increasingly scarce input: electricity.

AI uses an incredible amount of power, and even including “please” and “thank you” in replies to ChatGPT costs “millions of dollars” in energy costs, according to Sam Altman, CEO of OpenAI. Stakeholders across the space are steadily realizing that abundant, low-cost, and reliable energy may become the main constraint in AI growth.

When talking about the amount of energy needed to power data centers, Sam Altman said, “There’s no way to get there without an (energy) breakthrough,” he said. “It motivates us to go invest more in fusion.”

The Electric Power Research Institute recently increased its data center power consumption estimates after incorporating more AI growth, saying data centers could consume over 9% of US power in a few years. Meanwhile, the International Energy Agency (IEA) estimates that global electricity consumption from data centers, AI, and cryptocurrencies could double by 2026, exceeding the electricity use of Japan. Third-party forecasts such as this are steadily climbing, as many forecasters aren’t used to demand growing at this rate.

The power supply chain is also not prepared for this level of growth, and there are already constraints with a steady climb in prices for critical inputs such as natural gas and over 100-week order times for relevant equipment.

Countries trying to be leaders in AI are competing in what is arguably the greatest technological race in the last 100 years, as it will inform everything from GDP to defense capabilities. These countries are now adopting an ‘all of the above’ energy strategy to compete, removing regulations and incentivizing energy supply. This is because these countries now view the energy and AI race as existential for the quality of life of their citizens, and this is spilling into support for energy to a level not seen in decades.

This shift has made energy security a front-line issue in the AI race. Just as Cold War powers once scrambled for uranium, today’s geopolitical scramble revolves around electricity, transmission capacity, and stable grid operations needed to produce it.

In the United States, Virginia, Georgia, Ohio, and other states are seeing fast-tracked transmission upgrades and subsidies for natural gas plants to ensure power reliability. Natural gas has been the recent winner as jurisdictions require reliable power above all else, given that a blackout could negate all the AI model training up to that point in time.

China has taken this one step further and continues to accelerate coal plant construction to avoid blackouts that could derail compute-heavy training clusters. The government there has prioritized power reliability over emissions concerns.

Jurisdictions with an unsupportive energy policy risk being left behind. In Germany, where energy prices soared after Russia invaded Ukraine, tech firms are still hesitant to locate new data centers. In response, Berlin has started offering incentives for data centers to co-locate with industrial-scale battery storage and power plants. Norway and Finland, meanwhile, are attracting data center developers with promises of low-cost hydropower. Without cheap and reliable power, countries risk becoming entirely dependent on someone else’s infrastructure, a significant strategic risk.

Canada is a recent example of a country taking a renewed focus on both AI and energy, where the country appointed its first-ever Federal AI minister, with an aim to be an AI leader. Increased support for energy development has gone hand in hand. The current Liberal government, historically not a cheerleader for large-scale energy projects in the country, has also changed its tune, referencing a new all-of-the-above energy policy like China and the US. “Canada has a tremendous opportunity to be the world’s leading energy superpower, in both clean and conventional energy.” It remains to be seen if this promise will play out, but it likely must if the country wants to be a leader in attracting AI investment.

Jurisdictions, such as Alberta, where most of the energy production takes place in Canada, aren’t waiting around, with dozens of data center projects announced and in queue. But even in Alberta, a province that has favored development, the legacy policies are still likely not aggressive enough to meet the moment. The Alberta Electric System Operator (AESO) has limited total load connections and proposed allocating the load across projects. This pro-rata approach means the larger projects, typically with the most reliable backers, and with the scale that AI counterparties want, likely move on to other jurisdictions. Multiple gigawatt-scale projects, without caps, are what competing jurisdictions are trying to attract.

Countries and jurisdictions have rarely had to make policy that will inform everything from regional quality of life to country-level strategic advantages, this quickly. Expect to see policy get looser and looser as countries realize they are trying to capture the largest investment wave in the past century, and also determine whether they will rely on other jurisdictions for the most important tech breakthrough in our lifetimes.