Can AI solve the energy problem it created?

It costs just $5 and takes about 15 minutes to train an AI model that can help determine where America's next data centers should be built. This affordable solution is part of the ongoing debate over investing hundreds of billions of dollars in AI, as well as the warning that we need new power plants. Before constructing additional plants, we should first assess how much of the grid we are currently wasting.

Much attention has been given to AI's energy requirements, which is understandable. According to Bloomberg projections, data centers could consume up to one-fifth of all power in the U.S. by 2035, amounting to up to 200 gigawatts. Many of these demands will likely occur in regions already facing power constraints. Power generation wait times on aging grids can exceed five years.

The pressure is immense, and it arrives faster than utilities can plan using current systems and processes. However, what many people overlook is that our existing grid overproduces the energy we need, with much of it sitting idle. Power systems are primarily designed for peak demand seasons, with the rest of the time seeing only half of the capacity being utilized.

This situation presents a planning problem, a task where AI excels. New models can pinpoint where and when the grid has unused power and optimize the costs and timelines for building the infrastructure needed to route it. This "capacity mining" approach allows the AI revolution to progress swiftly while optimizing long-term backbone investments and minimizing backlash.

To achieve this, we should first apply AI models to align data center demand with the headroom already available, identifying locations with available power and operational flexibility. This approach avoids forcing utilities to build new generation and transmission capacity around a site chosen for other reasons. If matching alone isn't sufficient, AI can be used to optimize the expansion of transmission lines and battery systems, routing power to where it is needed or storing it until needed.

New power plants, which are more expensive, slower to build, and potentially polluting, should be a last resort. In most cases, they are avoidable. The outcome is a buildout that moves at the speed of software, not steel and permitting. By getting this order right, utilities can also meet their recent White House pledge to shield consumers from rising electric bills due to data center growth.

Utilizing existing capacity avoids paying for it twice, and identifying this capacity costs relatively little. However, the larger point is that data centers don't have to raise bills at all, and in fact, can lower them. At one large investor-owned utility, large loads, including data centers, can generate around $1 million per megawatt per year in new revenue, equating to about $1 billion per gigawatt.

Structured appropriately, this revenue can offset fixed grid costs shared by all customers. A well-executed buildout would invest in stronger transmission, better monitoring, and improved protection and control systems, benefiting every household on the line, not just the hyperscaler connected to it. Moreover, I believe hyperscalers would be willing to cover more of the cost in exchange for quicker access to power.

In essence, a data center should be viewed as a mini-utility. Its expansion should inject money back into the community it is located in, generating quality local jobs and providing modern, reliable infrastructure for all on that grid. There's no reason for regular people to pay for electricity that data centers consume, and clarifying this point will reduce backlash.

This is a pivotal moment for utilities, which are not historically growth businesses and have experienced stagnation for decades. If they manage data centers effectively, they can achieve the highest revenue growth in a generation while optimizing their grids for all customers. Getting this wrong would result in stranded assets, increased rates, hyperscalers building generation behind-the-meter, a loss of revenue, and missing a significant modernization opportunity.

We don't have to choose between winning the AI race and protecting communities from unnecessary costs. The real question is whether AI can solve the problems AI is creating, and the answer is yes, with tools readily available today.