Tech companies charging ahead with artificial intelligence have a problem: AI’s rapid growth is colliding headlong with a finite amount of available energy and computing power.

AI is evolving beyond chatbots and into autonomous AI agents that demand far more computing power and electricity, leaving companies scrambling for more energy. OpenAI recently shuttered its video-generating app Sora, in part because it was gobbling up the company’s computational supplies.

The data centers that make up the backbone of AI technology need loads of energy – both to keep their servers humming and to prevent overheating. But the United States’ electrical grid isn’t exactly up to the task of powering the 21st century economy. It’s old and janky – a loosely connected collection of three grids: East, West and Texas. Experts have long warned that it is outdated and ill-equipped to respond to everything from severe weather to AI’s insatiable new power demand.

“Basically, we have run out of headroom, largely speaking, in the US,” said Ben Hertz-Shargel, an expert on electrification and data centers at energy research firm Wood Mackenzie. “There is a land grab happening, where companies believe that access to more capacity for compute will be necessary to win the future battle over AI services.”

Companies like Microsoft, Amazon, Alphabet, and Meta are ramping up investments in data centers and electricity generation to power new AI models and services, betting that more energy will be needed to keep up with future AI-drive products. OpenAI last year warned the White House of an “electron gap” that puts US leadership in AI at risk, writing that “electrons are the new oil.”

Elon Musk, CEO of xAI, Tesla and SpaceX, said at the World Economic Forum in January that he expects that more chips will be produced than can be turned on because of power constraints.

A Google spokesperson told CNN that the current pace of energy growth is not yet meeting the potential of AI, and that the US needs to go all-in on energy development to keep pace.

Solutions to the power problem exist: increasing renewable energy and battery development, scaling up traditional energy sources like gas and nuclear power, and expanding the current energy grid’s capacity.

But they all face roadblocks, both politically and practically.

The industry is lobbying at both the state and federal level for policies that accelerate permitting and maintain incentives for private sector investment, and the Trump administration recommended relaxing federal rules and expediting construction permits in an AI action plan last year.

But delivering more electric infrastructure is more easily said than done. The most immediate solution is getting more power out of the current electrical grid, experts said.

Both the Biden and Trump administrations threw federal dollars at ideas to modernize and expand the grid, including “re-conductoring,” which replaces existing transmission lines with ones that have greater capacity, allowing more electrical current to travel through them. It’s faster than building a new transmission line entirely, which could take seven to ten years to complete.

Data center companies are looking for spare power capacity at every electrical utility they can find, “including very non-traditional utilities and cooperatives whose doors have never been knocked on before,” Hertz-Shargel said.

One answer could be renewable energy sources like wind and solar, which are often much cheaper and faster to build than the large turbines that are the backbone of the generators in gas-fired power plants. The manufacturers who build these turbines have their order books filled; current wait times for a new turbine are over five years.

But renewables have faced a one-two punch lately, with longer permitting times from the Trump administration combined with Republicans in Congress ending tax credits that would have made it cheaper to build them.

These changes “definitely killed perfectly good wind and solar projects that would have reduced energy prices,” said Andrew Levitt, a consultant focused on wholesale electricity at research firm Brattle. However, Levitt said he expected to see growth in renewable projects going forward because the sheer demand for electricity is so great.

AI companies are pouring money into futuristic technology like fusion power and fueling the big battery industry in an effort to satisfy their hunger for electricity.

OpenAI CEO Sam Altman is backing a $5.4 billion nuclear fusion start-up called Helion Energy, which claims it will supply Microsoft with electricity by 2028. Google, likewise, has signed a strategic partnership with company Commonwealth Fusion Systems, which is racing to build a donut-shaped machine called a tokamak that could eventually generate 10 million times the energy of coal or natural gas, all without greenhouse gas emissions.

Fusion – replicating the power of the sun – has made major advances over the past few years. However, there’s still a big question on whether companies and governments can crack the thorniest problem: getting more energy out than you put into it and building the right machine to sustain that power.

AI companies are also seeking ways to make their own technology more efficient. Google, for example, has been expanding its ability to ramp up or down its operations depending on other users’ demand for power.

Big batteries that store wind and solar energy are another solution, since they can be used to supplement gas turbines. Hertz-Shargel said batteries are becoming a necessity for data centers: The way these facilities consume power can cause issues with electrical currents, which damages gas turbines and electric infrastructure. In these cases, batteries can act as a buffer between the data center and the grid to protect its infrastructure.

Hertz-Shargel said data center demand will be a “huge” boon for the battery storage industry, especially for companies working on future battery technology. The big spending on data centers provides a reliable revenue stream that the “long duration storage industry really needs,” he said.

Despite the hurdles to getting more power, others are more optimistic. Google DeepMind CEO Demis Hassabis told a CNBC podcast earlier this year that he believes AI will end up helping solve its own energy crises.

“I think there are multiple breakthroughs that AI could come up with and help us come up with that would help with the energy situation,” he said.