Why Cheap Electricity, Not Chips, Will Decide the AI Race

For a century, the global energy mix has been built around fossil fuels: coal for electricity and oil for movement — with a recent LNG boom filling the gaps. The US is riding high on this fossil wave: it set record crude exports in 2024 and remained the world’s largest LNG exporter, even as domestic output hit fresh highs. That’s great for today’s balance of trade. But the AI era introduces a new question into the strategic calculus around energy: how cheaply can you power data centers domestically?

I’ve written recently about the AI factory build-out: the data centers, chip plants, and battery lines that will determine where frontier AI models are trained. The IEA expects data-center electricity demand to more than double by 2030, with AI the main driver. Washington has started to clear a path toward competitiveness in this era: a White House AI Action Plan and a Permitting Technology Action Plan to modernize approvals. These aren’t enough to provide energy at scale, fast.

Solar is now Serious

Here’s what’s changed: solar is viable now and likely to displace fossil fuels globally for many uses. Lazard’s 2025 Levelized Cost of Energy shows utility-scale solar (often paired with storage) as the cheapest new build in much of the US; 2024 also saw solar add more capacity than any technology in two decades. Batteries are racing in behind it — nearly 30 GW will be online by mid-2025, so noon sunshine can be shifted into the evening peaks that stress the grid (and data centers). This is a serious shift in the American and global energy landscape.

Solar is also exciting because it can be decentralized. You can bolt it to rooftops, parking lots, warehouses, and brownfields and then federate it with storage. That modularity shortens timelines and spreads political risk: a thousand 5–50 MW projects are harder to block than a single 1 GW plant. Yes, curtailment (intentional reductions in solar output to reduce grid congestion) is a growing issue in parts of California, but that’s precisely what storage and better transmission solve-and both are scaling.

The US is missing a golden opportunity to capitalize on this revolution.

China is Racing Ahead in Solar

China sees the opening. In 2024 alone, it added roughly 278 GW of solar -about as much as the entire world installed in 2022 -and it’s on track to keep compounding from there. Beijing’s strategy couples this solar surge with a rapid EV transition (nearly half of new cars sold in China 2024 were electric), explicitly to reduce reliance on oil imports (a key Chinese geopolitical weakness). On current trajectories, China could add on the order of a terawatt of solar this decade — comparable to the entire US grid’s capacity — giving it the world’s cheapest on-demand electricity when paired with storage. (China is also adding new coal capacity when suitable — its focus is on cheap power, whatever the source).

US Attention is Elsewhere

Meanwhile, the US has passed a new bill that would unwind clean-energy credits and even tax new solar, wind, storage — and, confusingly, parts of nuclear and geothermal — while tossing subsidies at coal. Analysts estimate this could cancel hundreds of gigawatts of planned capacity and increase power bills, while making the grid less reliable by choking off the very solar-plus-battery combo that helped Texas avoid blackouts. If the US does not correct course, US data centers (as they train the most advanced AI models in the world) will face pricier, dirtier, and less reliable power than is available in China.

If AI is the new space race, then cheap, abundant electricity is the rocket fuel. Coal remains huge globally and will not go away, but it’s no longer the least-cost growth option in the US. Oil will still dominate transport, but it doesn’t run data centers. LNG will remain important, but is becoming more expensive while solar-plus-storage costs keep sliding. The winning policy isn’t anti-fossil. Rather, it’s pro-cheap power.

There’s still time for the US to correct course. To do so, it would need to add back tax credits for solar and storage while domestic manufacturing ramps and economies of scale materialize (like they have in other parts of the world). It would also need to have a laser focus on enabling interconnection and transmission. It would need to incentivize data centers to run the most compute-intensive training during solar-rich hours (and to co-site storage. The IEA’s demand forecasts demonstrate that this would be a grid-reliability play. Additionally, it would need to let fast-track small and community solar-plus-storage with standardized contracts and template environmental reviews. Thousands of small wins beat one mega-project that dies in court.

Overcoming Politics

The politics of solar are difficult in the US. Established energy interests portray solar as unreliable, while conservatives associate it with liberalism and climate activism.

But statesmanship requires navigating politics in the interest of long-term, positive outcomes: lower electricity bills for people and cheap, domestic power for AI data centers, as they train the biggest and best models in the world. Everything else — chips, campuses, and export controls — relies on the availability of cheap power.

Cheap solar is now a key part of the energy mix. Countries ignore it at their own peril.