What you’re about to read is is my draft of a guest post I recently wrote for Michael Spencer of AI Supremacy, a leading AI news Substack. It was published there on February 11, 2025.
You may have seen many of the graphs and news items in my previous newsletters. In this article, I’m putting them together to help provide context for the large number of folks who are interested in AI but aren’t yet really aware of the concurrent clean energy revolution, so I’m framing these news items slightly differently here with an eye to a primarily AI-interested audience.
Now, I’m republishing it to my own Substack! Here it is.
AI’s rise is happening in a context where the key global energy trend is the Dawn of the Solar Age
AI is an exciting and transformative technology with potential utility across multitudes of domains, and its impact on human civilization is probably only just beginning. In this context, it’s easy and understandable to assume that for any given field, call it Field X, “AI’s impact on Field X” is going to be the most important thing going on in Field X. A lot of articles discussing AI’s potential impact on energy demand and climate policy seem to carry this implicit assumption.
However when you look at the real-world data, AI’s energy demand surge is relatively tiny compared to the ongoing and exponentially accelerating transformation of humanity’s energy system due to cheap clean electricity generation. This is one of the most important events of the 21st century — and may someday be seen as even more important than AI — but it’s received relatively little coverage due to its comparatively workaday and undramatic nature.
Solar power is now the cheapest electricity source in history, and it’s growing exponentially — kicking off what is already by far the fastest energy build-out in history. Here’s a chart from the recent The Economist special “Sun Machines” that shows just how much solar has massively outperformed essentially every everyone’s expectations.
It’s just plain incredible how far we’ve come in the last few decades. Key “price crossover” points being passed in the last few years have kicked the Solar Age into high gear, and these trends show no sign of stopping.
“In 2015 BloombergNEF estimated that the levelised cost of electricity (LCOE) for solar, on a global basis, was $122 per MWh, almost half as high again as the LCOE for onshore wind, then $83. The LCOE for coal in places without carbon prices at the time was $50-$75. Today both solar and onshore wind are in the low $40s, while coal remains much where it was.”
-Sun Machines
Markets, governments, and homeowners the world over are responding to the economics incentives of the emerging Solar Age. In a world where a cheap zero-emissions no-moving-parts energy-generation technology that you can order by the truckload is readily available, you’d expect to see more and more of it being built around the world. And that’s exactly what we’re seeing.
“Buying and installing solar panels is currently the largest single category of investment in electricity generation…solar power is on track to generate more electricity than all the world’s nuclear power plants in 2026, than its wind turbines in 2027, than its dams in 2028, its gas-fired power plants in 2030 and its coal-fired ones in 2032.”
-Sun Machines
Here’s another chart from the International Energy Agency, showing how their solar forecasts have consistently been revised upwards as solar continues to outperform expectations.
Note that as of 2023, human civilization had already built more solar than what the 2015-era IEA expected would be built by 2050. Think about that for a moment! Ten years ago, some of the top energy experts on the planet thought we’d have a certain amount of solar power generating capacity by 2050, and we got it by 2023.
Most of the new electricity-generating capacity being built everywhere on Earth is now renewable energy — a truly titanic shift that occurred in a few short years while the world was, by and large, distracted by other matters. Intermittency, the much-discussed bugbear of renewable energy sources (“what about when the sun isn’t shining?”) has quietly become a solved problem over the same timescale as grid-scale battery technologies began their own exponential surge. We will simply store excess clean energy in ever-more-capacious batteries and discharge it to the grid as needed, with other solutions like pumped hydro energy storage, long-distance transmission lines, and clean baseload power sources chiming in to help.
Markets like the European Union, where renewable energy has been pursued in earnest since the oil crisis of the 1970s, are leading indicators proving that a modern developed-society standard of living can rapidly transition to clean energy as a primary electricity source.
Notably, renewable energy is providing a strong majority of new demand growth and cutting into incumbent fossil fuels’ market share in both America and China, which at the time of writing are generally considered the world-leading nations in AI research.
Chinese solar and wind deployment has been breaking all records for years in a row (driving much of the global cleantech growth), and it’s still accelerating. As one headline put it, “China is installing the wind and solar equivalent of five large nuclear power stations per week.” China met its renewables installation target for 2030 by the end of July 2024, six and a half years early.
All this clean energy growth has pushed coal down to providing a record-low 53% of China’s electricity in May 2024, down from 60% in May 2023. Solar and wind together provided a record-high 23% of China’s electricity in May 2024, up from 7% in May 2016.
And in America as worldwide, the overwhelming majority of all new electricity-generating capacity being built these days is clean energy, led by solar and batteries.
Solar and batteries together, the exponentially growing workhorses of the renewables revolution, accounted for a whopping 83% of new electricity-generating capacity coming online in America in 2024. Wind accounts for another 10%. Natural gas, the only fossil fuel on the list (the U.S. hasn’t built any new coal plants since 2013), accounts for just 4%. Nuclear, just 2%.
This is a U.S. EIA map of new grid-scale power from all sources, set to come online across America from June 2024 through May 2025. It’s a sea of yellow solar farm icons and green wind farm icons, with a few scattered burnt-orange natural gas icons and zero coal icons.
The underlying economics are so strong that solar energy, at the time of writing, is forecast to continue meeting almost all U.S. demand growth despite strong political headwinds.
Note that this U.S. Energy Information Administration forecast was released in late January 2025, during the second Trump Administration, and well after the 2024 election.
It’s worth discussing at this point that there’s a lot of excitement around new nuclear reactor designs and their potential use as a power source for AI, especially in the U.S. energy market. Nuclear is indeed technologically fascinating, and relative to fossil fuels is a very clean and safe energy source. It would likely have been much better for the world if we had built thousands more nuclear fission reactors in the 1960s and 70s instead of succumbing to misguided activist pressure and building more coal plants. New nuclear reactor designs are an interesting research area that might prove useful at scale someday. But in context, nuclear gets a wildly disproportionate amount of hype given that it’s simply not economically competitive at scale anymore in the modern Solar Age. Nuclear provided just 2% of new U.S. electricity-generating capacity in 2024, and the EIA doesn’t expect that to change much. China, the country building the most nuclear power plants of any nation in the world, is building far more solar and wind. The renowned Astral Codex Ten blog has an excellent summary of the solar vs. nuclear debate in the “Progress Studies” field — check it out in full.
“If these trends continue, solar power could reach $10/megawatt-hour in the next few years, and maybe even $1/megawatt hour a few years after that. This would make it 10-100x cheaper than coal, and end almost all of our energy-related problems….
Despite this being a conference about the future, the pro-nuclear faction seemed comparatively stuck in the past…
The pro-solar faction countered that…Solar has so many other advantages - easier to install, more practical for poor countries, harder for regulators to thwart, less likely to irradiate cities in case of disaster. The nuclear faction is cheating by comparing real solar now to ideal nuclear in twenty years. Compare like-to-like, and solar is obviously better.
I thought solar won: I’ve spent my whole life as an extending-lines-on-graphs fan, and it would be hypocritical to stop now.”
— Scott Alexander on Astral Codex Ten
AI may be simultaneously a major factor in the global energy space in absolute terms (lots of new power demand) and a relatively small factor in relative terms (i.e. compared to everything else going on), because human civilization’s energy use is already undergoing unprecedented and accelerating transformations. We currently have a clean, safe, reliable energy source that’s growing exponentially and transforming the world, and it’s solar backed up by batteries. We’re already seeing skyrocketing electricity demand growth at both local and global scales, and it’s due to mass electrification of sectors like transport and heating through new technologies like EV and heat pumps, coupled with rising electricity access in Asia and Africa. AI is a huge, world-changing story, but it’s also rising in the context of another simultaneous huge, world-changing story — the clean energy revolution.
AI will likely use less energy than expected
The International Energy Agency expects the AI-driven data centers boom to drive only 3% of electricity demand growth to 2030 worldwide. Note that that is the *worldwide* number: in particularly data center-heavy areas like northern Virginia, that percentage will likely be much higher and could have profound local impacts.
"Rising data centre electricity use, linked in part to growing use of AI, is already having some strong local impacts, but the potential implications of AI for energy are broader and include improved systems coordination in the power sector and shorter innovation cycles.
There are more than 11 000 data centres registered worldwide and they are often spatially concentrated, so local effects on electricity markets can be substantial.
However, at a global level, data centres account for a relatively small share of overall electricity demand growth to 2030.”
—World Energy Outlook 2024 - IEA
Renowned data scientist Dr. Hannah Ritchie of Oxford and Our World in Data recently wrote a great Substack commentary about these IEA numbers, and created a publicly available chart of the IEA forecast visualizing how data centers fit in with other fast-growing sources of electricity demand.
Furthermore, there are reasons to think that even the IEA’s relatively low forecast may be an overestimate. Leading energy analyst Michael Thomas (writer of Distilled on Substack) has an excellent article summarizing the structural reasons why AI energy use estimates may overshoot, and how this situation has very similar dynamics to the Internet boom of the 1990s — which was also widely believed to be on the verge of sending America’s energy needs sky-high, but ended up having relatively little effect. (Dr. Ritchie’s article also discusses this parallel, as well as emphasizing that a small global impact on energy demand can also mean a large local impact).
“Long before the current AI power panic, there was another panic about a new technology with an insatiable thirst for electricity: the internet.
In 1999, Peter Huber and Mark Mills wrote an article for Forbes Magazine titled “Dig more coal -- the PCs are coming.” Reading that story today is eery in its similarities to the current moment….
By 2009, a decade after the article was published, data centers powering the internet consumed a couple percent of the country’s electricity—nowhere near the 50% growth forecasted in the article.
So how did internet usage grow by many orders of magnitude while electricity demand remained flat? The short answer is data centers learned to use more with less.”
— Michael Thomas
We’re already seeing very similar dynamics with AI. China’s DeepSeek recently shocked the world with its highly energy-efficient yet high-performing LLM, sending U.S. energy stocks tumbling. It’s likely just a matter of time before some American innovator develops a similarly energy-efficient, high-performing AI model, completely changing the underlying assumptions of all current projections of AI energy use in America.
AI can help spur clean energy build-outs
In addition to efficiency progress, data center growth is already helping drive corporate demand for clean energy build-outs and new innovations. Impacts may be counterintuitive. Trade media source UtilityDive recently a good overview article entitled “The AI paradox: Energy-hungry technology could speed clean energy transition.” Here are just a few recent examples (there are many, many more!) of U.S. AI projects directly stimulating clean energy build-outs by providing a high-paying new demand source:
The titanic new Stargate AI partnership has data centers on the way, and they’re set to be located in clean energy-rich Texas. Unsurprisingly, reports are already coming in that they plan to use solar and batteries, by far the cheapest option, as a power supply. The contradiction between President Trump’s support of Stargate and opposition to solar is striking.
The Virginia state government is working to transform former coal mining land into a clean energy-powered “Data Center Ridge,” with bipartisan support from Republican Governor Glenn Youngkin.
A portfolio of solar farms in Oklahoma with a combined generation capacity of 724 megawatts are in part financially sustained by providing power to Google data centers, while providing substantial benefits to local communities as well.
A developer in Utah recently quadrupled the size of their battery storage project to help meet demands from data centers and AI, providing extra capacity to help meet EV demand at the same time.
In December 2024, Google announced a new partnership set to spend $20 billion to build “gigawatts” of new clean energy and storage colocated with data centers.
Major Midwestern utility Xcel Energy told Canary Media that they plan to “attract 1.3 gigawatts worth of data centers to its territory by 2032” while still meeting the Minnesota state requirement for 100% clean electricity by 2040.
An early geothermal drilling rig at Cape Station, Utah. Get used to seeing emissions-free “clean drilling” as a new contributor to decarbonization! Pioneering startup Fervo Energy connected “Project Red,” the world’s first commercial-scale enhanced geothermal system, to the grid in Nevada in 2023 — to sell its power to nearby Google data centers. Since then, the Bureau of Land Management has already approved Fervo’s much bigger Cape Geothermal Power Project in Beaver County, Utah, with the potential to generate up to 2 GW (2,000 MW), and the company reports that its new wells are already achieving record-breaking flow rates! Enhanced geothermal uses much of the same technology and expertise as fracking, and if it continues to scale up could become a perfect contributor of clean baseload power tailor-made for the American economy. It’s now at commercial scale and growing fast. Data center demand directly helped make that happen.
But even this doesn’t capture the full story of AI’s potential synergies with the ongoing clean energy revolution.
AI consumes energy, but can also help produce it.
It’s fairly intuitive how AI growth can help spur energy innovation indirectly, by providing a new demand source that will pay large sums for reliable baseload power. But several recent examples show that it can also contribute to energy innovation very directly, with AI itself being used to find new resources and technologies to produce, store, and transmit energy. Here are just a few examples from a fast-growing list:
A Bay Area startup called KoBold Metals is using AI analysis of the vast existing stockpile of research about Earth’s crust to discover overlooked deposits of critical battery minerals like lithium, cobalt, copper, and nickel. They’ve already found new lithium deposits in South Korea, Australia, Namibia, Quebec and Nevada, but their crown jewel so far is the discovery of one gigantic “motherlode” of copper in Zambia. KoBold is beginning their first mine at that site in Zambia, and the lode their AI found could soon supply a major chunk of U.S. demand thanks to the Lobito Corridor project.
AI can enhance the efficiency of solar farms in a multitude of ways. Machine learning-enabled predictive maintenance can improve productivity by up to 25%, which synergizes nicely with the fast-growing sector of AI-enabled solar farm inspector drone fleets, while AI-enabled trading can find the ideal timing to maximize the market value of releasing electricity from battery storage. The World Economic Forum recently published an overview article.
Similar dynamics mean that AI can help improve the efficiency of the entire grid — and, as Bloomberg reports, encourage investment in new grid-related tech in general. The U.S. government has funded several “smart grid” research projects (though their political future may be in doubt) and early potential applications include predicting EV charging times, forecasting the risk of blackouts from extreme weather events, and the already-available Gridshare software providing customized household-scale energy use data.
The Lawrence Berkeley National Laboratory is developing an extraordinary automation of the entire process of scientific and technological discovery, with their A-Lab already capable of creating novel useful materials 24-7 with no humans in the loop. At the A-Lab, an AI is directing robots to conduct experiments trying to identify new high-functioning materials for battery electrodes, superconductors, and more. This is still very much early days, with critiques identifying early A-Lab errors already entering scientific literature, but the long-term potential is incredible. Here’s a description of how the autonomous “A-Lab” (pictured) works.
“The AI starts by coming up with a plausible way to synthesize a material, using its understanding of chemistry. It guides robotic arms to select among nearly 200 different powdery starting materials, containing elements such as lithium, nickel, copper, iron, and manganese. After mixing the precursors, another robot parcels out the mix into a set of crucibles, which are loaded into furnaces where they can be mixed with gases such as nitrogen, oxygen, and hydrogen. The AI then determines how long to bake the different mixes, the temperatures, drying times, and so on.
After the baking, a gumball-like dispenser adds a ball bearing to each crucible and shakes it to grind the new substance into a fine powder that’s loaded onto a slide. A robot arm then grabs each sample and slides it into an x-ray machine or other equipment for analysis. Results are fed back into the Materials Project database of materials structures and properties, and if the outcome isn’t what was predicted, the AI setup iterates the reaction conditions and starts anew.”
-Science magazine
And it’s starting to look like the A-Lab was just the beginning, like the Enigma code-breakers and ENIAC heralding the age of personal computers. On January 16, 2025, Microsoft unveiled MatterGen, a pioneering attempt to scale up generative AI for material science discovery and synthesis. They claim that it presents an entirely new state-of-the-art paradigm, explicitly addressing some of the A-Lab issues (like computational disorder) and already having generated the structure of a novel material (TaCr2O6) that was then successfully synthesized in a lab. Microsoft has already released the MatterGen source code and training data to the public under an MIT license.
Even in a highly implausible extreme bear case for AI, in which model capabilities top out at January 2025 levels and no new innovations are made, simply deploying and fine-tuning all the existing potential applications would yield substantial benefits for global clean energy development.
AI will help determine human civilization’s energy future
The future of the energy that powers AI data centers is going to be clean and green, because the future of the energy that powers all of human civilization is eventually going to be clean and green.
But the big questions in serious energy policy today aren’t really about the civilization-wide direction of travel, but about which nations and industries will pull ahead or fall behind — and how that speeds up or slows down the progress of the entire world. Will the second Trump administration hobble American energy abundance, reindustrialization and AI innovation by kneecapping the clean energy sectors that already provide the vast majority of the nation’s new electricity-generating capacity? Will China become a world-dominating “electro-state” if Western nations’ efforts to build up domestic solar, battery, and EV manufacturing efforts are hijacked by culture war polarization? Will India be able to stand up its own cleantech manufacturing industries while providing abundant energy to its citizens? Will the industrial heartlands of Japan and Germany be hollowed out by their star automakers’ failure to compete in battery technology? Will less industrialized nations from South Africa to Pakistan be able to use the flood of Chinese-made solar panels to achieve a decentralized electrification despite weak central governments?
Cumulatively, will these trends net out to eventually reduce greenhouse gas emissions enough to stabilize Earth’s atmosphere? Can we stay below 2°C of warming by the end of the century, or even get back down below 1.5°C? Will the Amazon Rainforest, Great Barrier Reef, and a stable West Antarctic Ice Sheet survive into the 2100s?
The AI field is gaining substantial economic and political influence, and there are many vectors by which it can help decide the answer to these questions. Strategic allocation of AI’s demand surge can help advance national industrialization in fast-growing cleantech instead of doubling down on the dying fossil fuel industry. Clever new uses of AI can help speed the already-accelerating cleantech revolution. As AI progresses, there will likely be many more connections that we haven’t even thought of yet. To paraphrase the apocryphal quote from Napoleonic France, attempts to pair AI development with fossil fuel use are worse than a climate “crime” — they are an economic and geopolitical “blunder.”
Every choice around clean energy policy matters for determining the future of a nation’s economy and security. Every fraction of a degree of warming avoided thanks to clean energy deployment matters immensely for protecting Earth’s communities and ecosystems. AI is an industry of the future, and has an unprecedented opportunity — and responsibility — to help make that future bright.
 | A guest post by
|
Great piece! My biggest takeaway is the risks of overestimating AI's power demand. As we're seeing in the US, utility companies are using that demand as an excuse to delay the retirement of coal assets in some cases. As we've seen with exponential renewable growth, however, if we don't panic, maybe tech companies will see that they have time to take the renewable route. As we're seeing in IL this week, legislators are working on adding 3GW of battery storage in the state to help integrate renewables with the grid.
The energy bottleneck you identify is real, but I'm curious why you assume AI training must happen in concentrated bursts. What if we reverse the paradigm and distribute training across renewable energy peaks, turning every solar panel and wind turbine into a node in a planetary compute fabric? The constraint might not be how much energy we can generate, but how intelligently we can choreograph computation to dance with energy's natural rhythms.