For much of 2024, we have heard about how the massive demands for electricity from AI and Data Centers was going to upend plans for an energy transition, and mandate massive new spends on fossil gas in particular, but also modular nuclear.
Those of us with a long view of the energy demand roller coaster have expressed some caution on the predictions.
If the new breathless predictions were wrong, well, this wouldn’t be the first time. Electricity projections in the 60s and early 70s were giving gimongous numbers for demand later in the century, which proved to be grossly in error, as energy efficiency, sparked by the oil embargoes of the 70s, took off.
It now seems possible that new technology is again about to make innovation much more efficient.
Mark Zuckerberg’s Meta assembled four war rooms of engineers to determine how a Chinese hedge fund managed to release an AI game-changer that may already rival its own technology, The Information reported.
DeepSeek, an AI startup backed by hedge fund High-Flyer Capital Management, this month released a version of its AI chatbot, R1, that it says can perform just as well as competing models such as ChatGPT at a fraction of the cost.
- Shares of some independent power producers fell sharply Monday amid a broader selloff in technology and AI infrastructure stocks. Shares in companies with significant nuclear and gas generation fleets in unregulated markets were particularly hard-hit, with Vistra Corp. falling by more than 28%, Talen Energy more than 21% and Constellation Energy more than 21%.
- The rout came days after Chinese AI startup DeepSeek released two high-performing AI models that may have cost 45 times less to train than leading-edge products from U.S. companies like OpenAI and Anthropic, technology investor and entrepreneur Jeffrey Emanuel said in a Saturday blog post. By Monday, DeepSeek’s AI assistant had surpassed OpenAI’s ChatGPT as Apple’s most-downloaded free app.
- DeepSeek’s success “calls into question the significant electric demand projections for the U.S. [as] AI represents ~75% of overall U.S. demand forecasts through 2030-35 in most projections,” investment bank Jefferies’ power and utilities research team said in a Monday note.
Advanced nuclear technology companies Oklo and NuScale have also notched impressive gains over the past year, with Oklo more than doubling in value since its May 2024 IPO and NuScale gaining 580% since January 2024. Shares of both companies were down more than 20% on Monday.
In December, Oklo announced a 20-year deal with data center developer Switch for up to 12 GW of power at an undisclosed price. Constellation in September announced plans to reopen the undamaged, prematurely retired first unit at the Three Mile Island nuclear power plant on the back of a 20-year Microsoft power purchase agreement that reportedly places a significant premium on the 835-MW facility’s output.
And early last year, Amazon Web Services purchased an up to 960-MW data center campus from Talen on the expectation that it would buy power from Talen’s 2,228-MW stake in the adjacent Susquehanna nuclear generating station. That arrangement has since come under intense regulatory scrutiny.
Amazon and Google have partnered with privately held nuclear technology companies X-energy and Kairos Power to power data centers beginning in the early 2030s. Amazon gained 0.3% and Google parent Alphabet declined 4% in Monday trading.
These deals came amid steadily escalating projections for future load growth. In a pair of reports published last year, consulting and technology services firm ICF forecast U.S. electricity demand growing by an average of 2% annually through 2033, while the Electric Power Research Institute said data centers’ share of U.S. load could double to 9% by 2030.
Those estimates may already be out of date, said Himali Parmar, vice president of energy advisory services, interconnection and transmission at ICF.
“Relative to ICF’s September report, which projected a 9% increase in U.S. electricity demand by 2028, there is a huge increase in demand,” Parmar said.
A preliminary load forecast presented Dec. 9 by the PJM Interconnection, which hosts proportionally more data center capacity than any other load balancing authority, showed its summer and winter peak load growing by averages of 2% and 3.2% annually through 2045, up from 1.6% and 1.8% growth in its 2023 forecast.
But DeepSeek’s apparent dramatic improvements in efficiency suggests further AI performance gains may require less energy-intensive “compute” than assumed. That threatens “the bull thesis on independent power producers and most integrated utilities [that] is entirely dependent on data centers,” Jefferies said.
The last time independent power producers were this excited about an electricity-boom cycle was in the late 1990s and early 2000s, when demand-growth expectations were fueled in part by the growth of Silicon Valley. It didn’t end well.
In a piece published on Forbes.com in 1999, Peter Huber and Mark Mills wrote: “Southern California Edison, meet Amazon.com. Somewhere in America, a lump of coal is burned every time a book is ordered online.” The two authors, who co-wrote books about energy, including “The Bottomless Well,” estimated that one billion PCs on the Web would represent electrical demand equal to the total power capacity of the U.S. at that time. The piece drew much attention and pushback, including from scientists at the Lawrence Berkeley National Laboratory, who said the authors were overstating the impact.
In a 2001 PBS interview, Peter Cartwright, who was then chief executive of Calpine, said, “Silicon Valley, as everybody knows, is well aware—is one of the fastest-growing demand centers in the state, and we have very—hardly any power generated in this area.” Calpine was one of the most aggressive independent power producers of that era. Its installed base grew at a compound annual growth rate of 63% between 1998 and 2002 through both new construction and acquisitions. In addition to demand growth, Calpine and other developers believed that their new gas-fired power plants would come to replace older, less-efficient generators that were built by monopolistic utilities rather than competitive developers.
Here’s what Marc Andreesen, a big Trump supporter, thinks about AI:
https://futurism.com/the-byte/ai-investor-goal-crash-human-wages
https://www.businessinsider.com/marc-andreessen-ben-horowitz-donated-millions-pro-trump-pac-2024-10
I’ll go a bit off from the meaning of this post (AI might not require as much power as initially predicted) and into one of my ‘question technology’ meanderings. Something that hasn’t been mentioned by the press with the DeepSeek news is that it puts pressure on all the AI companies to speed up production, giving them less incentive to try and regulate a technology that almost certainly needs it, and desperately so. There’s even more pressure on the heads of these companies to ‘beat’ their competitors and ignore any and all warnings about the tech.
This was news today (would’ve chosen the Fortune article, but its paywalled):
https://www.dailymail.co.uk/news/article-14335985/OpenAI-steven-adler-quits-warning-risky-gamble-huge-downside-AI-safety.html
‘In my opinion, an AGI race is a very risky gamble, with huge downside,’ he wrote. ‘No lab has a solution to AI alignment today. And the faster we race, the less likely that anyone finds one in time.
‘Today, it seems like we’re stuck in a really bad equilibrium. Even if a lab truly wants to develop AGI responsibly, others can still cut corners to catch up, maybe disastrously,’ he added.
‘And this pushes all to speed up. I hope labs can be candid about real safety regs needed to stop this.’
‘Daniel Kokotajlo, a former OpenAI governance researcher, told Fortune in August that nearly half of the company’s staff – specifically the ones that focused on the long-term risks of superpowerful AI – had left the company. ‘
I don’t think regulation of AI tech would ever have been practically feasible. First of all, legislators truly suck at understanding technologies or the myriad implications they may portend, and secondly, there’s quite a lot of lobbying and international back doors to further complicate the matter. DeepSeek popping up is almost the perfect example of this.
And one more article:
https://www.theguardian.com/technology/ng-interactive/2025/jan/29/silicon-valley-rightwing-technofascism
Is it a wonder that these same guys have been supporting Trump? Has financial success gone to their heads? Or is it that they’ve always had oligarchic and authoritarian leanings?
Imagine a future where citizens can be tightly monitored (and therefore controlled), where the vast majority of people can no longer truly move up socio-economically because the job market has been decimated, true wealth has already been locked down by the few, and we have a government and corporate sector intent on keeping it that way. (Some would say we’re already there, I would say it can always get worse – much worse. Andreesen would call me a communist for the above, when I’m not. I’m just saying perhaps the path we’re on isn’t the best for either us or those who follow us.)
While some of those VCs rub their hands with glee at the thought of saving money on the lowly laborer, none seem to realize that the very predictable behaviors of VCs and those in the C-suites make them prime candidates for replacement by AI.
Whenever I read headlines about electrification of fossil fuel heating in the US with heat pumps or the potential for EVs replacing ICE vehicles, it was always how impossible it was to supply the energy and grid improvements to support it. Then when it came to the increase demand from planned datacenters and AI farms, it was we can do this, we have to do this, no problem!
I think “follow the money” more than explains the support or non-support of any policy. EVs are a threat to the US legacy car companies and the oil industry, and the grid can’t keep up. Now that AI has been the darling of US’ premier (i.e., rich) technology companies, We Can Do It!
Likely, perhaps, but the guy who has been making $10 million/year for over a decade and investing it in real estate and assets might not be impacted as much as the many more making a small fraction of that and spending all their money on living costs. Andreesen sees a future where everything is cheap because labor costs have disappeared. But who has the money then to buy things? Andreesen seems to just be assuming his reality is everyone else’s reality – or maybe he doesn’t do that and just doesn’t care.
After reading that the Chinese government was filtering responses about politically sensitive subjects I decided to try some questions. The responses about Tibet and Uyghurs in camps pretty much matched between deepseek and chatgpt. I think I need to ask more sensitive questions.
Runs on a laptop, lol. Still not ‘intelligent’, still can’t make a joke, connect a canard
Example: I run a talk to text translator on a tablet for my deaf spouse. At first it wasn’t very good; time passed it got better. Learned our unique accents, syntax and nuance. Machine learning. Big database. Not intelligent
Runs on a laptop, lol. Leon’s, Gate’s and Zuck need ((( atomic power )))
DeepSeek is truly a sputnik moment and nothing more. What shocks me are the number of people who claim (falsely) that China steals technology. On a related note, 99% of all the social media companies got rich by using open-source software (Linux, mysql, etc) which saved tons of money by not licensing proprietary software. Then when those same companies began working on LLM, they moved to a closed source model (this is also true of OpenAI where the word “Open” is only in the name). Meanwhile, DeepSeek is open-source and (like all open source nerds) publish their code to the world on one of the usual GIT platforms. Check this out: https://github.com/deepseek-ai
“What shocks me are the number of people who claim (falsely) that China steals technology.”
Really? There are still things that shock you at this point?
As for China stealing technology, that was true for a time decades back, as the United States did after it split from the British Empire. Then they started investing in and educating the hell out of their researchers and engineers. The US industrial model isn’t keeping up, in part because they’ve had easy access to the richest consumer base in the world, and didn’t have to work as hard.
The fact that DeepSeek is based on open source is an extremely big deal. There’s no basis for fear-mongering based on secret proprietary tech that the Chinese can use as a Trojan Horse. (There will probably be such fear-mongering, but there’s no real-world basis for it.)