Meta did something right, and it deserves recognition and replication.
Recently, the American-owned technology company announced a private partnership with three nuclear firms, Vistra, Terra Power, and Oklo, to expand electricity production supporting their own data center needs and generating surplus power to benefit the regional grid.
This should happen anywhere data centers are in the discussion. It is what every politician courting the tech communities to build data centers in their city or state must expect going forward, because right now data centers face massive opposition, and no amount of spin will fix it.
The main concern about data centers is their incredible electricity need. Of course, we all like and use technology, and your average American even fundamentally embraces the patriotic need to “beat China” in the AI race. None of that denies the “affordability” issue, and the frustration of ratepayer who have yet to see a tangible benefit ubiquitous data centers bring to his personal life. To highlight this, let me use the example of a rural Virginia farmer: me.
For years, Virginia enjoyed bipartisan enthusiasm for data centers. A third of the nation’s data centers are in the Commonwealth, and every politician relishes ribbon cutting photos and self-congratulatory press releases promoting “investment” and “jobs” and “tax revenue.”
What do I see? I see my farm’s electric bill 40% higher than it was in 2020 with a projected 14% increase this year. For all of the historic revenue, my tax burden remains the same (or higher). From my perspective, data centers are a burden.
There is a general good, and I am happy for all the construction jobs, electricians, cement and steel workers. There is new money for “education.” I can find dozens of statements celebrating essential programs for “mental health” and “nutrition” and whatever else sounds caring.
None of this benefits me. In fact, it costs me several hundred dollars more per year. I would at least like a thank you. Maybe a fruit basket.
Meta did something right, and it deserves recognition and replication. It did not just announce a plan to build expecting ratepayers would finance it through higher monthly bills. It brought its own money and government coordination to tell the locals things could actually… get better. Imagine that.
Our elected leaders should be building nuclear power plants (and coal power plants and natural gas power plants) because we are in desperate need of increased baseline power, something Energy Secretary Chris Wright deeply understands and regularly addresses. During recent remarks to Meta, the Secretary mentioned that decades ago, before Meta founder Mark Zuckerberg was even born, he was studying nuclear power in college, and here he is, decades later, helping usher it into reality.
Meta has the money. Vistra, Terra Power, and Oklo have the know-how, and Chris Wright has the humility to let it happen. Government as a willing facilitator to online new energy technologies for the benefit of us all. Imagine that.
I named my organization Power The Future deliberately. What will power the future? “God only knows…” to quote The Beach Boys, but one thing which I do know, of which I am certain, the geniuses who will power the future with new technologies and new energy sources are, right now, in need of reliable, affordable, abundant energy, American energy, and if we punish ourselves now by punishing energy, we are not hastening that better tomorrow. We are conscripting ourselves to the status quo.
AI is likely going to figure out how to power the future, but AI will not happen unless more power comes online fast. Meta did it the right way, and every tech company with data center desires should take note. We may just be simple farmers, but we vote and we fight. You can win us over if you treat us right, and Meta is the first to do so.
Daniel Turner is the founder and executive director of Power The Future, a national nonprofit organization that advocates for American energy jobs. Contact him at daniel@powerthefuture.com and follow him on Twitter @DanielTurnerPTF
This article was originally published by RealClearEnergy and made available via RealClearWire.
Plain and simple…
With Nuclear, AI is a piece of cake.
With Renewables only, AI is a piece of carp!
Sounds fishy to me!
In the local river, and iirc, anywhere in NSW, if you catch a carp, it is illegal to throw it back.
You must kill it and bin it.
Why can’t you take it home and eat it?
For the same reason you wouldn’t eat a cane toad – they’d taste like shit.
Yep.. a fisher friend of mine says even cats won’t touch them !
Down the local fishing place, some people leave them on the grass…
… and even my dog won’t roll in them.
Ever heard of Gefilte fish, a classic Jewish dish.
Interesting. Maybe USA carp are not as nice as Polish carp which is a traditional meal at Christmas.
I’ve read that the Brits make a stew of carp and some sort of river grass.
I reckon they’d have strain the liquid off for soup, drown it with Worcester sauce, and throw the grassy fish mush onto the garden as fertilizer.
I know a few people who eat carp in the US, they are mostly from the southern states where “some” will eat anything that moves and aren’t afraid to scrape fresh road kill up and eat that too. Being from the PNW we usually toss them on the bank but some take them home to use as fertilizer in their gardens.
I’ve never had carp myself but I’ve known people that do take them home and eat them.
The trick is that, while cleaning them, you need to carefully remove what they called “the mud vein”.
There are different types of carp but none are native to the US.
(There are native fish that fill the “bottom feeder” role of some types of carp that have less destructive feeding and spawning habits.)
“Grass carp” can be a big problem in some areas.
In Ohio it is legal to stock them for aquatic plant control but they must be sterile. (Here they are called “while amur”)
https://fwfarms.com/what-is-the-difference-between-white-amur-and-grass-carp/?srsltid=AfmBOopWXE9SfC8oX1qPOKG5UJi1IQ3tSChfXi7nhort8cLpVE_OjXFd
Kill it and throw it back is a viable option….
They are also a test case for their peers who are rushing ahead faster than the 49er Gold Rush. The rate of announcements and deals is just incredible.
25 of your 40% increase is inflation from Biden’s flooding of the money supply. The rest is likely due to state energy policies having more to do with forcing renewables and discouraging inexpensive gas generation. Data centers probably aren’t the driving factors. VA already gets about a third of its electricity from nuclear. What’s needed is to make nuclear cheaper by reforming NEPA, which is what allows “citizen lawsuits” which have been weaponized to cause expensive delays by green activists. The main problem with nuclear now is it takes too long and costs too much, both entirely due to environmental lawfare.
Just make Renewables “Right Priced” by eliminating ALL subsidies as well as “Take or Pay” so they have to charge their actual price.
The FED regulation of the moneysupply is independent of politicians.
That’s why Trump is squawking because Fed wont flood the money supply at his demands
Nothing is independent of politicians.
“Nothing is independent of politicians.” Unless there’s no money in it. Don’t forget, we’ve got the best government money can buy :<)
Yes, legacy mega-scale nuclear is not an option in the U.S. The well funded sophisticated anti-nuclear opponents, with help from activist liberal judges more than willing to entertain frivolous delay tactic lawsuits, won that battle. Move on.
The answer is assembly line factories producing one NRC design approval fits all cookie cutter identical small modular reactors (SMR’s) delivered on semi-trailers for plug and play to sites prepared in advance for their arrival. Flange up, test, commission, and in operation in months not years. The Administration needs to incentivize industry to put together a consortium of energy companies and investment banks to get such a factory built and in operation ASAP.
AI isn’t the Virginia power problem, it’s wind and solar.
and here’s why they did it:
Is Cap-and-Trade Saving the Planet—or Just Creating the next Bubble?
Cap-and-trade sounds like an elegant solution: set a cap on emissions, let companies trade allowances, and watch the market drive efficiency. In reality, it’s less about physics and more about finance.
Here’s how it works: Governments set an emissions cap and issue allowances. Companies that emit less than their allowance can sell credits; those that exceed must buy credits or pay penalties. This creates an artificial market for carbon credits—value driven by regulation, not intrinsic utility. The money flow starts with regulated industries purchasing credits, moves to credit holders and trading desks, and governments skim auction revenue.
Consumers ultimately absorb the cost through higher energy and goods prices. Financial intermediaries—banks, brokers, carbon funds—often profit more than technology innovators, turning this into a trillion-dollar speculative market.
Compare that to a carbon tax: A tax is simple, transparent, and predictable—fixed cost per ton of CO₂. Cap-and-trade offers certainty on emissions but not on price, and it’s prone to lobbying, rent-seeking, and opaque trading. A carbon tax doesn’t guarantee emissions reductions, but it avoids the complexity and corruption risks of a cap-and-trade system. In short: cap-and-trade is a compliance-driven financial ecosystem that incentivizes trading over real decarbonization.
Conclusion:
Cap-and-trade isn’t an environmental solution—it’s an economic distortion. It creates a trillion-dollar compliance market that rewards financial engineering over real emissions reduction, while consumers and manufacturers shoulder the cost. If the goal is genuine decarbonization, transparency and technology—not speculative trading—should lead the way.
Very nice Daniel. This only makes sense. Any outfit not just AI that is going to use massive amounts of energy should be made to join in and insure that we all have plenty of energy. Not just outfits that want to connect to the grid but also those who are currently connected to it. That doesn’t mean each outfit must build their own power plant there are likely dozens of investments that can be made to insure plentiful power for your region. One stipulation should be you must invest in a system that will supply your region. If you choose wind or solar that will be your sole source of power so you better be able to operate on intermittent power. Choose wisely the required investment could make you a bunch of money.
From the above article:
“Meta did something right, and it deserves recognition and replication.
Recently, the American-owned technology company announced a private partnership with three nuclear firms, Vistra, Terra Power, and Oklo, to expand electricity production supporting their own data center needs and generating surplus power to benefit the regional grid.”
“Something right”??? . . . well, that remains to be seen/proven.
Neither Vistra nor Terra Power nor Oklo have built a single nuclear reactor power plant. Vistra has acquired ownership and operates previously built nuclear fission power plants. Terra Power and Okla are known for their individual plans to build and operate new types of nuclear fission power plants (e.g., SMRs, traveling wave reactors, molten-salt reactors, sodium-cooled fast reactors) but have not completed any such construction to date.
There should be alarm bells going off in Meta headquarters for partnering with relatively immature companies having no previous experience with building any nuclear power plants from the ground up to bring such on line to produce industry-scale electricity.
Hmmmm . . . perhaps Meta has an AI bot advising them that such prudence is no longer required in today’s world? Sure, I’ll give it recognition for such thinking.
Maybe, just Maybe these are the only available options in your country and that real alternative are only available in heavily sanctioned countries.
Maybe . . . we’ll see. There was—once upon a time in the USA—belief and encouragement of individual initiative and individual investment for proceeding in “new directions”.
But sadly, within the last 50 years of so, that initiative has been replaced by belief that anything deemed technically or financially risky should be first SUBSIDIZED or LOAN-GUARANTEED by
governmenttaxpayers to demonstrate it is even feasible.There’s a lack of companies with experience in building civilian nuclear plants in the US since we are not exactly building them. To put it in perspective Nuscale I believe is the farthest along in licensing SMR’s for use in the US and they’ve yet to build one.
FYI, the two most recent large capacity, civilian power plant nuclear reactors to be brought on-line in the US were Westinghouse AP-1000 reactors, each rated at 1117 MWe, as Vogtl Unit 3 (commissioned in 2023) and Unit 4 (commissioned in 2024).
“. . . Georgia Power Company reached a contract agreement for two AP1000 reactors designed by Westinghouse, owned by Toshiba. Westinghouse partnered with the Shaw Group (Baton Rouge, LA) and its Stone & Webster division to manage the project with Westinghouse responsible for engineering, design, and overall management, and Shaw responsible for manufacturing the pre-fabricated component modules and managing the on-site construction.”
(https://en.wikipedia.org/wiki/Vogtle_Electric_Generating_Plant )
I do not know what US-based companies could build a full nuclear power plant today (beyond just contracting a supplier for the nuclear reactor(s)), but the following have announced plans to do so:
— Duke energy
— UniStar Nuclear Energy (a joint venture of EDF and Constellation)
— Westinghouse Electric Company
(https://en.wikipedia.org/wiki/List_of_companies_in_the_nuclear_sector ).
A good article raising the issue of power rates. The energy demands of hyperscale datacenters are not mentioned in this article. Neither are the demands for water cooling.
But it is more than the largest cities in each state.
Another rising cost to consider — the cost of buying personal computers. Companies that have a significant market share in consumer electronics are shifting production to compute — for datacenters.
So jeff bezos thinks you should just compute with the cloud, <b>not own a pc.</b>
That’s water, electricity, and computers. All for a technology which at the moment, it seems, users are more concerned with finding out how to disable it.
Data centres doent use the same components that are in home or office computers
AI chips could be $5k-10K each
Home computer prices rising because they include simple AI software technology
My annual microsoft license went up recently because they included AI software and its links to the data centres. I was able to disable in my MS account this and get a reduced annual payment
“I was able to disable in my MS account this and get a reduced annual payment”
YEP…. Me too 🙂
Vistra is the only one with actual nuclear power plants
Terra Power, and Oklo are just working on possible designs with different technology than currently used.
I wish them well but its a long journey
Story tip
Germany appears to be giving up on renewable energy (although they pretend not to)
https://www.tagesschau.de/wirtschaft/energie-neue-gaskraftwerke-100.html
Did you ever wonder from where Germany will get the gas?
If you blow up a few more of their pipelines you’ll be the main provider.
And if you don’t have enough gas – do what you always do.
Steal some ships from foreign countries(,national security, sanctions, drugs ,WMD’s – any excuse will do, just as always)and sell it to them.
Meta is not innovating.
It does not deserve “recognition and replication”.
It is only starting to do what traditional energy intensive “heavy” industries have done since just about forever.
It is not at all unusual, in fact it is quite normal, for traditional energy intensive “heavy” industries to provide for their own electricity generation on-site with grid connections only to provide emergency back up.
There are several reasons for this in no particular order:
==> Control
==> Certainty that the power will be available when needed for initial start-up
==> Speed. Power generation included as part of the facility initial investment is quicker than getting it included in the power utility’s investment planning and capital project execution programme
etc., etc…….
There is no good reason why data centres with energy demands comparable to traditional energy intensive heavy industry should not be required to provide for their energy requirements by themselves.
So, now that they have decided they need to provide for their own power, the only question is about which power generation technology to use.
Renewables are out of the question because to build enough to provide for full annual demand given capacity factors would require at least 6 to 8 times nominal capacity over-build compared to demand and would also require a very large (currently unproven) LDES system.
Meta et al. have worked this out.
They are not going to go renewable.
And they are not going to go nuclear, the new technologies will not be tested and approved in line with their energy needs.
The nuclear option is a wedge they will use to get permission to rely exclusively on gas turbine systems, with maybe some purely nominal BS contributions to planned renewable projects.
Cynical? Or a realistic assessment? You decide!
“Renewables are out of the question”
Yep.
Good luck trying to run a Data Center on windmills and solar.
From the article: “This should happen anywhere data centers are in the discussion. It is what every politician courting the tech communities to build data centers in their city or state must expect going forward, because right now data centers face massive opposition, and no amount of spin will fix it.”
That’s exactly right. No amount of spin will keep people from fighting against higher electric bills.
President Trump emphasized that new Data Centers should pay for their own power needs and not put the costs on the general public.
If Data Centers do not do this, I don’t think they are going to get built because of the public opposition.
President Trump has laid out the path to success and Data Centers should follow it.
Only problem with that is the IEA say investment in data centres worldwide was expected to reach USD 580bn in 2025, surpassing the 540bn spent on global oil supply and over 85% of new data centre capacity growth in the next 10 years will be in the US, China and Europe.
IEA ‘World Energy Outlook 2025’ (Nov 2025)
Here’s a simple question for the WUWT group mind: Why do AI “data centers” require so much continuous power in the first place?
So much power was not previously required for Google’s remarkable web search engine (as well as other later competing search engines) nor for Wikipedia’s amazingly broad and deep and searchable encyclopedic data base.
Likewise, so much power was not previously required for computers to perform some of the most complex, mathematically-intensive tasks in such fields as:
— cryptography
— weather modeling/forecasting
— calculating spacecraft interplanetary trajectories to high accuracy
— analyzing collisions and byproducts from particle accelerators
— performing 3D rendering in the arts and sciences
— photo and video image processing
— performing electronic circuit analysis and PCB layout optimization
— performing CAD/CAM, FEA and CFD engineering design/analyses
— signal detection/analysis (especially in very low signal-to-noise environments)
— real time video gaming.
For reference and consideration, a single human mind is powered by metabolic power of 80–100 W in a “resting” (but consciously thinking) state.
Cryptography has always required lots of electrical power and has always required cooling for that power. Supercomputers went to water cooling fairly early. It is not a new issue.
But the issue I raised wasn’t at all about the means of cooling of AI data centers, but instead about why they require so much continuous power in the first place.
You seriously underestimate the electricity and cooling demands of IT.
A significant part of the reason why many 1950’s and 1960’s office buildings have been demolished rather than refurbished is that their floor-to-floor heights were insufficient to permit upgrading their HVAC systems (if they had them at all, e.g. in the UK) to handle the loads required in a modern office.
The question that I asked was specific to AI (artificial intelligence) “data centers” as implemented on modern computers. Such do not date back to the 1950’s and 1960’s, nor are AI computers installed in “office buildings” as that term is commonly used.
Good question.
I heard someone on tv say the other day that there were about 4,000 Data Centers in the United States, and the introduction of AI would double that number.
But those initial 4,000 Data Centers are not the energy hogs that AI Data Centers will become.
As for why AI Data Centers need more electricity than the older Data Centers, I guess it takes lots of power to try to think like a human being. 🙂
Well, perhaps.
But isn’t the whole point of AI to develop and implement computer coding/self-coding (“learning”) that DOES NOT “think like a human being”, but to instead enable near-instantaneous access to most of the data that exists in the world while at the same time not being subject to logic failures, confirmation bias, memory lapses, hallucinations, math calculation errors, and other common failings of the human mind . . . even the best of them?
You literally could ask any of a dozen AI’s that question:
The short answer is: modern AI data centers are doing fundamentally different work, in a fundamentally different way, than classic web search or Wikipedia ever did—and that difference translates directly into massive, continuous power draw.
Below is the precise breakdown.
1. Search and Wikipedia mostly retrieve data; AI mostly computes
Classic search (Google, Bing, etc.)
A traditional search query primarily does this:
Key point:
Most of the “hard work” was done offline, ahead of time:
At query time, the system mostly performs:
This is I/O-bound, not compute-bound.
Wikipedia
Wikipedia is even lighter:
A page view is often just:
That’s why Wikipedia famously runs on a comparatively tiny infrastructure.
2. AI inference is real-time numerical simulation, not lookup
Modern AI (especially large language models) does something very different:
A single prompt can involve:
This is compute-bound, not I/O-bound.
3. Why GPUs (and TPUs) are power-hungry
AI workloads rely on accelerators designed for maximum parallel math:
They are intentionally pushed near thermal limits to:
Typical numbers:
Search servers, by contrast:
4. AI workloads are continuous, not bursty
Search traffic is:
AI inference is:
There is no equivalent of:
Every interaction:
5. Training dwarfs everything that came before
Everything above is just inference.
Training is orders of magnitude worse:
Google Search never required:
AI training does.
6. Why this wasn’t obvious from earlier “AI” hype
Earlier AI systems:
They scaled with:
Large models flipped the equation:
Power consumption is not a bug; it is the mechanism.
7. Bottom line
Search engines and Wikipedia:
Modern AI:
That simulation:
Hence:
If you want, I can also:
CTM,
Thank you, honestly and deeply, for taking the time to provide such a comprehensive reply.
I appreciate the reasoning that AI offers for its own existence.
Just these closing comments:
Under its “7. Bottom line” comment, the AI you employed states:
“Hence, AI data centers consume enormous, continuous power because they are running real-time numerical engines, not information libraries.”
This begs the question of “Well, what was wrong with information libraries?”, other than they didn’t tell people how to think (e.g., “synthesizes new output”, “Power consumption is not a bug; it is the mechanism”), instead only provided information to think about.
But perhaps this distinction is too philosophical/metaphysical to contemplate further?
Thanks again for all you do to make WUWT the great site that it is!
I took a job at a DC a while ago. While I can’t give specifics I can say when I power on a new rack, without even looking at rack type I can tell you if it’s for AI or not. How? Sitting idle without even being plugged into anything but power it’s pulling roughly 50% more amperage.
“Pulling” power without doing anything productive?
Where is DOGE when one really needs them? /sarc
If you’re handing out Kudos might consider that AWS announced plans in 2024 that it was going in partnership with Xenergy and Energy Northwest in Washington state to build a new nuclear. Meta is actually kind of late to the game.
Minor edit:
We are conscripting ourselves to the status quo or worse.