“The ban on research and development into artificial intelligence is, as we all know, a holdover from the Cylon Wars.” – Battlestar Galactica (2004)
When I asked my mom if I was ugly, she said, “I’ve told you not to talk to me in public.” (All memes as found.)
I remember the dotcom bubble.
Back in the late ’90s, everyone was throwing cash at anything with a “.com” slapped on it. Anything. Take Pets.com™, which had the idea that they could take orders for dog food online and that would lead to them being worth a trillion dollars. Instead? They spent $11.8 million on ads which resulted in $619,000 of total sales. But wait, there’s more! Their business strategy was to sell their products at 30% of what they paid for them!
Genius! I suppose they thought they could make it up on volume?
That’s just one example, and there are thousands of companies that burned through money like cocaine-addled chipmunks going through nuts. Billions of dollars vanished, but hey, at least we got Jeff Bezos managed to get a slightly used wife out of it.
Fast-forward to 2025, and we just may be in Dotcom 2.0: the AI edition.
This time, it’s not websites filled with dancing hamsters. Nope. Data centers are sprouting like marijuana in a Colorado hippie’s backyard. Chipsets are piling up like Indians in Canada. The spending is insane on this bubble, and if history’s any guide, the pop could echo for decades.
The source of this frothy mess?
Massive investments in AI infrastructure. In the first half of 2025 alone, spending on AI data centers and related gear added more to U.S. GDP growth than all consumer spending combined. This is about $75 billion from AI infra versus $69 billion from folks buying lattes and lawnmowers.
I tried to get the lid of my pen for ten minutes. Nothing was working. Then it clicked.
That’s right: Big Tech’s server farms are propping up the economy more than shopping. Companies like Microsoft®, Google®, and Meta® are pouring trillions into building these behemoths, buying up NVIDIA® chips like they’re the last Twinkies® in a zombie apocalypse. It’s not just servers; it’s cooling systems, fiber optics, and enough wiring for George Bailey to finally lasso the Moon.
Why?
Because AI needs compute power like a teenager needs a cell phone: continually and without gratitude.
So, how long can this bender go on before someone yells “last call”?
Analysts are projecting explosive growth through 2030 but they also told people that Pets.com® made sense. Bubbles don’t burst on schedule, they pop when reality bites. McKinsey estimates we’ll need $6.7 trillion worldwide by 2030 just to keep up with compute demands from the various AI products, while the global AI data center market is forecasted to balloon from $236 billion in 2025 to $933 billion by 2030, growing at a scorching 31.6% yearly.
Where will the power come from? 10 gigawatts of new data center capacity will break ground this year alone, with construction at record levels and power transmission delays stretching to four years in some spots.
Before electricity, were people sentenced to death in the acoustic chair?
Let’s extrapolate this:
If spending keeps doubling every couple of years, as it has since ChatGPT lit the fuse, we’re looking at a timeline where the frenzy peaks around 2028-2030. By then, data centers could consume as much electricity as Gavin Newsom’s blow dryer, and the supply chain for chips and rare earth metals starts buckling.
Analysts predict data center power demand surging, but what if AI hits diminishing returns? We’ve seen it before: the dotcom buildout assumed infinite internet growth, but when the stunning genius of selling products for 70% less than you bought them for didn’t pay off, the house of cards folded. Rapidly.
If AI doesn’t deliver massive productivity gains or the company can’t figure out how to make it up on volume, investors pull the plug. My guess? This bubble could inflate for another 3-5 years, then deflates when ROI reports come in looking like a kid’s lemonade stand profits for some companies.
Salmon don’t watch cable TV. They prefer streams.
It’s not just the data centers themselves; the ripple effects are creating mini-bubbles in related bits of the economy. AI’s thirst for electric power is turning it into the new oil. The International Energy Agency projects global data center electricity demand more than doubling by 2030 to 945 terawatt-hours, enough to power Australia several times over if they ever figure out electricity.
This means billions funneled into new power plants, grid upgrades, just to keep the lights on in these silicon sweatshops. Utilities are scrambling: nuclear restarts, solar fields the size of small states, and even deals with fusion startups that sound more sci-fi than spreadsheet. This is trillions spent on infrastructure, from transmission lines to cooling systems that guzzle water like a camel in the Sahara. If the bubble bursts, we’re left with ghost grids and stranded assets, much like the fiber optic cables buried post-dotcom that still haunt telecom balance sheets.
What do a ring, a baby, and a threesome have in common? None of them are going to save a relationship.
What happens if AI reaches its mature end-state? We’re talking Artificial General Intelligence (AGI) where machines that can do any intellectual task a human can, not to mention Artificial Superintelligence (ASI), where they outthink us like we’re Mexican mall lawyers trying to fix a copier.
Some whisper we might already be there, with models like Grok™ or whatever OpenAI®’s cooking up blurring the lines. But assuming we hit it soon, the economy does a backflip.
In the AGI/ASI world, productivity explodes: AI handles everything from coding to curing cancer, slashing costs and boosting output.
But jobs? Poof.
Hey, let’s see it take a 15 minute coffee break.
Economists at AEI outline scenarios where AGI displaces masses of workers: truck drivers, lawyers, artists. Optimists say it will augment humanity, creating new gigs in “AI wrangling” or whatever.
The dark side for this case: inequality skyrockets. A few tech overlords own the AIs, reaping trillions, while the rest scramble for UBI scraps.
Civilization-wise, it’s transformative: endless innovation, but if ASI “solves” economics without humans, we enter a post-scarcity utopia . . . or dystopia, where labor is worthless and purpose is a luxury.
If we’ve hit AGI/ASI now (debatable, but let’s play along), the bubble accelerates short-term as companies race to integrate, then crashes when overcapacity hits. Data centers become obsolete overnight if ASI optimizes compute down to a laptop. The fallout? Trillions in sunk costs, like building railroads right before cars took over.
Scooby Doo® taught many kids that if they smoked enough pot, their dog would talk and help them look for snacks.
If AI fails (and there is no sign of this) we end up in, at least, a dotcom-style recession.
At least.
If AI succeeds, in the best case we end up in a strange, post-scarcity world, but a world that hardly needs us.
I guess we could make it up on volume?
I just started a new serial story called “Translation Error.” The first chapter is up now. Let me know how you like it.
https://open.substack.com/pub/zaklog/p/reunion?r=2nmhek&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true
Keep us posted!!!!
So, where’s the “juice” gonna come from to run AI? Duke Energy stopped work on a nuke plant in Upstate SC is the 1990s; it’s only use was for filming underwater scenes in “The Abyss”, as far as I know. And, SCE&G went bankrupt trying to build one in the Midlands around 10 years ago. Given permitting & construction (NIMBYs too), it’d be well over 10 years to generate sufficient power to handle AI, which could be “DOT.COM Bubble 2” by then.
Not to mention where’s the capital coming from?
Exactly. But more on that in another post . . .
Crypto-currency mining is the one that really annoys my engineering sensibilities. For our entire lifetimes we’ve been told that we’re running out of oil and we need to conserve. We’ve even fought wars over it.
But, out of the blue it is now suddenly okay to burn massive amounts of electricity for the sole purpose of running random number calculations to support digital money. No useful work being done…..just random number crunching. I guess conserving that limited energy supply is only necessary for the dirt people.
The good news….many communities are pushing back and winning. Crypto consortium set up a mining operation in a rural community near me but backlash was so bad that they are being forced to shut down. It is extremely noisy and did not provide the jobs that were promised, not to mention the perception that it is increasing the cost of electricity for the average consumer.
The current surge in AI implementation is really about control. All the current tech giants are afraid of being the next Ask Jeeves or Yahoo! also-rans in the AI arena come a decade from now. Whoever builds the biggest AI center now gets to write the most homework assignments for todays student-loan-financed college students. Seriously – research shows that research, er, cheating at school is the current primary use of AI.
https://www.yahoo.com/news/articles/openai-usage-plummets-summer-students-152156880.html
Whatever brand of AI becomes the de-facto standard of use among these Gen Z “scholars” becomes the gilded cage in which they live the rest of their online lives. It used to be all about fighting for shares of eyeballs glued to screens and the “attention” (and ultimately advertising revenue) that represented. The current AI gold rush is instead about fighting for shares of thoughts downloaded into minds and the “opinions” that are thereby absorbed.
Orwell thought the government would own all the surveillance cameras and totally missed that everybody would own one in the form of a cellphone to produce endless viral video clips of everybody’s most revealing moments. Likewise, he thought a totalitarian Party would have to push universal slogans to be absorbed by all: Peace is War, Freedom is Slavery, Ignorance is Strength. He never dreamed of a world where everybody would receive their own personalized propaganda by asking a trusted oracle.
Yet here we are.
But what about AltaVista?
Did You NOT ‘see’ the actual ‘Prioritization’ of Needs based Humanity ?
https://www.zerohedge.com/energy/texas-preparing-cut-power-data-centers-during-grid-emergencies
🤣 🤣 🤣
NOT a ‘State’ Issue.
The govt. WILL cut off Your Heating/Cooling, Your Water, Your lights,….EVEN Your Legs and…
https://israelect.com/reference/WillieMartin/P.L.102-14.htm
…WHEN it comes to that…BEFORE one data center looses a Watt.
ANY ‘spare’ capacity will be for bitcoin mining !
https://www.coinbase.com/price/official-trump
UBI ?
Why, non other than the vaunted thomas paine…
https://www.bing.com/search?q=Thomas+Paine+and+Universal+Basic+Income&form=APMCS1&PC=APMC
THOUGHT it would stop ‘Revolution’ ???
We Shall See, indubitably.
P.S. Stole your wild-caught Salmon and put it on TBP. Thank You
Golden rule, and they have billions.
As you well know, John, I could write enough about ES v. AI v. Thinking Machines to break your comment section. So I’ll talk about pursuing my enlightened self-interest, instead. Once upon a time my wife and I were both engineers in wildly divergent fields. Most of our friends were engineers. In the mid-90s, I learned to keep my eyes and ears open about tech matters. Thus, I was able to ride the following before they plateaued or cratered: dotcom, fiber optics, flat glass. All of which has me in the very comfortable place I am in now. Now, per your essay, I told my money guy to look into power generation funds to hold for the next 3-4 years. I am more than willing to exploit other’s uninformed enthusiasm.
Have him investigate Kairos. Prototype bleeding-edge-design molten salt cooled nuclear reactors hooked up to the TVA grid and turned on for the first time ever after only five years of construction – what could possibly go wrong?
https://www.cnbc.com/2025/08/18/google-kairos-nuclear-smr-tennessee-valley-authority-tva-data-center-ai.html
WCPGR? Um, this….
https://www.utilitydive.com/news/electric-transformer-shortage-nrel-niac/738947/
Many of my stories are “homed” in Knoxville-Oak Ridge for reasons such as this. When they start to push back against the collapse of civilization, they tend to use smaller thorium pebble-bed reactors.
I’ve always been interested in nuclear tech from my half-decade of working in Oak Ridge and I too have long thought pebble bed reactor (PBR) tech is a much better and safer idea than the insane commercial light-water reactors (LWRs) used today. I think the main gotcha with pebble bed (besides its relative thermal inefficiency) is the traditional use of helium gas for coolant, a resource in increasingly short supply today. Plus, the mechanics of pebble motion in a PBR is no picnic, either.
https://lvenneri.com/blog/pebble-bed-nukegumball
https://hackaday.com/2023/12/14/how-germanys-troubled-pebble-bed-reactor-came-of-age-in-china/
Both the US and German test PBRs from decades ago and the current Chinese PBRs (their HTR-10 series) use helium gas at temperatures up to 1400 degrees F. However, the Google / Kairos pebble bed reactor scheduled for turn-on in 2030 Oak Ridge is gonna use so-called FLiBe molten salt as the coolant (more accurately, the heat transfer fluid), which can theoretically work all the way up to 2600 degrees F with corresponding increase in thermodynamic efficiencies. Kairos recently opened their plant in NM to make this stuff.
https://www.energy.gov/ne/articles/kairos-power-breaks-ground-molten-salt-production-facility
It’s worth noting that other small nuclear reactor companies like Starcore Nuclear are splitting the difference – TRISO fuel in more thermally efficient prismatic graphite blocks instead of pebbles, but still cooled by helium. This is the “high temperature gas reactor” or HTGR / “Prismatic” approach, as opposed to PBRs. Bottom line, at this point, nobody knows what mix of nuke tech is “optimal”.
https://starcorenuclearpower.com/technology/
So…new nuclear design using new coolant material boldly going where no operating temperature range has gone before? All to power new Star-Trek-style M-5 Multitronic Ultimate Computers dreamed up by Richard Daystrom, er, Sam Altman?
https://memory-alpha.fandom.com/wiki/The_Ultimate_Computer_(episode)
I repeat, what could possibly go wrong?
Given that the companies banging the drum for “AI” (God, they and I hate using that term) are all on the Left Coast, I think any of these reactors, established tech or semi-experimental, should be built in California. If something does go wrong, well, that’s several thousand birds with one pebble, er, stone.
My regards to Captain Dunsel.
CB-
Although my biz was based in Charlotte, won a contract to install groundwater monitoring wells in East TN, SW VA and South WV & for a national company 20 years ago. First week was in the Knoxvegas area. Always supervised my drilling crew the 1st week for a new client.
Stayed at a Holiday Inn in West Knoxville; trolled at Baker/Peters (Peters/Baker?). Cougar Slut Central, both places. Fun mid-50ish days.
I wouldn’t mind spending some of my late July/early August time there in Downtown. But not much. Charleston & Beaufort are too much fun.
Exactly! Embrace the bubble, and get out when it’s right. But never trust an internet humorist for financial advice!
It cracks me up that after growing up in the era when everyone was terrified about nuclear power, now we are going to restart shuttered plants and build new mini-nuclear power plants just so I can make racist AI images.
“Just?”
We’re talking vitally-important work here.
Fusion. Microsoft is claiming fusion. I don’t buy it.
Uhm, did y’all forget? (((They))) are gunna kill the useless eaters? Don’t take it personal, unless (((They))) consider YOU or YOURS, are useless.
Tree Mike
8 to 5 you are in there
Exactly.
There was a press release last week where an AI startup called Perplexity had made an offer to buy Chrome from Google for around $30 billion. I’d never heard of the company so was surprised that a no-name could come up with that kind of money. Did a little searching and it turns out they are a recent startup offering yet another AI engine that isn’t really any different from every other AI engine. But as with the Dot Com bust, never let reality get in the way of expanding that hype bubble.
How did they get the money you ask? Turns out they were valued at $400M a couple of years ago but because of all of the AI hype, their value literally shot up overnight and is now at $18 billion dollars.
Our whole financial system is now based on retards and unicorn farts.
JB
Yup, full dotcom hype.
“That’s kind of one of the other reasons we wanted to build a browser, is we want to get data even outside the app to better understand you,” Preplexity CEO Aravind Srinivas said to Techcrunch 2025.04.24
https://techcrunch.com/2025/04/24/perplexity-ceo-says-its-browser-will-track-everything-users-do-online-to-sell-hyper-personalized-ads/
His new browser will be called comet. His AI is called Perplexity. Motorola, announced Thursday, where their app will be pre-installed on the Razr series — out of the box spy machine. Perplexity is also in talks with Samsung
It is just another control device.
As long as they control the motion, they don’t care where it goes.
I live in a swamp surrounded by firewood on-the-hoof.
I’m good with cutting the big cords that run near my house, those big towers are on a single point with guy wires and I own an acetylene torch.
Hahaha, that does tend to make quite a sound when it hits.
AI is just Spell-Correct, UniBlab, and HAL 9000, gone global.
That should have told anyone all they needed to know about it.
We’ll see . . . it’ll still unemploy Indians.