We May Not Get Our Recession until the AI Spending Bubble Implodes

But this stuff takes years. Until then, the massive spending from corporate balance sheets and cash flows that circulates & circulates is very stimulative of the economy.

By Wolf Richter for WOLF STREET.

The amount of money that gets spent on AI is just mind-boggling. This spending shows up in two categories on corporate financial statements: One on the income statement where it drives up expenses and pushes down earnings; and the other on the balance sheet, where it’s called “capital expenditures,” an asset, with no impact on expenses and earnings right now, but at a later date.

AI-related spending on the income statement that push down earnings include salaries of AI-related staff, additional office space, software services, electricity to power and cool AI data centers, costs of leasing AI servers, the engineering and design work that goes into all this, etc.

AI-related capital expenditures that go on the balance sheet as an asset include the purchases of AI servers with chips from Nvidia and others, construction or purchase of data centers – the size of AI data centers is expressed in the amount of electricity they can consume at capacity, such as a “500-megawatt data center” – the construction of powerplants and grid infrastructure to supply electricity to these data centers, the costs of reactivating nuclear reactors and retired fossil fuel plants, the land acquisition costs, etc.

These are vast amounts of money that are now getting spent. The Magnificent 7 talk a lot about it, Nvidia hyping the “insane demand” for its chips, and the others spending that money on those chips. Tesla has announced that it would expand its data centers to 500-megawatt centers, and then to 1,000 megawatt centers.

Microsoft said that it spent $20 billion on capital expenditures in the last quarter, not counting the stuff that it expensed to the income statement, such as the AI-related compensation costs. Meta said that it anticipates capital expenditures “of approximately $38 billion to $40 billion in 2024,” and that it expects “significant” growth of those capital expenditures in 2025, all related to AI. Amazon is on track to spend $75 billion on capital expenditures in 2024, up from $48 billion last year, and said it will spend even more next year. Amazon CEO Andy Jassy called AI a “maybe once-in-a-lifetime type of opportunity.” And then what?

But it’s not just these big companies that are spending the 10s of billions of dollars each on AI, and in ever larger amounts. It’s also less huge and smaller companies that are heavily spending on AI infrastructure, AI services, and AI-related compensation costs. AI startups are getting billions of dollars of funding. Then there are the utilities that are spending to provide the power to data centers, and construction companies building the AI infrastructure, AI consultants, AI conferences, AI whatever.

Companies are burning through their huge cash piles right now to do this, and they’re directing their operating cashflows into this, and some are borrowing to do this.

Since this spending comes from corporate balance sheets and from cashflows from other operations, they’re essentially recycling that cash that they would otherwise have invested in T-bills and other securities and they’re plowing it into the economy, and it begins to circulate.

AI workers and construction workers and the people that make their sandwiches are spending their income, and people and companies that receive that spending are also spending it, and the cash that was stuck on corporate balance sheets begins to circulate, and operating cashflows circulate, and borrowed funds circulate, and it all circulates and circulates and creates economic activity, from restaurants to freight companies that ship the AI servers.

This is the Dotcom Bubble all over again, but bigger, and spread wider. This amount of spending that is circulating around the economy is very stimulative.

And what came out of the Dotcom Bubble was an immensely useable internet, where nearly everyone has broadband, even on their cellphones, and a lot of economic activity has moved to the internet, eventually decimating entire activities and replacing them with new activities.

But when the huge amounts of investments and spending by companies and venture capital firms slowed, and the slowdown circulated through the economy, just as the cash had circulated through it, the Dotcom Bust ensued, amid massive job destruction in internet-related tech.



During the Dotcom Bust, the affected companies cut spending, and those that went out of business stopped spending altogether. Tech workers saw their wealth vanish when their company’s stock went to zero or near zero, and then they lost their jobs, and they cut spending.

And people who weren’t in tech, but with lots of money in the stock market got beaten up, and those that were leveraged got crushed, and they cut spending too.

This reduction in spending and investment circulated, just like the cash had circulated, and it triggered the recession from March 2001 through November 2001. Ultimately the Nasdaq plunged 78%, and the S&P 500 50% over a two-and-half year period from March 2000 through September 2002. It took the S&P 500 13 years, and the Nasdaq 15 years – plus trillions of dollars of money printing – to surpass the March 2000 highs, and lots of stocks just vanished. But the internet has become an economy-changing force.

What could cause the next recession? AI spending too will slow someday, and then the hangover starts. We’ll hear about it during the earnings calls from Meta, Alphabet, etc. They’re already under pressure to show results and slow down this AI-related cash burn. And we’ll hear about it during the earnings calls from companies that sell the chips and servers.

And when Corporate America slows down AI-related capital expenditures, and “right-sizes” AI-related staff and other spending, it will circulate through the economy, leading to slowdowns at the companies that received that spending, and it circulates from there, accompanied by a significant selloff in stocks that whacks people’s sense of wealth, causing them to cut spending as well, even if they’re not in tech. And that may be when we finally get our recession.

But this stuff takes years. Until then, that kind of widespread spending from corporate balance sheets and cash flows that circulates and circulates is very stimulative of the economy.

Enjoy reading WOLF STREET and want to support it? You can donate. I appreciate it immensely. Click on the beer and iced-tea mug to find out how:

Would you like to be notified via email when WOLF STREET publishes a new article? Sign up here.



  186 comments for “We May Not Get Our Recession until the AI Spending Bubble Implodes

  1. Random guy 62 says:

    Chatting with a friend last night who works maintaining nuclear power plant turbines across a half dozen plants. He said the mood at work flipped overnight from “don’t spend anything” to “whatever it takes to ramp up production”.

    He said they were recently harvesting salvage parts from three mile island, basically treating it like a junk yard car, and now it is going through a total recertification to produce again, at great expense.

    Data centers and AI power demand are driving the shift in investment mentality.

    Our local nuke plant has been buying up property aggressively in anticipating of increased demand too. Word has it there is an expansion in the works, with talk of a big new data center nearby.

    • Doc says:

      Bitcoin farms consume a substantial amount of power. As DJT seeks to replace the Federal Reserve with Bitcoin, we will need an ample energy supply to keep the network operational and online. This absurd idea was shared with me by my boss, who doesn’t have a credible opinion on any topic. So take it with a grain of sodium chloride. Circulates and circulates!

      • tom20 says:

        So, Microsoft, Google, and Amazon are big into Bitcoin farming?

        • Doc says:

          Ultimately, the AI gold rush is reshaping the energy landscape and forcing both tech giants and cryptocurrency miners to adapt to a rapidly changing environment. As the demand for AI and cloud computing continues to grow, the race for energy resources is likely to intensify, with significant consequences for the global economy and the environment.

      • spencer says:

        That would destroy the black market, the underground economy. And it’s too big to ignore.

        • ru82 says:

          Trump met with BTC miners and said he would like all BTC mining to be in the US. Think about that. It would now be a way to sanction BTC which has been used by rogue countries bypassing US Dollar sanctions. In addition, crack down on the cartel money laundering. Brilliant.

        • Escierto says:

          Using Bitcoin mining to sanction countries using it should be the LAST thing any crypto fan should want. The day that happens Bitcoin would cease to be an alternative to the dollar and immediately crash. The appeal of Bitcoin is the fact that it is NOT controlled by any government.

        • sufferinsucatash says:

          Bitcoin is a grift.

          FTX proved that.

      • WB says:

        “Bitcoin farms consume a substantial amount of power.”

        Yes, so completely unsustainable in a capitalist system. In a free market, all that energy would go to something with true demand, like food and housing. The level of moral hazard that has been unleashed into the world is truly unprecedented. “Free money” (i.e. low interest rates and ZIRP) leads to capital misallocation and malinvestment. Bitcoin is one such example. Paypal still works just fine for cross-border payments. Thank Elon.

        • Wolf Richter says:

          “… true demand, like food and housing”

          LOL, after you have food and housing, what do you do then???? Run around naked and twiddle your thumbs?

          You would not be able to post this comment without “true demand” for services, such as for internet service providers, cellphone operators, layers and layers of software to run it all, and “true demand” for goods, such as electronic equipment, including your device, and servers that run sites like Wolf Street, and “true demand” for more services such as the layers of software required to make the Wolf Street server work, and then on top of it the services that yours truly provides as part of his business??? JUST SO THAT YOU CAN POST THIS ONE COMMENT.

        • Biker says:

          I agree. Wasteful by design.

        • WB says:

          LMFAO!!! Touch a nerve Wolf? You didn’t address the point. Paypal works just fine, no need for bitcon.

        • WB says:

          What do you do?!? LOL! Just about anything you want! Innovation requires energy, so does building things. Very disappointing, but not surprising for an old paper-pusher.

          The point is, there are millions of much better uses for that power/energy. Much better than people getting angry and starving in the streets, but maybe you disagree.

      • RSDallas says:

        Bitcoin is never going to replace the dollar or the Fed (as much as I hate the Fed). Bitcoin is an extremely high net worth status investment. It is going to burn down to nothing, just like most investments did in the dot com explosion.

    • sufferinsucatash says:

      I don’t see why they do not just do solar panels in the desert with lots of lithium to store it.

      But perhaps too costly or not enough power. Still, lotta free power there.

      • Kent says:

        The cost of the infrastructure to move that energy to where it’s needed is significant.

      • Anthony A. says:

        Not really free power, plus solar panels get damaged, need cleaning, have required maintenance, need occasional repairs, take up huge land area, and do not generate much power on an individual basis.

        It’s a nice idea, kind of like a perpetual motion machine, but not cost effective or typically practical. And at night, not much happens without very large storage batteries that have needs too.

        • Wolf Richter says:

          All power plants — nuclear, fossil fuel, hydro, wind, solar, biomass, etc. — have large construction costs, equipment costs, maintenance costs, etc. with nuclear being the most expensive to build by far. But with wind and solar, at least the “fuel” is free and already here, ready to be used, delivered free of charge to the plant. Why not take advantage of it for some of the power needed?

      • Matt B says:

        They are building a ton of solar out there. You see a lot of it between Las Vegas and LA, plus the wind farms. CA also just opened the Ivanlah solar thermal plant there – they’re able to run overnight off of the residual heat collected during the day. Those plants incidentally look more sci-fi than anything the tech bros are building – pretty cool to see from the highway.

        At certain times of year the entire state of CA runs on renewable (includ

        • Matt B says:

          omg…. including large hydro.

          It is NOT fantastic however that we’re now going to be wasting all of this power on AI.

      • John H. says:

        Except for the initial capital costs, annual maintenance costs, transmission costs, and back-end site restoration costs, the “free power” idea is very attractive and timely! /s

      • Tina says:

        It’s not like “there’s nothing there” these solar farms destroy local biomes-so no moral superiority vis an vis biofuels as their proponents like to assume. And solar is not all that great-nothing beats biofuels and nuclear. Certainly not the windmills…

        • Pea Sea says:

          This is correct. Giant desert solar (and wind) farms can be incredibly environmentally destructive.

          But that’s more of an argument against huge, remote solar farms than it is against solar.

        • Mirage Owner says:

          Biofuels are incredibly inefficient to convert into useful energy. Photosynthesis only converts about 1% of the energy from sunlight into plant mass that can be converted into fuel. Solar panels typically have conversion efficiencies in the mid 20% range. Even wind turbines are far more efficient than photosynthesis and both use far less water, no fertilizer, and don’t deplete the soil of nutrients like intensive farming for biofuel crops.

          On top of that, there are then the conversion losses when crops are transformed into fuel and then are burnt in highly inefficient combustion engines that get maybe 35% of the energy consumed converted to useful work compared to 90% efficiency or higher for electric motors.

          It’s far more efficient and better for the environment to burn natural gas and make electricity for EVs than it is to use biofuels for transportation. Nuclear or renewable energy to produce clean electricity are the best options for the environment, but just about anything is better than biofuels.

        • Wolf Richter says:

          Biofuels are great if the biomass comes from what would be waste otherwise. Papermills have been doing this for a long time with their own waste. Feedlots can collect the gas from manure (digester gas, which includes methane) and supply it to powerplants or use it themselves for heat and power. Agricultural waste can be turned into biofuels. Old cooking oil and fats from restaurants and homes can be processed into biodiesel (for example the city of San Francisco does this for its fleet of biodiesel buses (the new ones are biodiesel-electric hybrids). They run on B20, which is a mix of biodiesel and regular diesel. There are lots of applications of turning organic waste into energy. They work great.

          But growing food explicitly to make energy is terrible nonsense.

      • Sc73 says:

        Solar fields need cool air to be efficient, fresh water to clean and maintain dust free surfaces for light access. Power converters need less ambient temp and low humidity to avoid break downs. Finally grid infra to evacuate power. So long story short, deserts aren’t the best choice for solar.

        • sufferinsucatash says:

          In my best Monty Python type voice:

          Do they not have car washes in Vegas?!

          Does the casino bird not fly south?

          He’s using coconuts!

          Hehe

      • ShortTLT says:

        Data centers run 24/7 but the sun doesn’t shine at night or on cloudy days.

      • Matt says:

        Not enough energy. That’s why Google is getting into nuclear investment for powering data centers. Plus thet s the whole ecosystem in the desert that should be left alone.

      • Mirage Owner says:

        Lithium ion batteries tend to be expensive and have difficulty reaching the energy density required for long term energy storage(by this I mean several days or more, 4 hours is about the typical limit they plan for with lithium ion batteries).

        Newer battery technologies, like the iron-air batteries Form Energy is now starting to build for utilities, will be truly transformative for renewables. Since iron-air batteries don’t need expensive minerals, have a much higher energy density and are designed to output maximum power for ~100 hours at a time, they can be built to very large sizes affordably, allowing renewables to compete with sources of baseload power like coal or natural gas.

    • John says:

      Do we have a list of stocks or index that would be potential to short this AI frenzy?

      • crypt currency says:

        It can be tempting to try shorting from time to time, especially ridiculous stuff like bit coin at eighty eight grand, but then I remind myself that that involves wall street products and any short position is a target for them to try and squeeze so the way to win is to just not play their crooked game.

  2. mike says:

    question: where does this kind of boom leave the 10 year treasury yield rate? Does it go up or down?

    • B2321 says:

      It will increase constantly, and there are technical indicators and fundamental reasons that suggest it will test the 5.6% around the beginning of May.

      • WB says:

        LOL! How’s that going to work with the current debt? Please, educate us on these “technical indicators”…

        I do not see CONgress acting responsible, and balancing the budget, anytime soon.

  3. PF says:

    SMCI to be delisted soon?

    Aren’t they one of nvda’s top ‘customers’?

    • AiHasBeenHereFor10Years says:

      SMCI is a systems integrator just like DELL, HP, or countless others. They sell NVDA’s wares in a consumable product to end users, typically in the form of servers.

      NVDA has a long list of those partners, and any gap will be happily filled by them.

      Accounting malfeasance is what is going to kill SMCI, and I don’t see that having any impact on Nvidia.

  4. Home toad says:

    Many years on getting our recession is a bit much, I want the recession to happen sooner.
    Lots of good deals on display with a good recession, have your cash ready and be ready to swoop in for the deals.

    • Anonymous says:

      The mentality of waiting on the sidelines for a recession to scoop deals is how you miss bull runs

      Wolf has written extensively about this. There is plenty of money sloshing in the system. If there is even a pullback it will be bought quickly by other lemmings seeking discounts

      Just buy NVDA right now. It ain’t getting any cheaper

      • Wolf Richter says:

        “Just buy NVDA right now. It ain’t getting any cheaper,” says the guy that only posts to hype NVDA to make his shares go up so that he can sell them to you at the peak and let you take the loss?

        FYI, It’s cheaper right now, -1.5%, not much of a discount, but still.

        • LT says:

          At times, I suspect there are stock market players that are looking at the cash in T-Bills and money market funds as their exit liquidity.

    • Biker says:

      You can study what Buffet said, I know he said many things.

      But should be easy to google article when he was asked to invest now or wait for a major crash. He said strongly: don’t wait for a crash. He compared this to being a mortician waiting for a major epidemic to kill millions. A bad business.

      So, the only sane approach is keep investing all the time aka cost averaging. It always worked OK.
      Also, related. WB said that you should not invest in stock if you are not prepared for 50% drop. So both these make axioms for investing.

      • Chris8101 says:

        Make sure your asset allocation is correct for your age, your wealth, and your responsibilities to your family.

        The past 4-5 years have caused my equity positions to grow bigly, so I began slowly selling a few months ago.

        Buffet selling also influences me to reduce my stock market exposure a bit.

    • Biker says:

      A summary of what I posted above: Time in the market beats timing the market. This could be printed and posted above your bed.

      • sufferinsucatash says:

        Some cfp did a study,

        Even if you had invested at the height (or near?) of the dot com bubble. Like you received a windfall from grandma Burl on your dad’s side, you slap it in (at a really inopportune time).

        Your balance today would be 350-400% up.
        (Think this was discussed on YouTube in April).

        So yeah best not to time the old market too too much.

        Then again we could have a Great Depression type fall or even our whole civilization could falter. it’s all baked in to the odds. But you hope not.

  5. Redundant says:

    Re: “Companies are burning through their huge cash piles right now to do this, and they’re directing their operating cashflows into this, and some are borrowing to do this”.

    Fine article!

    One company not getting overhyped on AI cash burn is run by Uncle Warren B, who depends on cash flow reality.

    I totally get the argument and theory that AI is a perpetual future cash engine and a wormhole to a stimulated economy — but, after reviewing the Buffett Indictor with an electron microscope — I came to a conclusion, similar to my bitcoin outlook, which is — nothing in the world grows faster than the aggregate cumulative economy — GNP.

    Hence, the Buffett Indicator, informing us that the aggregate growth of the entire total market cap (overvalued) at $58 trillion will not sustainably grow faster than our $29 trillion GNP economy. Warren is selling because the current valuation levels are insane.

    Yes, there are booms and pops and mini tech spikes, but these cycles never are sustained.

    The internet bubble, dotcom bubble is a fine example of a stimulative boom that took years to ramp up, then crash, with an assortment of BIG winners that survived , who continue to dominate with monopoly power.

    Nonetheless, The Greatest example of the entire dotcom bubble is Cisco. If you would have been super smart and bought a few shares of that in the late 1980’s, you’d still be up like around 36,000% — but, if you were like the vast majority of people that got excited about it near the top of its glory days in 2000 — you’d have basically gone sideways for 25 years, wondering why.

    That actually is a great example of how time can distort performance.

    There’s a huge lesson there in perspective — primarily, if you get in something like nividia or msft or apple at dirt cheap levels, great, you probably can survive for decades of volatility and break even or better.

    I realize Buffett is ancient — but there’s a reason his indicator is extremely useful — it hilights times when excess exuberance is causing cash to be burned at accelerated rates — which more than likely, will never create sustained cash flow.

    In the near term, there’s a vast tsunami of cash going into AI — it’s undeniably seductive — but that temptation is likely to blow up, because AI is not bigger than the aggregate economy — money flowing into one area, flows from somewhere else — and it’ll be the unexpected imbalance that throws this perpetual machine into a nose dive.

    I think the last time I saw these types of profit margins was in bank balance sheets in 2021, SVB was bullet proof too.

  6. Dark Artist says:

    AI may be a once in a lifetime opportunity, but it may not be in our lifetime that the fruits of it come about.

    • AuHound says:

      @Dark Artist

      How true! I made a fair amount of $ in the mid-80s investing in “neural net” chipmakers (AI in hardware instead of software), and then lost it in the 87 stock market crash. Not one chipmaker survived.

      Of course this time it will be different! /s

  7. JK says:

    Wow. Really great analysis. I remember the Dot Com crash. I got clobbered and went from trading back to working my trade. Everybody is a genius until you get caught up in a crash and realize you’re not as smart as you thought, but a regular schmuck like everyone else. You’re absolutely right, the internet was an awakening that opened all kinds of possibilities. I use it daily for information at my job, purchasing, reading this blog, etc. When will we hit a top? We’ll have the answer after it happens.

    Really awesome article.

  8. BobC says:

    Its not happening as long as the government keeps spending $4 trillion per year more than it collects

    • Wolf Richter says:

      $2 trillion. The whole budget is a little over $6 trillion. Revenues are $4.4 trillion. Deficit ca. $2 trillion.

      A lot of government spending, such as interest expense, foreign spending, etc., doesn’t contribute to the economy at all. But all of corporate spending and investment does, and it’s multiplied by the circulation of the funds.

      • B2321 says:

        The total spending of Ai capital investment is excepted to be attracted the maximum around $200B, and a significant part of it goes to the TMSC and their employees. While the federal deficit net of debt service is around $1T, which is multiplied in the circulation same as that AI spending. It’s curious that you belive that less than $200B is more effective in boosting the aggregate demand than $1T!! Really interesting math you use…

        • Wolf Richter says:

          LOL, RTGDFA (why do people keep posting here without reading anything???)

          “The total spending of Ai capital investment is excepted to be attracted the maximum around $200B,”

          1. AI capital investment already went way over $200 billion. They’re not waiting for you to catch up.

          2. What are the TWO TYPES of AI spending? Fill in the blanks (Tip: the answers are near the top of the article in bold):

          A._________

          B._________

          Tip for your homework: Capital spending is only one of them.

          3. Maybe you’re thinking about investments by venture capital firms into AI startups? But that’s just a small sliver of it. Or maybe you’re thinking about the Mag 7 AI capital spending only, which could be about right. But their second type of AI spending (RTGDFA) is likely much more than capital spending.

          4. TMSC building a chip plant in the US is not part of AI spending. It’s on top of it. That’s just construction & equipment spending on semiconductor fabs. If the plants will also make AI chips, then that portion would be AI spending.

          5. You need to revisit the difference in multiplier effects of government spending and business spending.

  9. BuySome says:

    Ha! Modern “Recessions”? People with cushy positions filling easy money cream between wafers who then recieve endless “extensions” of unemployment checks so they don’t have to get off their tush and flip burgers to pay the bills. Meanwhile, the ones who worked their backs into the ground for peanuts but managed to save up a bag are rewarded with dilution of their face value by banksters who ensure they can never exchange it for anything of real value without having the government colonoscopy camera peeking up their a-holes and taking notes on how best to cornhole them. Nah, this won’t end until a new Sherman arises to march to the sea and burn (AI)tlanta to the ground again. Screw ‘em all. Make Money Great Again. Kancel “their” Kulture.

  10. James says:

    Your article is a stupendous way to start a new week Wolf! You really lay it out what the macro, mid & incoming-immediate picture in our always evolving evonomy!
    Today should be a wake up for those who were not cognizant during DOT com bust 2001.

    My question is will everyone need their own generators or Small Nucleur Reactors shared with like minded folks because the Mag 7 & other too big to fail Corps all over the world will price the hoi polloi out of what used to be a functioning market?

  11. Adeel says:

    Wolf…..if you were to take a guess….. how many years….. will the current Upswing in AI spending Last ??

    • Wolf Richter says:

      Someone else take a guess 🤣

      • VintageVNvet says:

        According to the conjunction of Jupiter with the other Great Spirits, especially Venus, WE, in this case the GLOBAL WE are already far into World War 3 that will be continued by the AI Hun Robots after the radiation wipes out all live beyond amoeba level, and that will take precisely 3.76853 years MOL…
        While certainly open to challenge, LOL, that guess is just as good a guess as any other, as it should be clear to anyone with any cognition at all that NOBODY KNOWS.
        We can at least hope that someone who claimed recently NOT to have ANY cognitive will find some, eh

      • TAP says:

        My guess is that the upswing is going to last a long time. At least 10-years and likely much much longer. Here is the thought process:

        Challenge: Traditional Data Centers (Edge) are growing at 10-12% per annum. Add in AI and it is much higher growth rate. The obstacle to higher growth is the entire USA market is short reliable base load power generation sufficient to handle the growth.

        Generative AI Data Centers require 8-12x more power than a traditional Edge data centers. So if an Edge DC requires a manageable ~20MWs of power for a Utility, an AI DC requires ~200MW of power, and that probably immediately makes them the Utility’s largest customer.

        Logic: A Data Center takes 24 months to construct (not including development time – siting, permitting, etc). And there is currently a backlog on equipment of 18-24 months (servers, chips, etc) . . . a power facility is a longer construction timeline (30-40 months, depending on where constructed) and the development timeline is longer for a power facility (1-3 years, depending on location and prior to commencing construction). That is, if you can get the equipment (turbines, switches, transformers, etc – which are currently backlogged ~30 months). And it takes years and significant CapEx for a turbine manufacturer to add production lines. The timelines mentioned are typical for “fast moving” independent power producers (IPPs). A typical Utility “planning cycle” is 8-years. Utilities do not move quickly and are not really sure to make of these Data Center folks. IMO the Utilities alone cannot handle it all and growth will be augmented/driven by the IPPs.

        I know the “silver bullet Nuke” comment is coming from someone, so I will address now. The latest hype are small medium sized reactors or “SMRs” which is unproven equipment, many from new unproven manufactures – that means they are not financeable by the traditional finance markets and this expensive tech will be unfinanceable in the project finance market for many years. So SMRs are not a real solution on their own. Sure maybe some risky tech companies will fund some of this unproven equipment with full equity commitment – god help them and their shareholders. But do you want a Nuke in your backyard or next to your kid’s school? It may be zero carbon, but as soon as there is a problem with a Nuke, it is no longer “green”. Just ask the residents of Fukushima their opinion. Permitting a Nuke is a nightmare. At a minimum, a new SMR will likely take at least 10-years to permit and construct, probably longer. And they produce less power than a traditional, proven, reliable and much cheaper combined cycle gas turbine (CCGT), which megawatt for megawatt is much faster and to construct than any other type of equipment available. Gas turbines will not be the only answer but it is the most reliable base load power to augment intermittent renewable power – which are also needed and required. We will need all types of power to facilitate the build out of the Data Center boom.

        So will this all happen? Yes. But it is going to take time. The USA currently has ~50% of the Global Data Center market. The rest of the global data centers are . . . everywhere else. The boom is happening everywhere. However, if the USA wants to remain a leader in data/information, the biggest impact will be in the USA.

        No matter the country, the practicality of speed of construction, will be the bottleneck for Data Center revolution. It is a boom, and it is a hockey stick boom, but the limits of constructing new power will extend/[moderate?] the growth over at least 10-years, probably longer. At least that is my guess.

        • Nemi5150 says:

          You are only looking at half of the picture. The ‘supply’ half. Your unwritten assumption is that ‘demand’ will stay the same for the next 10 years. If demand dries up before then, all of your calculations go out the window. This is what Wolf is talking about. Eventually stockholders start pressuring companies to ‘stop the spend’. When that happens, the whole house of cards comes crumbling down.

        • Anthony A. says:

          Great post, and confirms my feeling that natural gas will be in demand for quite some time fueling these generation facilities.

          That’s good news for the producers and transmission companies.

        • ShortTLT says:

          IMO, combined cycle natgas plants are the only feasible power source for these data centers in the short term (next decade-ish).

        • SteveG says:

          Much too gloomy on smr’s. The Trump admim., if they do anything, will reduce regulation and permitting . This will bring down costs hugely. The funding will come as well, just as it miraculously materialized for fiber cable, the inet, biotech startups and R&D, etc. The grid. in many respects is much harder to achieve. Material, i.e.copper cable, transformers etc. along with much greater permitting challenges, i.e. nimby, environment, etc.will be difficult but data centers will avoid this by colocation. Yes, there will be data data centers just as there was permanently dark cable, but no big deal

        • windowpane says:

          TAP, Holy S

      • Louie says:

        It likely ends when there is a realization by these companies that monetizing their efforts will not work out as they planned. That’s kind of how these events end historically.

        • Franz G says:

          100% agreed. investors are already starting to ask what they’re getting for this spending. the “put up or shut up” moment will come sooner than people think.

        • Matt B says:

          I’ve argued broadly before that without some kind of hard grounding in the same reality that humanity lives in then it’s usefulness is going to be pretty limited. Apparently the plan to overcome that is to just keep throwing more data at it, but I think the result is going to be a kind of virtual Kessler syndrome rather than some kind of emergent intelligence. I can’t see how this is going to end up being ” the next internet”.

          But of course a lot of this investment is probably just bubble dynamics – it’s impossible to know how much the investors themselves believe in it.

      • Andrew pepper says:

        Perhaps something as great as or greater than AI will come along in the interim and we will never get that recession. With enough innovation we may become recession proof! What a thought.

      • Aman says:

        While the impact in the real economy will last sufficiently long, the patience of investors will run out sooner.
        They will start to question the returns from the investment. Then the following will happen
        1. Stock prices fall
        2. Negative wealth effects reduce consumer spending
        3. Investments slow down and earnings drop
        4. Repeat 1-3
        5. A recession

        While slow down can cause stocks to fall, falling of stocks can also cause a slowdown.

        But as WR said, it is not possible to estimate the patience of investors or to time it.

        But non-tech economy is showing sufficient signs of normalcy. In aggregate things seem to be fine.

      • Ross says:

        Two to three years. That’s my rule of thumb for how long it takes everyone on the planet who might be interested to “sample” a new product or technology. Either it will keep growing (Internet) or collapse to a mere shadow of it’s hype (WeWork, Groupon). For the current crop of tools that people label AI (LLM, deep nets, etc) I predict the latter, because these tools are merely statistical prediction engines. The are not general intelligence, which is what most think of when they hear “AI” The tools will prove useful in a few niche applications, but Skynet or Colossus they are not.

    • Biker says:

      Hard to tell. But also emerging sub trends that can be pushing it more forward. The generative AI should be peaking up pretty soon, say a year. But there is a race for General AI, AI in robotics etc. Only if available resources could support all that.

      • Nick Kelly says:

        Will the companion doll of the future be an automaton or a robot? Will the chit chat be on board AI or will it be via link to remote AI or human?

    • Russell says:

      Adeel – Ask ChatGPT…

      • Russell says:

        Adeel – Did the legwork for you.

        ChatGPT – “Ultimately, the rally will likely persist as long as AI continues to show economic potential and investor sentiment remains positive. However, most analysts expect increased volatility and potential pullbacks in the coming months.”

  12. Kentucky says:

    Yes, AI will help companies, like Dell, do away with call centers that answer the same mundane questions over and over. But, will it really, exponentially benefit our lives on a daily basis?

    Do I really want to send or receive a thank you card written by AI? I feel like we, as a society, are quickly becoming domesticated farm animals getting fattened up for slaughter. Food is effortlessly brought to us in our barns via apps and we stare at screens instead of seeking daylight. Ultimately, if truly “successful,” AI will skin its clients – not shear them. Akin to the tiger eating its tail.

    I think it won’t take years for the bubble to pop to find out the AI king isn’t wearing clothes from a direct consumer standpoint. See etoys.com.

    At what point is frictionless living not living? AI is going to run into the tech old-age problem of fixing problems that weren’t a real problem.

    • Kent says:

      AI has the capacity over the long run to produce machines to cook your food, clean your toilets, rotate your tires, etc… It is much, much bigger than lots of people are thinking, which is why there are billions being poured into it. But you’re right. Just like most folks wouldn’t have a clue how to successfully grow their own food, folks will forget how to make their beds.

    • Brian says:

      The benefits of AI are not text generation. It’s the advancement in thinking that is to come. Already it solves protein folding which greatly aids medicine. Soon it will do complete medical diagnosis better than Dr. House and team.

      But that’s peanuts. Next will be designing the next generation of AI systems. That’s when things really take off. Human intelligence is fixed, though our knowledge and tools are cumulative. Machine intelligence has no such bounds. It’ll be a true intelligence, different but not “artificial” and 10s, 100s, an eventually millions of times smarter than ourselves. It won’t just find answers beyond our comprehension; it’ll discover and solve problems beyond our comprehension.

      That’s the future of AI… and it’s within decades, not generations.

      • Jeff says:

        Agreed. The question then becomes “what happens after AI – what will be the next big thing?” I believe the answer is NOTHING – then what happen?

      • Nick Kelly says:

        I doubt it will think ‘outside the box’. Example: in 1905 fundamental discovery in theoretical physics was considered to be over: job done. Only technical detail needed to be added.

        Then a 26 year- old patent examiner in Switzerland , second grade, who had just been denied promotion, submitted a paper in German that overturned the whole field. It would have received an F grade from the AI examiner. Its idea that even time was relative, and affected by gravity, might have led to a psychiatric referral.

        • phillip jeffreys says:

          And even that intellectual giant got it wrong on the cosmological constant.

          Science, very often, excuse the license taken here, makes up **** to explain the gaps in theory/prediction and actual measures. The measures themselves can be proxies. After the fact, itt seeks confirmation. Chicken/egg kinda thing.

          Did the Universe expand faster than the speed of light? Scientists say it did during so called “inflation”.

          Don’t know why macro predictions aren’t always correct? Must be dark energy!

          How many sub-atomic particles are known/postulated at this time? 32 last time I checked. Does this somehow conflict with the dictum that complexity is always the least likely answer? Whjy less anti-matter than matter? “Just lucky I guess”!

        • Brian says:

          “I doubt it will think ‘outside the box’.”

          Of course it will. How do I know? Because it already does! Machine intelligence has already shaken up every field it has touched from medicine to the game of go.

        • Wolf Richter says:

          Brian,

          It “hallucinates” outside the box (makes up stuff), which is why AI is so unreliable. But then humans do too, so I’m not sure it makes any difference.

      • VintageVNvet says:

        Brian,
        Your statement, ”Human intelligence is fixed,” is absolutely false!
        Human intelligence is only limited by our own individual choices and genetics, and IMHO mostly by the former rather than the latter. BOTH can and will change, as in, “The only constant is change.”
        When and only when WE, in this case our species, choose to increase both our individual and collective intelligences – and contrary to popular thought, there are many as has been postulated beginning with Gardner et al many decades ago – WE will do so and probably very quickly relative to the general timeline of the past several millennia.
        The internet and WWW has certainly helped this process already, and the end results of AI are very likely to help even more sooner and later.

        • Biker says:

          We are getting dumber and dumber every last few generations, a study stated.

        • Brian says:

          Genetic change is so naturally slow that it might as well not exist in comparison to everything else. The rest is “knowledge and tools” which I already mentioned.

          We may begin to grow our individual raw intelligence with genetic alterations but only after much exploration and refinement… using knowledge and tools.

      • Coil says:

        Yes, but will it ever be able to tell me what happened to to the other sock that went into the dryer and never came out?

        • ApartmentInvestor says:

          @Coil your missing sock is probably under the drum…

          I just took apart a dead coin op apartment dryer and found a bunch of socks. Recology won’t let me put “appliaces” in my two yard dumpsters and taking apart a dryer with an impact driver and reciprocating saw to dump “random pieces of metal” in the dumpster takes less time than going to the dump…

      • Anthony A. says:

        This all sounds like a load of crap to me.

        I’ll believe it when AI can give me the winning lottery ticket numbers a week in advance. Now that problem and answer is “beyond our comprehension”.

      • Franz G says:

        lol. i always find these type of posts entertaining.

        look, technology will continue to advance. that’s been the case since the first primitive humans walked the earth.

        i haven’t seen any evidence that ai is anything special.

        • Brian says:

          People have advanced technology since we first appeared. But soon, technology will advance itself. That is a BIG shift.

      • Matt B says:

        I’ll believe it when I see it. Meanwhile, the latest news we have on how that’s going with basic audio transcription:

        “Although Whisper’s creators have claimed that the tool possesses “human-level robustness and accuracy,” multiple studies have shown otherwise.

        In one study of public meetings cited by AP, a researcher from the University of Michigan found hallucinations in eight of every 10 audio transcriptions. Another machine learning engineer reported hallucinations in about half of over 100 hours of transcriptions inspected. A third study identified hallucinations in nearly every one of 26,000 transcripts generated using Whisper, AP said.

        Microsoft, which offers Whisper as part of its cloud computing services, advises companies incorporating it in the solutions they offer to ‘obtain appropriate legal advice to review your solution, particularly if you will use it in sensitive or high-risk applications.'”

  13. Michael Engel says:

    The nerd-7 are all $3T. $30B AI capex : $3,000 market liquidity = 1%.
    CL rise will deflate Europe and the US in frog cooking. Putin and Iran
    will subsidized the BRICS nation. They will herd together, cemented to
    each other. When the nerd-7 dihydrate their investors will realize that
    the nerds produced bad items

  14. sine99 says:

    I’ve posted on the dot-com bubble/crash here before and I’m glad Wolf wrote this article. I see a lot of similarities between then and now as well (and a difference or two). I’ve been saying this for a while, so I guess I’m a “perma-alarmist” on this topic because this party just keeps going and going, admittedly beyond what I thought it would.

    But let me say a few simple things.

    One, the most recent run-up is bringing the SP500 very close to valuation levels not seen since the top in March 2000. Do your own research, make your own conclusions.

    Two, my number one take-away of the March 2000 top was that virtually every bear disappeared. This one is interesting, because there are still a few bears left that still exist right now the way I see it. Look at price of gold lately. (It’s interesting to research what gold did it in 1999. Spoiler: It wasn’t acting the same as it right now.)

    Three, I’m a nobody. So instead, look what Buffett’s been doing lately. Look how he did during the “dot-com crash” era. Look at Berkshire’s performance in the wake of that crash. And look what he’s doing now.

    I am not predicting the future. Instead, I take a more mathematical look at this. I look at the data, as things stand now vs then. I have NO idea what humans are going to do. I think there is a fair possibility that humans take this bubble further than the dot-com bubble, possibly WAY further judging from how people are acting right now. But the fact is I don’t really know.

    • Harry says:

      I believe the main difference is that during dot.com there were dedicated companies dealing with it. With AI this is driven mainly from large companies with deep pockets and AI operations is a small fraction of their operations.

      Take alphabet: Annual free cash flow of 70b. Invest in AI approx 30b per year.

      On the other hand, those big techs were hoarding employees and office space as WR pointed out. After releasing such resources they have free cash flow increases paying 50% of AI investment. A large bet but when the bet is not successful none of these companies is on risk.

      And when they are successful they will be able to sell / rent their AI models to SMB companies which will have the largest benefits.

      • Franz G says:

        harry, this is a fair point, but one thing to consider is that if the general economy starts to slow, as i believe it has, the cash flow for the bigtech machines will drop. they are heavily reliant on cloud services and advertising, both of which dry up when that happens.

    • Franz G says:

      the other thing is that people like to point to the fed as justification for this bubble.

      but look at it. wolf demonstrated in his last article that the fed’s balance sheet is back to where it was over 4 years ago. in may of 2020, the s&p was half of where it is now, and that was with interest rates of 0% and $120 billion a year of printing.

      that is now gone. this is not the fed. this is pure bubble mania, which has existed for hundreds of years.

      • Bobber says:

        I think the bubble is expanding because people believe the Fed will come to the rescue with QE in the face of a significant asset price drop, rendering all losses temporary. It doesn’t matter that the Fed is applying QT now.

        What will the Fed do when faced with crisis like the 2000 stock price drop? The answer will be unknown until the next test arises.

        The article implies the Fed might stand by and let stock prices drop. That’s one realistic outcome and the most fair outcome in the face of the next economic challenge, but it’s an austerity outcome. Will our “leaders” accept austerity outcomes when opportunities for money printing and rate suppression are available? I have my doubts. Until I see austerity hit, I’ll never make an unhedged bet against the willingness of “leaders” to sacrifice long term health for short term gain.

        Over the past several decades, authorities have intervened more and more to dictate pricing and outcomes. Instabilities have built up as a result. Even the Fed admits our current fiscal path is unsustainable, which implies current asset values and debt growth is unsustainable, given the interrelationship of everything.

        I will not go out on a limb in this environment. I maintain a barbell investment strategy that hedges all outcomes. Monetary profligacy v. austerity/recession. Inflation v. deflation.

        • crypt currency says:

          I guess some people probably they can but I dont think the fed can maintain bubble prices and I think money printing interventions actually make the inevitable bust even stronger and hopefully the money supply continues to decrease. I think austerity is the inevitable outcome whether we choose to commence it willingly or whether we have no choice. Countries that were once rich but became poor didnt want it to get to that but they eventually couldnt stop it.

  15. BigBird says:

    Wolf, in your opinion, would the deflation of the housing bubble be another possible trigger for the next recession? Not talking a banking crisis, as the banks have offloaded the risk. I just mean a consumer spending recession. Consumption is the biggest part of the economy, and when home prices decline, people start to feel poorer…

    • David S says:

      Not a trigger, but maybe an effect. You don’t see a frenzy of economic activity in housing. Not the mad house flipping, commission making, loan making, loan packaging like 2005-2007. If anything, there is already a big pullback there.

      The recession will come when frenzied economic activity slows. This article points out where that frenzied activity resides : in AI. The next recession will likely come when the AI frenzy pulls back.

    • Wolf Richter says:

      BigBird,

      Just to add to what David S said…

      In late 2007 and in 2008, as the housing bubble was deflating, corporate spending also began to decline — Cisco’s CEO John Chambers infamously called corporate demand “lumpy” in Nov 2007, which sent stocks reeling. Then came the banking crisis (mortgage crisis). So it was a big mix of things.

      The 2001 recession was a mild recession, no biggie. And the Greenspan Fed overacted to it and thereby created the housing bubble.

      So if the housing bubble deflates enough to create problems with foreclosures, it would be a contributor to a recession, but not a big one.

      Also, what I’m saying is that an implosion of the AI spending bubble may cause a MILD recession, of which we have had many, such as in 2001, and not a financial crisis type recession.

      • sine99 says:

        Wolf said: “The 2001 recession was a mild recession, no biggie. And the Greenspan Fed overacted to it and thereby created the housing bubble.”

        I 100% agree with that, but to state the obvious for any young folks reading this: Just because the recession was mild doesn’t mean the stock market correction was mild.

        To quote Wolf again: “Ultimately the Nasdaq plunged 78%, and the S&P 500 50% over a two-and-half year period from March 2000 through September 2002.”

        And…it wasn’t that mild for people who worked in tech, particularly in Northern California. But the good news is after just a year (or two), the jobs came back in force and the next Greenspan-induced boom was on…leading to the housing bubble, exactly as Wolf says.

  16. Sandy says:

    Hi Wolf. Based on this it seems sector specific equities in US will benefit whereas there will be limited benefit to international and emerging market equities like India, China, Brazil etc. Any thoughts?
    Thanks.

  17. sufferinsucatash says:

    One would think they are making the internet 3.0 with all this infrastructure.

    It’ll be amazing. But you’ll have to pay to play.

    Betcha our homes become addicted to it as well.

    Maybe the draw will be it anticipates our needs and does a lot of leg work for you.

    But who knows. Maybe our cars will run on trash too! Marty! :)

    • Dirty Work says:

      The next step in data center capacity should be 1.21 gigawatts.

    • Tony says:

      If you want to invest in something, invest in quantum computing and networks, that is the true future. No, it’s not science fiction anymore, and it solves the energy crisis. It may be a few years yet, but it’s coming.

  18. Flo says:

    Wolf question for you or anyone that knows…. Is there accounting rules for depreciation for data centres? Like how much creative accounting takes place for them you reckon? Cause being forced to depreciate diminishing value 20% per year would quickly show up on income statements, but I’m assuming they can do it much slower even if the asset is worth nothing in 5- 10 years?

    • Wolf Richter says:

      GAAP accounting rules offer some flexibility and choices for depreciation. And there are IRS rules. IRS rules specify 39 years straight line depreciation for commercial buildings. Land doesn’t get depreciated at all but is carried at acquisition cost. Equipment depreciation is much shorter, such as 7 years, but that varies with the type of equipment. And most of the value in a fully equipped and operating AI data center would be the equipment.

      I don’t remember all the GAAP options. Maybe a CPA with experience in commercial building and equipment depreciation (computers, electrical equipment, HVAC, etc.) under GAAP can weigh in here.

      • esop says:

        What about the capital account called Developed Software? Usually around 5 years depreciation. And then there are the accounts of Acquired Software that may be part of Goodwill from an AI company acquisition, these accounts are given certain years of life in Purchase Price Accounting rules. And all are subject to future write-offs when annually revalued and losses are sent through the income statement. Huge Goodwill write-offs occurred after the Dot Com bust, as well as zillions of lawsuits.

    • Franz G says:

      i believe it’s 5 years straight-line, but it’s been a while.

      as a disclaimer, i don’t know if the chips act or some other law implemented accelerated depreciation. i don’t believe it did, but who knows?

    • MM1 says:

      I don’t think there’s a ton of room for manipulation in depreciation. Buildings are buildings, computer equipment is computer equipment, furniture is furniture and so forth.

      Also there’s lots of easy less obvious ways to play with the number – like the new revenue pronouncements. Also when these big companies start investing in/funding little privately held companies with complicated hard to value financial instruments like convertible debt, warrants, etc the valuation becomes highly subjective.

      Not financial advice just my opinion.

  19. Kenny Logouts says:

    I’ve noticed Google already giving AI answers.

    So you either get adverts to businesses, or an AI answer which is now the aggregate result of all the answers on the internet, essentially… and could be convincing but wrong, or shallow as a puddle for insight.

    Oh progress. Very expensive “progress”

    Y2K seemed to trigger after the world didn’t stop on 1st Jan 2020.

    We just need a trigger. SMCI might be it.

    • RichardW says:

      I sometimes read the Google AI answers, but I wouldn’t rely on them for anything important. The other day it gave me a surprising answer, so I checked out some of the linked articles below, and confirmed my suspicion that the AI got it wrong.

      • Kenny Logouts says:

        Yeah I’ve just been doing some stuff and as it’s scraped from all sorts of sources where USA and UK relevant policy are mashed together (those generic websites filled with what feels like scraped data itself)… the AI answers are just a blurred mess of USA and UK policy/rules/laws.

        Then the rest of the Google results are just links to said websites often with amalgamated info.

        All these AI searches are going to increasingly bias themselves to their own biased training, as they regurgitate their own outputs.

    • Max Power says:

      In the existing web search model, search results lead users to sites and those sites collect ad, subscription, sponsorship and other revenues. However, this model cannot work under Microsoft, OpenAI, Google’s, etc. vision for AI.

      The problem Google and the other AI firms are eventually going to run into is that the original content creators whose data was used (or one could argue, ‘stolen’) to create the AI models have a disincentive to have their information shared with the models since the models don’t drive users to their site.

      Why, for example, would Wolf Richter want to allow OpenAI to steal his content to be used in an answer where no ad benefiting WF is displayed?

      Now I suppose the AI companies could pay content creators directly but if you thought building data centers were expensive, wait until you have to pay tens of millions of creators for their content… not to mention the extremely complex process behind it trying to figure out who should pay what and how much and the whole logistical process behind it. It’s a compete paradigm shift from how the web operates today.

      People are not paying attention enough to this issue until one day it’s going to blow up and upend the whole AI ecosystem.

      • Redundant says:

        “Elon Musk’s social network X just quietly gave itself permission to use everyone’s posts on the former Twitter to train Grok, an AI model built and released by X.ai, another Elon Musk company. It was first discovered by a Twitter user.”

        The motivation behind buying twitter as a training resource should make people question the value of ai data — I think this is the big picture big elephant, if all these ai databases and data centers all have basically the same scraped data, where’s the advantage of meta stupidity?

        So yes, will ai companies end up paying for useful data, or is ai going to assume it’s got all the best data — this entire concept is stupid. Predictive analytics may be responsive to specific queries, but it’s not FRIGGN magic !!!

        I know WR goes ballistic with EV and cars, I go ballistic with ai — lol

    • Escierto says:

      For this reason I have stopped using Google search. They can take their AI and shove it where the Sun don’t shine!

      • crypt currency says:

        Yes, same as when goo gull tries to get me to log in to read the text messages on my own phone. They can go straight to hell. Ironic their motto was once dont be evil cause theyve turned to the dark side with a lot of their invasive spying and trashy ad business.

  20. THEWILLMAN says:

    It’s a massive reshuffling and the billions in GPUs are being spent out of fear, not greed.

    For many companies it’s already started (see Chegg) – for others business is good for now but a very obvious existential risk is around the corner.

    A better version of everything from search to self driving cars to auditing public companies is about to be a commodity.

  21. Kent says:

    A recession may come along if investors don’t see the massive capital expenditures producing adequate returns. Already Microsoft has given up on charging extra for their AI interface (Copilot) to corporations. They’ve just rolled it into their Office 365 package.

  22. Dani says:

    What if this explosion of investment in AI actually delivers returns and then it is no longer a bubble? Is that scenario plausible?

    • sine99 says:

      Dani, this kind of logic is indicative of manias. AI is here to stay, just like the Internet was here to stay in March of 2000. But just because a disruptive technology emerges and is here to stay doesn’t mean any valuation on a business is warranted. Mathematics still matter. When markets throw mathematics out the window, sometimes in spectacular fashion, that’s the hallmark of a market bubble. As Wolf pointed out in another comment, the 2001 recession was mild…and two, the Internet stuck around. But the market correction was not mild. The math came back down to reality.

  23. Random50 says:

    If AI takes off in the productivity sense, our future looks really bleak.

    The mega companies will own the world even more than they already do, because the financial requirements alone form an unassailable moat to competition. They even have a secondary moat from their vast and private data resources.

    • Franz G says:

      they won’t. the nations of the world won’t tolerate that much power being concentrated in the hands of a few american companies.

      we’ll see antitrust enforcement like we’ve never seen before.

    • Jonzo says:

      Watch or rewatch 1975 “Rollerball”. The game is a good sideshow, but the premise is that Corporations have replaced Countries and run the entire world.

    • crypt currency says:

      Doubtful AI is going to take off much productively. “AI” cant and will never do manual labor.

      • Wolf Richter says:

        “AI cant and will never do manual labor.”

        Robots excel at doing manual labor. They do it faster than humans, 24/day without complaining and without asking for a raise or having to go to the bathroom. Whole entire factories are equipped with robots. And they’re all run by software, and some of that is AI or has been written by AI.

    • Ben R says:

      “The mega companies will own the world”

      No one will notice or care, because we’re already like 70% there and we’re taking it like collective champs.

  24. Viktor69 says:

    “This is the Dotcom Bubble all over again, but bigger, and spread wider” @Wolf, do you have any estimated figure behind this statement, like what % of GDP goes into AI in 2024 vs the dotcom investment in 2000?

  25. phillip jeffreys says:

    Wolf, in my mind, is spot on vis the the likelihood of Dot Com II – only at an even massively larger scale.

    Wall Street/business world constantly latch on to these memes that gather irrational momentum.

    The “some are borrowing” part of the equation raises the eyebrow and pauses one to consider the systemic risk.

    From a pure mathematical/probability point-of-view, I understand the mechanics of AI.

    And I still believe the ole saw “Life has a fat tail” holds true no matter the computing power and math thrown at a solution space.

    This is not going to end well.

  26. Mister E. Miyagi says:

    I can’t fucking believe this shit. Where the hell did I put my god damn car keys. I’m going to be lake for work. Fuck Brian in accounting. I’ve had it.

  27. Economic Armageddon says:

    The AI hype / gravy train is going to go on for as long as its promoters and evangelists can push it, because after AI…there’s no discernable ‘next big thing’ waiting in the wings.

    • phillip jeffreys says:

      Quantum Computing around 2030 will be ready for commercialization – so the pundits say.

  28. Redundant says:

    This sums up the importance of Buffett Indicator very nicely —

    “At its height, Cisco’s market cap was worth 5.5% of the U.S. GDP at the time. At press time, Nvidia’s market cap represents 11.7% of the nation’s current GDP, more than twice Cisco’s peak relative to GDP.”

    I think GDP will be revised down as nividia value goes higher — I’m absolutely convinced Warren’s recent selling of Apple is on the money. You can be super bulls and ignore valuation, but this nividia ai run is not sustainable.

    • Wolf Richter says:

      You’ve been saying GDP would be revised down for over a year. But it was massively revised up last month, going back 5 years, with the last 2 years being huge upward revisions.

      • Jonzo says:

        Regarding GDP, when do you think the US Gov’t will be forced to address the US National Debt by 1st running on a balanced budget ?

      • Walter Floeter says:

        Is there a difference between the AI boom and the dot com boom in that hopefully the AI boom is creating higher productivity there for real value where as many of the dot com companies did not.

        • Wolf Richter says:

          You may not realize this, but the dotcom companies figured out how to commercialize the internet. Many companies died trying it and took billions of dollars of investor money down with them. But what came out of it was the biggest productivity enhancer since the broad commercialization of electricity. And it changed the economy.

  29. Bob says:

    I guess “investing” in possible productivity gains from AI is better than using all their cash for stock buybacks. Lord knows they won’t pay a decent dividend to their shareholders. Comparing the impact of the AI bubble to the dot com bubble is a lot of speculation now. The massive opening of new communications and information sharing of the Internet built a foundation for productivity gains that continues to pay off. AI doesn’t seem to function at that lower level of fundamental transactional communications , it seems more like icing on top of Internet cake. And maybe a lot of sprinkles.

    I just hope they can use AI to make those piles of rentable internet connected electric scooters littered all over some cities much smarter and show more relevant ads. /s

    • VintageVNvet says:

      Bob,
      Your comment: ”AI doesn’t seem to function at that lower level of fundamental transactional communications , ”
      Is way too early, as in ”maybe SO,,, maybe NO.”
      While hope is in the air ( and investors minds far shore )
      IMO it is still WAAAYY too early to know with any rational degree of certainty what this latest allegedly advance of digital and electronic ability will actually deliver.
      ”Gamble” if you choose, but be clear it is gambling, eh

    • Boobster says:

      Agree with the icing on the cake and sprinkles analogy. AI advancements were here long time ago. Speech translation, image recognition, protein folding and what’s not, were unnoticed by the public, until clever VC guy scaped some portion of web, trained on it and opened their chatbot to public. This thing is on its late stage but presented as the second coming, which of course it is not.

  30. spencer says:

    Greenspan was just concerned about Y2K and goosed the economy as a result. That created the bubble and then he burst the bubble. Money flows didn’t bottom until November 2002 (the same time stocks bottomed).

    AI generated growth makes good sense. So, it doesn’t have to dovetail the FED.

    • Wolf Richter says:

      You got your dates wrong. In Jun 1999, Greenspan HIKED to 5.0%, HIKED in Aug 1999 to 5.25%, HIKED in Nov 1999 to 5.5%. And that’s where rates stayed through Y2K.

      Then in Feb 2000, he hiked again and then again and again, and in May 2000 to 6.5% where they stayed until Dec 2000 when he started cutting just before the recession in 2001.

  31. One might wonder if the intense demand for AI is merely a short-term rush to secure a first-mover advantage or safeguard market share and meet growth targets for investors. Will this surge soon lose momentum?

    The drive for profit and incentive structures suggests it won’t. Online commerce and advertising have arguably become the most successful business models in history, as evidenced by impressive revenues, profits, and market valuations.

    In online consumer commerce, a substantial amount of money is spent by providers of goods and services to channel traffic to a limited number of websites out of millions. AI can enrich and expand these search functions and open up new transactional opportunities. Why would this growth stop? Imagine the density of potential transactions and services. Consider the back-end processing needed to map billions of consumers’ precise locations and instantly align that data with their individual, constantly updating preferences—like a person’s local interests in petting zoos, children’s stores, health-conscious family restaurants, and GPS directions home, plus real-time updates on speed traps, stalled cars, and roadside McDonalds.

    These preferences shift over time, requiring continuous updates. Even a simple Google search now includes a brief, AI-generated summary for each suggested page or article. That’s AI at work. When deciding whether to read an article, a book, or watch a movie, wouldn’t faster transactions follow if the content were already summarized and categorized for you? Higher transaction frequency means more incremental profits for someone.

    The possibilities for AI applications are limitless, with many still beyond our current imagination. However, in every sector, someone can envision an AI-driven solution that could increase profits, cut costs, enhance products, or achieve other valuable outcomes. The demand for data centers may very well never be fully satisfied.

    • random50 says:

      The US military funds all sorts of crackpot ideas purely because of the extreme advantage they would yield if they actually worked.

      AI currently feels a bit like that for the mega corporations.

    • Franz G says:

      interesting theory. kind of like the companies that rented tons of office space that they didn’t need so that they could get it before someone else did.

      we all know how that worked out.

    • Biker says:

      Looks AI generated.

  32. Jon says:

    Well-written article as always, Wolf. I work in AI at one of everybody’s favorite AI company and can provide some insight. I tend to look at a handful of things to try and read the tealeaves on when AI spending will slow down or mature:

    1. We must recognize that no Fortune 1000 company can afford to NOT invest in AI – particularly big tech. Whether it’s for development, licensing, or products/services, there’s an arms-race effect going on with real business value (how much value and the ROI remains to be seen in corporate). You can’t compete if your technology is inferior.

    2. Consider the global datacenter market: 25% of the entire market gets refreshed every year as has been the standard practice for compute capacity planning. AI is driving double-digit growth of this market annually and is not projected to slow down by any forecast measure including my next point:

    3. Energy forecasts are proving to be a solid indicator of forecasting compute demand. You can extrapolate AI’s long term viability by looking at energy forecasts. A meta-signal of this strength can also be found in Big Tech’s ability to meet their pre-AI climate objectives (hint: AI ruined their objectives).

    In summary, AI investment and the cottage industry around it are here to drive the economy for a while and won’t be a factor of a recession ay time soon.

  33. Bear Hunter says:

    No question that more money will move and be created. The real question is the value of that money!

    It inflation comes back, you have till the next election to dump your dollars and get into real assets or means of production.

    • Franz G says:

      the fact that bitcoin is now at 85,000 dollars is not a good sign.

      the fact is, despite the fed having done most of the right things in the past couple years, their inability or unwillingness to tamp down on speculation does not bode well for the dollar.

      • Depth Charge says:

        BitCON is at $85,000 because the billionaire crowd piled in and drove it there, then corporations and everybody else (politicians) followed suit. Now, crypto is a in a worldwide gambling bubble. They have essentially assigned massive value to pet rocks without the actual pet rocks.

        • phleep says:

          Interesting that the same guys have been hawking it, who will now be its apex predators, I mean, traffic cops, regulators.

        • Escierto says:

          Get with the program. Now that our Great Leader has been elected, Bitcoin has now been anointed with his blessing. To Infinity and Beyond! Is this a great country or what?

        • Bobber says:

          Agree. I have to wonder if there is there anything stopping another anonymous player from developing a platform just like Bitcoin? An identical new platform would offer much more value in terms of potential growth than what Bitcoin offers now, because all the big money has been made.

          I don’t think Bitcoin’s acceptance as a payment platform is anywhere near what was expected five years ago. By that standard, I’d say it is a huge failure so far. The only thing that seems to be driving it higher is price momentum and perceived scarcity.

        • sufferinsucatash says:

          “HEY GUYS! 👋!! Buy My Bit Trash! 🗑️ “

  34. Benton says:

    Right now most end-users are getting some experience with AI for “free” or highly subsidized cost rates. The question I have is will AI in whatever form it stabilizes into be cost effective? The investment to power AI is just beyond anything I have ever seen in IT – even during the dot com era which was an investment in people to build stuff whether it made sense or not. In this case there is a rush to build data centers which you can’t easily downsize or “right-size”. Once you sink billions into a AI data center it is a sunk cost. And companies are gambling that new technology isn’t going to make a brand new AI data center design obsolete overnight. The closest example in recent time I can think of is fiber optic cable mania of the early 2000s. Companies and speculators poured billions into laying fiber optic lines all over the country with the promise of future projects based on bad assumptions of need. The technological mistake was they modeled their projections on the assumptions that only a single light signal could be transmitted down a single strand of fiber optic cable. Technology advances very shortly emerged to allow many signals down a single strand of cable. That killed that speculative funding model and the fiber optic investments died overnight. Several very large organizations went out of business. Due to a misunderstanding of technology, we suddenly had way too much fiber optic capacity due to poor planning and wild speculative investment. I just can’t help to see some parallels in the AI investment model.

    • phleep says:

      I hope we don’t wind up with the nuclear power plant equivalent of orphan gas wells, abandoned by bankrupt companies.

      • AndyfromLV says:

        No there’s no need to worry about that. There shouldn’t be any abandoned nuclear power plants anywhere. If we are going to ween ourselves from fossil fuels, nuclear has to be part of the mix. To think otherwise is delusional.

        More people die every year from accidents in the oil and gas industry than the number of people who have died from all the nuclear power accidents…ever. 31 died in Chernobyl and one death attributed to Fukushima. ZERO deaths for 3-mile Island.

        Heck, in 2019 alone the oil and gas industry had 114 deaths attributed to it.

        • Benton says:

          Instant deaths are not the only metric and far more than 31 people died from Chernobyl. I mean we are talking USSR data reporting here. Even recently a bunch of Russian soldiers died when they dug into highly radioactive ground around Chernobyl during the recent invasion. The problem is nuclear contamination as everyone knows unless you stick your head in the sand. The contamination zones around both sites is large and deadly and last a very long time. That said, nuclear is of course part of the solution due to the huge power production from nuclear. Managing nuclear is always going to be a huge problem because industrial mistakes are not just an oil spill or fire.

    • Redundant says:

      I think debt is what eventually makes the ai party fall apart — everyone apparently has endless access to cash today, but the desire to get ahead of the curve will end up burning up too much cash.

      The commingling of football stadium size data centers with a tsunami of obsolete mega chips — to power cool new shopping sites, so that consumers can find new socks, is utterly stupid as hell — but we need nuclear power plants to help make that more efficient…

      Building out useless capacity is the main attraction of a bubble:

      “Global Crossing has never reported an annual profit since it was created in an ambitious plan to extend its sole resource — a fiber optic cable traversing the Atlantic Ocean — into a 100,000-mile network connecting 27 countries in the Americas, Europe and Asia. The company has $22.4 billion in assets and $12.4 billion in debt, making its filing the largest bankruptcy by a telecommunications company.”

      I kinda recall JDSU had enough optical fiber to go to the moon — before it crashed, with other companies racing to get ahead of the curve.

  35. Michael Engel says:

    SPY and QQQ might have a major correction.

  36. AV8R says:

    R2K to 2750.

  37. Andrew Stanton says:

    Hope this works out better than Global Crossing.

  38. Michael Engel says:

    SPY and QQQ might have a correction.

    • phleep says:

      This statement is as verifiable or falsifiable as it would be on any of the last 10,000 days. I draw my own conclusions about its informational content.

  39. Redundant says:

    Probably the only person in the world to find this funny is WB — not sure about WR?

    The nividia mkt cap is now 12% of GNP

    That’s in a $29 trillion economy that’s growing….

    “The growth rate of the United States’ Gross Domestic Product (GDP) in the third quarter of 2024 was 2.8%. This is a decrease from the 3.0% growth rate in the second quarter.”

    I think that implies nividia will be worth a trillion more next year and just keep sucking investment growth out of everything else — of course, that’s just one company, so total value of Mag 7….

    And the funniest thing of all, is:

    The ai mkt is projected to be about $800 billion next year ….. hmmm?

    • sufferinsucatash says:

      You just don’t realize how much NVDA stock is owned by US retirement savers, billionaires and lots of other people.

      I think it has split 4 times. It may split again.

      There are several other semi conductor plays as well. Like TSMC, Micron, Marvel.

      One of them sells at like $700 a share.

      They’re not making cupcakes.🧁

  40. RickV says:

    In a recent article, I Talk to Robots While Driving, the writer, tomtungus.com/working-with-ai, describes a professional manager who has written a script so AI reads his morning Emails and drafts responses which he later reviews. Says it saves many hours. By the way, Wolf could use a similar script to review WS comments for unacceptable posts. Would probably also save many hours. I offer free advice in return for your many outstanding posts.

    • Wolf Richter says:

      That’s the last thing I’m going to do, is put the fate of comments into the hands of AI.

      Google’s AI has produced consistently shitty responses in my experience so far.

      But look, a machine reading stuff to me is old hat. My computers have been doing that for over 15 years. Hearing my own stuff read back to me helps me catch typos and garbled syntax (but I don’t usually do it for comments).

  41. Redundant says:

    One issue that exposes an ai-type company to Buffett Indicator valuation reality, is earnings reports — and the exciting game of keeping played spinning, versus crashing.

    Apparently the entire world waits in awe as nividia reports earnings in a few weeks — praying that it can hit its 0.07 cent EPS.

    I reckon if they nail it, the shares will explode towards $156, getting it near $3.8 trillion mkt cap or 13% of us GND.

    The concept of crowding out becomes more interesting going forward — the concept, that as more goes into feeding ai, money is bleeding out somewhere else — and as the ai monster eats more, there’s a greater disconnect in the Buffett Indicator.

  42. Joe says:

    “But this stuff takes years”! Well it’s already been a solid 3 years of insane AI spending (not counting the considerable spending prior to the last 3 years). I don’t see how this lasts too much longer before the crack and soon to follow crumble.

  43. Biker says:

    The article implies that the top is not in (a least not yet, a bit of change of heart?).
    My thought after the read: that means that the top might be in. LOL

  44. Wages v CPI says:

    Wolf – where are wages in relation to overall inflation since pre-pandemic? You’ve covered both but can’t recall if you ever plotted the two against one another to compare?

    Thanks!

    • Wolf Richter says:

      Wage growth was slower than inflation in 2021 through mid-2022 but then accelerated as inflation decelerated and since mid-2022 wage growth has been higher than inflation. But wages are personal (not everyone gets the same job or raise), and inflation rates are personal too, so per household, this varies. The biggest percentage wage increases were at the lower-wage jobs initially, though that has now changed.

      The way we plot this is to use “real” wages, meaning wages adjusted for inflation. There are different measures of “real wages” and “real household income” out there, producing different results.

      I use the personal income without transfer payments, which includes wages and salaries, and income from small businesses, rentals, dividends, interest, etc. adjusted for inflation, but without transfer payments such as Social Security, VA benefits, unemployment insurances, Welfare, etc.

      https://wolfstreet.com/2024/09/27/consumer-income-savings-rate-were-revised-massively-higher-going-back-2-years-spending-revised-up-too-stunning-numbers/

      And real disposable income (after-tax income from all sources, including from transfer payments). This chart shows the year-over-year percent growth, so you can see where it fell behind, and where started catching back up:

Leave a Reply

Your email address will not be published. Required fields are marked *