To be clear, I am not asserting here that the hyperscalers (Google, Microsoft, Amazon, Facebook, and Oracle) are morally equivalent to Putin. My assertion is only that they are making a logical error similar to Putin's behavior towards his invasion of Ukraine.
I don't normally approve of the phrase, "...on steroids", because steroids are normally evidence of a problem. Taking steroids is messing with your body's chemistry, and while occasionally there is a big problem which steroids may help you with that makes it worth the side effects, "on steroids" should not be taken to mean "...except better". If you're taking steroids, either there's a big problem which you need help with, or there wasn't a big problem until now, and you're making one. "On steroids" really means, "and you're desperate".
However in this case, the metaphor is apt. What the hyperscalers and Putin are doing, can be rightfully described as the sunk-cost fallacy, on steroids.
Most of us have had that experience of beginning something, with the expectation that it would turn out to be much easier than it actually was. Sometimes, we don't care that much, and quickly stop. Often, that is a mistake, and we will need to stiffen our spine and persevere more if we want a fulfilling life. It is important to recognize, though, that there is an opposite mistake, which can be just as severe; refusing to give up and change course, when it is apparent that things aren't working out.
The "sunk cost fallacy" refers to the error in logic where you make a mistake which you would otherwise have avoided, because avoiding it would require admitting that something we did previously was a mistake. Few people like to admit making mistakes, even to themselves. The classic example is spending more to try to repair a car which is now actually too far gone to be worth spending any more on. You shouldn't pay for a new transmission, but you replaced the air conditioning last month and you can't throw that away now, so you throw good money after bad. After paying for the new transmission, something else breaks, and then something else, and eventually you have to admit that what you need is a new car. You can probably insert something from your own life into this, it's a common error.
However, there is a final barrier to us falling into an abyss of bad judgement, and that is the fact that we only have so much money to lose. Eventually, no matter how unwilling we are to admit that the costs we have already sunk are gone, we have to stop, because we don't have the ability to continue. Imagine what might happen, if you had an near-infinite credit line. You could pile up more debt than you could ever get out of.
When Vladimir Putin made the decision (and it does appear to have been entirely his) to try to conquer all of Ukraine in a 3-day special military operation, he no doubt expected Ukraine's government to fold quickly. When it did not, it is possible that he secretly regretted his decision. What he did not do was decide, "well then, it's not worth it, let's pull back". He could not admit that he had made a mistake, therefore he could not admit that it would never work out no matter how much they persisted, therefore he kept pouring more resources into a doomed attempt to make it work. Men, money, and more or less the entirety of the Soviet-era stockpile of tanks, shells, and other military gear, all got thrown away rather than admit the initial error. When the "sunk cost fallacy" has a near bottomless pit of resources to burn through, it can reach a point of no return, because the higher the costs that have been sunk, the harder it is to admit that it will never work.
One hopes to be incorrect, but it appears that Putin will never admit his mistake, and never cease in pouring all of Russia's resources into an endeavor destined to fail, until he is entirely out of men, money, and material to burn. Perhaps it is possible for him to admit his mistake at this point and still survive, but perhaps it is not, and even if it is he may not wish to take that chance. Throwing good money (and all the rest) after bad may not buy him victory, but it may buy him another year or two in charge.
We arrive at the hyperscalers.
No doubt they had some honest belief that by pouring hundreds of billions of dollars into Large Language Models (what we usually mean now when we say "AI"), they would take us into a Brave New World. They may not honestly have believed, as some OpenAI investment documents suggested, that AI would make money unimportant because it would make us all so prosperous, but they at least believed that the investment would pay off with a technology so "transformative" or "disruptive" that it would justify the cost.
I don't claim to know for certain what they think of that decision now, but the signs are that they will get very little for their efforts when all is said and done. None of Microsoft's products will make more money because of the addition of AI, which has proved to be about as popular with users as Clippy. Google's crown jewel, its search business, is getting eroded by LLMs with nothing remotely as big to fill the hole in its balance sheet. Oracle is now rated one step above "junk" by the bond market. Yet they have all spent so much by this point, and taken on so much debt, that they cannot admit that it will not pay off.
If I made a mistake this bad, I would run out of money in short order, and have to admit my mistake (or, even if I didn't admit them, I would at least have to stop spending). However, the hyperscalers were sitting on such enormous reserves, that they can continue to prop up the bubble by filling it with the smoke from their incinerating piles of cash. Until, one day, they won't be able to.
The AI bubbles in South Korea (Hynix, Samsung) and China have already deflated, but while there have been rough days, so far the US bubble has proven unwilling to stay popped. We know that the hyperscalers are burning through their cash rapidly, we know that they have engaged in complex and even opaque transfers of money between different players, either as investments or stock purchases or loans or weirder financial instruments. Given how tied up they all are in each other's fate by this point, it seems possible that they may be using that money to prop up each others stocks as well.
Many commenters on the AI bubble have concluded that it is less likely to end in a fiscal crisis than the mortgage problems that lead to the Great Fiscal Crisis of 2008, because the hyperscalers all have fundamentally sound businesses that they can fall back on. Unlike Countrywide or Bear Stearns, when the bubble pops they will survive, and thus won't bring down the economy. This presumes, however, that they are willing and able to admit that the bubble has popped. What if, like Putin and his generals continuing to throw a generation of young (and even not-so-young) men into the meat grinder long after there was any reasonable chance of winning even the Donbas region of Ukraine, the hyperscalers keep throwing their money into datacenters that have no realistic chance of ever paying back the cost of building and equipping them?
Financial analysts saying that the AI bubble will not bring down the economy, are assuming a minimum of rational behavior by the hyperscalers. The assumption is that they will give up on making AI pay off, before they wreck the legacy businesses that they are meant to fall back on. Is there any proof that will happen?
And, if the tech titans of our economy (who employ about 1.5 million people in the US as of mid-2026) manage to spend their way into the ground? If they run out of money to spend, and borrow from the bond market until it won't lend them any more, and then sell more stock until the share price crashes? It would be highly irrational. How do we know that they can't be that irrational?