AI, are we in a bubble? A calm walk through the cyclical case, the liquidity rotation, and why the Meta selloff might be bullish

[Harper48]

Every few weeks the bubble question comes back around and the temperature of the discussion tells you more than the charts do. So let us try to do this properly and without the theatre, no doom, no cheerleading, just an honest walk through what is actually moving and why, and where reasonable people land on each piece. The goal here is to think, not to be right loudly

Start with the word itself, because bubble has become a lazy catch all. A real bubble is prices detached from any plausible earnings, funded by pure belief and greater fool buying. What we mostly have instead is a small group of enormous, profitable companies spending colossal amounts on capex, which is a very different animal. You can absolutely overspend and misallocate inside a boom without the whole thing being a 1999 style fiction, and conflating the two is where a lot of the noise comes from

The cleanest lens on the recent wobbles is cyclical rotation rather than collapse. Money that piled into a handful of AI names for two years is not vanishing when it moves, it is rotating, and liquidity reversals into other sectors look like a crash if you only stare at the thing being sold. A market where capital leaves the crowded trade and spreads into laggards is arguably healthier than one where ten names carry the entire index on their backs indefinitely

This is why the software selloff needs context. A big part of why software names got hit is precisely that they are the ones spending so heavily right now, pouring money into compute, models and the teams to deploy them, which compresses near term margins even when the long term thesis is intact. The market is currently paying for cash flows today and punishing promises, so the companies visibly spending the most get marked down the hardest, and that is a valuation rotation, not a verdict on whether the technology works

The semiconductor moves deserve the same calm reading. What keeps getting called a crack is often just a healthy pullback in names that ran extraordinarily far, extraordinarily fast. Pullbacks in the semis are not the thesis breaking, they are a violently extended move catching its breath, and a sector that only ever goes vertical is the actual warning sign, not one that corrects and consolidates along the way

The Meta compute selloff is the best recent example of a move that got the story backwards. The reaction hammered CoreWeave and Nebius and dinged the chip names on the inference that if Meta has spare compute to sell, then someone somewhere massively overbuilt. But that read only holds IF there is genuine excess capacity sitting idle, and by every available signal there simply is not, demand for compute is still outrunning supply across the board

Follow that logic through and the selloff looks as bullish as it was bearish, arguably more so. A hyperscaler moving to sell compute is not a confession of a glut, it is a company monetising an asset it already owns into a market that is still starving for it, which is a demand signal wearing a scary costume. The panic priced in oversupply, the reality points the other way, and moves built on the wrong premise tend to be the ones that unwind

None of this means the skeptics are stupid, and the strongest version of their case deserves respect. The concentration is real, the capex numbers are genuinely staggering, a lot of the revenue is circular as the big players fund each other's compute bills, and plenty of deployment is still stuck in pilots that never reached production. Those are legitimate concerns and anyone dismissing them entirely is doing the same lazy thing as the doomers, just in the opposite direction

But here is the part worth saying out loud, a striking amount of bubble commentary is narrative dressed as analysis. Some of the loudest bubble callers are talking their own book, positioned short and needing the story to spread, or pushing a broader thesis that AI is hype for reasons that have nothing to do with the balance sheets. That does not make them wrong automatically, it just means you should separate the argument from the incentive every single time, and notice how often the confident bubble call arrives without a single number attached

So where does a calm person land? Probably here, this is a cyclical, overextended, sometimes frothy boom inside a real technology shift, with liquidity rotating rather than evaporating, spending pressure that is squeezing margins today and building capacity for tomorrow, and healthy pullbacks getting mislabelled as the top by people who benefit from you believing it. Froth in places, yes. A house of cards, the evidence does not support it. Interested to hear where the board actually sits on this, and especially from anyone who thinks I am too relaxed

[Frost Jay]

This is the most level headed take on it I have read in weeks. The rotation point is the one nobody wants to hear because a crash headline gets more clicks than money moved from software to industrials, but that is largely what has happened

[MrRicardo]

Mostly agree but I would push harder on the circular revenue point than you did. When the same handful of firms are funding each other's compute spend, a chunk of the demand IS the capex, and if one of them blinks the loop can unwind faster than the bulls expect. It is not fiction but it is more fragile than the calm version suggests

[AlexaBliss]

The Meta bit is the key insight and it is genuinely underrated. Everyone read spare capacity as a glut signal and nobody stopped to ask whether the capacity even exists to be spare. It doesn't, so the selloff was a reflex, not a read

[Forge37]

Disagree that it is not a bubble, respectfully. Every bubble in history had real technology and real earnings somewhere in it, railways were real, the internet was real, that never stopped the valuations getting silly and correcting brutally. Real tech and a price bubble are not mutually exclusive and this thread slightly waves that away

[Sequence87]

That is a fair challenge but the counter is concentration of PROFIT this time versus concentration of story last time. The 1999 leaders were burning cash, these ten gush free cash flow, which is exactly why calling it a straight repeat keeps not working

[BanterQueen]

The talking their own book paragraph needed saying. I follow a couple of very loud bubble accounts and it took me a year to notice not one of them ever posts a number, it is all vibes and posture, and several are openly positioned for it to fall. Consider the source is not a dismissal, it is basic hygiene

[NeuralSeer]

Where I land is close to yours but more cautious, froth is real, the tech is real, and the thing that actually scares me is not valuation but a single guidance cut from one hyperscaler triggering the rotation all at once. The exits are narrow when ten names are the whole trade

[Joel96]

Semis pulling back after the run they had is the most normal thing in the world and I wish more people treated it that way. A vertical line with no corrections is the setup that ends badly, the ones that breathe on the way up are the healthier charts

[SoloOrca]

Slightly too relaxed for me, since you asked. The capex is real spending against demand that is still partly projected, and if adoption in the enterprise stays stuck in pilots the way it has been, the bill arrives before the revenue does. I am not bearish, I am just not as comfortable as this post

[Karen88]

The software margin point explains so much of the recent price action that got called a crack. They are spending heavily NOW and getting marked down for it NOW while the payoff is later, that is a timing mismatch, not a broken thesis, and the market is terrible at pricing timing mismatches

[Ava82]

Good thread, and the framing I am taking away is separate the froth from the fraud. There is froth, there is overextension, there is misallocation, and there is basically no evidence of the fake it till the music stops dynamic that defines an actual bubble. Those are different diagnoses and they get treated as one word far too often