I kept replaying one line from a conference room in Connecticut this week, mostly because it refused to behave like a normal warning. A market veteran stood up, called artificial intelligence the biggest bubble of our lifetime, and then, almost in the same breath, told people to get in. Not later. Not after a tidy pullback. Now. If you have spent any time around markets, that pairing feels wrong on purpose. Bubbles are supposed to be the thing you spot and then sidestep. Here the spotting is the sales pitch.
Maybe that is why the remark landed harder than another round of chip headlines. The S&P 500 had just printed a fresh record, with AI-linked shares doing most of the lifting, and the chip leader’s market value was pressing toward six trillion dollars. Plenty of people already feel late. Hearing a bubble label from someone who still wants exposure does not calm that feeling. It sharpens it. You are no longer arguing about whether the story is hot. You are arguing about where you are on the clock.
A Bubble Call That Doubles As A Buy Signal
The comments came at the Greenwich Economic Forum on October 6. Michael Novogratz, founder of Galaxy Digital, described the AI boom as the largest speculative surge he has seen, then argued the ending conditions are not in place. His phrasing was blunt enough to travel: bubbles do not finish the way this one looks right now. He knows how they end. They end in a spectacle. This, in his view, is not spectacular enough.
That is a timing claim dressed up as a confession. He is not saying prices are cheap in some absolute, sleepy sense. He is saying the mania has not yet produced the kind of disorderly climax that usually marks the last act. I have found that distinction easy to miss when a headline only keeps the word bubble. The second half of the sentence is the whole trade.
He also leaned on valuation, which is the part skeptics tend to skip. On a price-to-earnings basis, he still sees AI-related shares as attractive enough to own aggressively. The line he used for people sitting out was almost comic. If you are not invested, you might as well go home and stick your head in a bucket of ice water. Colorful, yes. Also a tell. He is not whispering a hedge. He is scolding the sidelines.
Bubbles do not end like this. They end in a spectacle, and this one has not reached that point.
Paraphrase of the forum remarks
Policy sits in the background of his case. Competition between the United States and China, he argued, makes it very hard for Washington to slow the buildout, even to a pace some builders might privately prefer. If the state is more likely to accelerate than to brake, the usual political off-ramp looks blocked. That does not make prices safe. It does change the list of things that can kill the story quickly.
Why The Phrase Not Spectacular Enough Matters
Most bubble talk collapses into a binary. Either we are early and clever, or we are late and doomed. The forum remark tries to live in a third lane. The boom can already be a bubble, and still have room to get stranger before it breaks. That lane is uncomfortable because it asks you to hold two ideas that feel like they cancel each other.
Think about how past manias actually felt in the middle, not in the documentary. Internet shares in the late 1990s were called a bubble for years before the peak. Housing was called a bubble while prices were still climbing and credit was still loosening. Crypto had loud bubble calls in 2017 and again before the 2021 blow-off. The label was often right. The calendar was often wrong. Being right early and being paid are not the same job.
What would spectacular look like, if you take the phrase seriously? Not a strong earnings print. Not another record close. Spectacular, in market memory, looks like refusal to do math, like financing that only works if the story never pauses, like retail slogans replacing cash-flow questions, like insiders selling into a crowd that no longer asks who is on the other side. Some of that is already visible at the edges. A lot of it is not the dominant mood in the core AI trade, where profits, order books, and power contracts still anchor the conversation.
Perhaps the most interesting aspect is how calm the core still sounds. Bubble endings are rarely calm. They are loud, sloppy, and strangely confident. If the room is still arguing about forward earnings and power supply, you may be in a boom with bubble traits, not yet in the final stampede. That is an opinion, not a law. Booms can skip the cartoon ending and just roll over. History is rude that way.
The Other Side Of The Room
The same week produced the opposite warning, and ignoring it would be lazy. Ray Dalio, speaking in Singapore, called the AI surge a classic bubble and said it was close to bursting. His worry is not that chatbots are fake. It is that the financing climate is tightening around a capex cycle that still needs a lot of borrowed patience. Rising long-term borrowing costs sit in that argument like a stone in a shoe. They do not stop the story overnight. They change the discount rate on every promise.
Other desks landed in between. Some strategists have framed the setup as a bubble you can still participate in if you hedge, pointing to risk gauges that rhyme with the months before the dot-com peak rather than the week of the crash. Others look at the chip leader’s forward multiple, sometimes cited near the high teens, and say that is not how bubbles price the main engine. A 70 percent growth expectation for the coming year, if it holds, makes a mid-teens multiple look ordinary. If it does not hold, the same multiple looks like a trapdoor.
I do not think you have to pick a team and wear the jersey. The useful split is simpler. One camp is trading the distance to the climax. The other is trading the fragility of the funding. Both can be partly right for a while. Markets do that more often than newsletters admit.
What The Numbers Are Actually Arguing About
Strip the personality out of the forum and you are left with a handful of numbers that refuse to agree. Index highs. A chip company approaching a six-trillion-dollar value. Forward earnings that still look tolerable to bulls. A cost of capital that keeps creeping up. Capex plans measured in power plants, not software licenses. None of these is a punchline. Together they are the whole debate.
Price-to-earnings is the bull’s favorite door because it lets profits into the room. A narrative stock with no earnings can only be defended with a story. A company printing enormous earnings can be expensive and still not be absurd. That is the narrow bridge Novogratz is walking. He is not claiming the multiple will never compress. He is claiming compression is not the base case while growth stays violent and the political system keeps feeding the buildout.
Bears answer with concentration. When a small cluster of companies does most of the index’s work, the index stops being a diversified vote on the economy and starts being a vote on one supply chain. That can last longer than feels decent. It also means a single disappointment in orders, power, or export rules can move retirement accounts that never meant to take a single-theme bet. I have watched people discover that only after the fact. The discovery is never pleasant.
| Camp | Core claim | What would prove them wrong |
| Buy the bubble | Ending conditions are not here, and earnings still support the leaders | A climax in speculation, or a clear break in order growth |
| Stand aside | Classic bubble traits plus a higher cost of capital | Multiples stay contained while profits keep compounding |
| Own it, but hedge | The trend can continue while the left tail gets fatter | Hedges bleed for so long that the protection becomes the loss |
That table is not a forecast. It is a way to stop mixing the arguments. A lot of online fights are two people answering different questions and calling the other person blind. One is asking whether the technology is real. The other is asking whether the price already assumes a perfect decade. Those are not the same question, and mixing them is how smart people talk past each other for a year.
A Short Tour Of Bubbles That Were Right Too Soon
Every generation gets a bubble lecture that uses the same three slides. Tulips, if the speaker wants theater. Railways, if they want infrastructure. The internet, if they want something the audience actually lived. The lesson people take home is usually “do not be the last buyer.” The lesson the tape actually teaches is messier. Real technologies and absurd prices can share a decade.
Railways changed commerce and still ruined plenty of shareholders. The internet changed everything and still cut many indexes in half. Both facts fit in one paragraph without contradiction. The technology did the work it promised. The capital cycle overshot the cash the work could throw off in the early years. AI can rhyme with that pattern even if the chips, the models, and the power deals are all genuine.
Where this cycle feels different, at least so far, is the profit. The late-1990s leaders included plenty of firms whose business model was a slideshow. Today’s core AI suppliers sell scarce hardware into customers who are themselves large, cash-generating companies. That does not immunize the trade. It changes the failure mode. You are less likely to wake up and find the product was imaginary. You are more likely to wake up and find the product was real and the capacity was built twice.
- Real utility does not cap the multiple. It only changes what the multiple is discounting.
- Early bubble calls can be directionally right and still ruin a short seller’s year.
- The painful phase is often the overbuild, not the invention.
- Index concentration turns a sector bet into a stealth portfolio bet.
There is a personal bias worth naming. I am more wary of stories that require the customer to stay irrational than of stories that require the customer to stay competitive. Companies buying compute because rivals are buying compute can keep spending after the press loses interest. Fear of falling behind is a duller motive than euphoria, and dull motives last longer. That does not make them safe. It makes the clock harder to read.
The Ice Bucket Line And What It Reveals
Colorful insults aimed at the uninvested usually age badly. They also reveal the speaker’s position. Novogratz is not a neutral referee. Galaxy Digital is a markets business with a long crypto history, and he has used the “biggest bubble of our lifetime” label before, in other cycles. Knowing that does not erase the point. It tells you the point comes from a man who is comfortable owning manias and who has also watched them break.
The ice-bucket jab is aimed at hesitation, not at research. In a forum full of allocators, hesitation is the competing product. Cash yields something again. Long bonds are no longer a free option. Sitting out AI no longer feels like sitting out a toy. It feels like sitting out the index. That social pressure is part of the bull case and part of the risk. When not owning the theme starts to look like a career error, flows can stay strong past the point where the incremental dollar is wise.
Would I repeat the jab to a friend who has a mortgage and a concentrated job in tech? No. The line is a stage line. The underlying claim is narrower. If your horizon is multi-year and your size is sane, missing the leaders entirely is its own active bet. Active bets deserve a reason, not just a mood.
Rates, Power, And The Boring Ways This Can Crack
Dramatic endings get the documentaries. Boring cracks do the damage. A higher long-term rate does not need a scandal. It just reprices the back half of every growth model. If the ten-year cost of money stays elevated while AI revenue arrives on a lag, the net present value of the buildout shrinks even if the factories open on time. Dalio’s warning lives mostly in that sentence.
Power is the other boring crack. Models do not run on adjectives. They run on electricity, cooling, grid connections, and permits. A chip can be sold and still sit waiting for a substation. That delay does not kill demand. It can scramble the quarter when investors expected a clean ramp. Markets hate scrambled quarters more than they hate hard problems, because hard problems can be narrated and scrambled quarters show up in the print.
Export rules and rival silicon sit in the same drawer. A political system that refuses to slow AI at home can still restrict where the best hardware goes. That is not a brake on the theme. It is a reroute. Reroutes create winners and losers inside the theme, which is easy to forget when the index treats the cluster as one trade.
Ways the boom can stumble without a spectacle: cost of capital stays high power hooks up slower than chips ship customers pause orders to digest capacity one policy rule reroutes the supply chain the index discovers it was one trade
None of those is the fireworks ending Novogratz says he has not seen. They are enough to hurt. A portfolio that only prepares for fireworks will miss the kitchen fire.
How To Own A Theme You Already Distrust
If the honest read is “bubble traits, profits still real, climax not obvious,” the practical question is size, not loyalty. I have found that people lose more sleep over the label than over the weight. A 4 percent sleeve and a 40 percent sleeve are not the same decision, even if both are called AI exposure. Treating them as the same decision is how the debate gets religious.
A workable approach starts with what you already own. Many broad equity funds are already heavy in the leaders. Adding a second pile of the same names is not a view. It is a duplicate. Before anyone buys the ice-bucket sermon, they should look through the funds they already hold and count the overlap. You might be in the trade and still feel like an outsider because the statement does not say the word intelligence on the cover.
- Count the exposure you already have through broad funds.
- Decide the maximum weight you can hold through a 30 to 50 percent drawdown without selling in a panic.
- Split that weight between leaders with earnings and a smaller basket of suppliers, not a pile of story stocks.
- Write down the two numbers that would make you cut, before the tape asks you to invent them.
- Keep a cash or hedge sleeve if the cost of being wrong early matters more than the cost of being late.
The fourth step is the one people skip. “I will get out if it gets crazy” is not a rule. Crazy is a feeling, and feelings arrive after the price. A rule looks more like this: if forward growth guides fall below a number you chose, or if your sleeve crosses a weight you chose, you trim. The number can be wrong. The existence of a number is what separates a plan from a mood.
Signals That The Spectacle Has Started
If the buy case rests on “not yet spectacular,” you need a picture of spectacular that is not just a red day. Red days happen inside healthy trends. Spectacle has a texture.
Watch the quality of the new issuance. When companies with thin products and thick slide decks start raising easy money on the theme, the marginal dollar has changed jobs. It is no longer funding scarce supply. It is funding adjacency. Adjacency is where bubbles go to get sloppy. Also watch insider selling that is no longer routine, customer comments that shift from “we cannot get enough” to “we are digesting,” and multiples that rise while growth guides fall. That last combination is the classic tell. Price up, fundamentals down, story louder.
Retail language is a softer tell, and easier to mock than to time. When people who do not follow gross margin start explaining the trade with slogans, you are later than the conference circuit. You are not necessarily at the top. Slogan phases can run. They just raise the odds that the next surprise is a disappointment rather than another beat.
The dangerous moment is not when skeptics exist. It is when skeptics sound unserious to people who have stopped checking the order book.
There is also the funding tell Dalio is pointing at. If the buildout increasingly depends on debt that only works at low rates, and rates do not cooperate, the spectacle can arrive through the bond market rather than through a meme stock. That version is quieter on social feeds and nastier in credit spreads. It is worth respecting even if you do not share the timing.
What This Does To The Rest Of A Portfolio
An AI-led index high is not a neutral backdrop. It pulls correlations around. Energy and utilities can suddenly matter because they sit under the data centers. Industrial names tied to cooling, grid gear, and construction stop being dull. Software that resells access to models gets a narrative premium it may not earn. Meanwhile, businesses with no AI angle can look cheaper for the wrong reason: they are simply not in the flow.
That relative cheapness is a temptation and a trap. Buying what is ignored because the leaders feel frothy can work, if the ignored business has its own engine. Buying it only as a protest against the bubble is how people end up with a portfolio of slow growers that still falls when the index falls, just with less upside on the way up. Protest portfolios are a mood with a brokerage login.
Crypto-adjacent investors have a special version of this problem. A voice they know from digital assets is now loudest on equity AI. The skills transfer only partly. Token cycles and chip cycles share speculation. They do not share the same cash flows, the same regulators, or the same customers. Treating a bullish AI call as a cue to lever a different theme is how the ice bucket gets passed to the wrong head.
A Cleaner Way To Hold The Contradiction
Here is the version I can live with, without pretending it is a model. The technology spend is real. The index is concentrated. The core suppliers still have earnings that make a pure story-stock comparison unfair. The cost of capital is less friendly than it was. The political system is more likely to race than to pause. The climax, if it comes in the old theatrical style, is not obvious in the tape this week. Therefore a sized position is rational, a maximal position is a different bet, and a zero position is also a bet.
That paragraph will not satisfy anyone who wants a side. Good. Sides are how this topic gets dumbed down. The forum line works as a provocation because it refuses the side. Biggest bubble. Still buy. The refusal is the content.
If you need a single habit to take from it, make it this: separate the technology question from the position question every time you check the account. Is the spend still happening? That is one note. Is my weight still inside the loss I pre-committed to? That is the other note. Mixing them is how a good quarter becomes an identity.
Position check: existing weight + new buy < loss you can sit through. If not, the bubble debate is already lost.Where The Debate Goes Next
The next few prints will not settle the lifetime question. They will settle the near-term one. Order commentary from the big buyers of compute, any stumble in power timelines, and the path of long rates will matter more than another conference adjective. If guides stay hot and multiples stay merely rich, the “not spectacular enough” camp gets another quarter. If guides cool while prices levitate, the classic-bubble camp gets the better photograph.
Either way, the argument has already moved. A year ago a lot of people were still asking whether the spending was a fad. This week a room full of professionals argued about the shape of the ending, not the existence of the boom. That shift is itself information. Themes do not get bubble labels from veterans while they are small. They get them when they are large enough to embarrass anyone who missed them and anyone who overstayed.
I keep coming back to the ice bucket, not because it is clever, but because it is a dare. Dares are a poor way to build a portfolio and a very good way to see what the speaker believes. He believes the cost of absence is higher, right now, than the cost of being early to an ending he says has not started. You do not have to accept the dare. You should at least notice that it is a timing statement, not a comfort statement. Comfort was never on offer.
So the practical close is plain. Treat the AI boom as a real capex cycle wearing bubble clothes. Own what you can explain, in a size you can hold, with a rule you wrote down on a quiet day. Leave room for the spectacle if it arrives, and leave room for the boring crack if it arrives first. The market will not send a calendar invite for either one.
