You really think the biggest technological breakthrough in human history won't create a bubble?
Come on.
There just ain't no way.
That is the thesis behind this website. Not that AI is fake. Not that the technology is overhyped nonsense. Not that the future is bleak. Almost the exact opposite: AI is so obviously important, the future it points toward is so obviously enormous, and the amount of economic value it could unlock is so large that people are going to lose their minds trying to own a piece of it.
Some will get rich. More people will see them getting rich. Then still more people will decide that they should be getting rich too. Prices will rise, confidence will compound, companies will issue stock into that demand, and leverage will make the whole thing more fragile than it looks.
That is how we get a stock-market bubble wrapped around a very real technological revolution.
This post is a snapshot of how I see that process unfolding today. I expect parts of it to be wrong. The point of the Four Horsemen is not to give me a prophecy. It is to give me a framework for watching the evidence as it changes.
The technology is real
Hell, an AI interviewed me to help write this piece.
AI also helped me build the website you are reading. Could I have built this site without it? No—not in anything close to its present form. Could I have developed the argument without AI? Yes, ish. It would not have been as good, and I probably would not have gone through the effort of publishing it.
This has been a beer project, not an obsessive, can't-sleep-until-it-is-finished project. AI lowered the cost of turning an idea in my head into something other people can actually use.
Today the most visible example is coding. It is already spreading into design, research, analysis, and ordinary knowledge work. OpenAI reports that its coding tools are being used beyond software engineering for research, data analysis, workflow automation, finance, legal work, and recruiting. That is a company describing its own product, so take the marketing discount you think is appropriate. Then look at what people are actually building and decide for yourself.
Soon enough, AI will be helping with your taxes and much, much more. If artificial superintelligence arrives, its consequences will be literally unimaginable in the strict sense of the word: we cannot confidently picture the economy on the other side of something smarter than us.
Our future is bright. That is the future I am betting on.
But a transformative technology and a good investment are not the same thing. Ray Dalio has made this distinction directly: AI can be in a bubble even though the technology will endure and transform the world. Michael Burry has likewise compared the recent rise in technology stocks with the final phase of the dot-com bubble. Credible people can be early. They can be wrong. They can also be worth listening to without being treated as oracles.
The underlying technology can win while many of the investments made in its name lose. The internet already taught us that.
The Netscape problem
Leopold Aschenbrenner's Situational Awareness essays laid out one of the most compelling versions of the bullish AI future. The core claim—that rapidly improving systems could drive an extraordinary industrial and geopolitical buildout—is not hard for me to believe. The revenue is no longer imaginary, either. Anthropic said in May that its run-rate revenue had crossed $47 billion while announcing a financing round that valued it at $965 billion.UPDATE 10/31 Anthropic's S-1 went public Oct 26. IFR (Oct 31) pencils a roadshow the week of Nov 9 and a Nov 20 debut. No price range yet. CAPE Multpl 40.13. See S-1 (EDGAR) and IFR.UPDATE 11/11 Terms are set: 825M Class A shares at $70–78 on Nasdaq (ANT), up to about $1.40T fully diluted, with pricing expected Nov 18 and trading Nov 19 (CNBC sources). Run-rate revenue was about $81B at Oct 31. CAPE Multpl 40.24 (Nov 11). See Anthropic, S-1/A (EDGAR) and CNBC.
But I keep returning to Netscape.
Netscape was how millions of people first got onto the web. Its 1995 IPO is often used as the opening bell for the dot-com bubble. Yet being the gateway to a technological revolution did not mean capturing most of the value created by that revolution. Microsoft bundled Internet Explorer into Windows, Netscape lost the browser war, and AOL acquired the company in 1999.
OpenAI and Anthropic could become two of the most valuable companies in history. They could also be our Netscapes: the products that show everyone the future while the durable economic value ultimately accumulates somewhere else. I do not pretend to know which outcome we will get.
The historical analogy also needs to be handled honestly. Netscape did not mark the top. Its IPO came in August 1995; the Nasdaq did not peak until March 2000. The final phase came years later, amid extraordinary IPO activity and first-day returns. There was no single, clean event that everyone agrees popped it.
That is useful humility for the comparison I am making today. A landmark IPO can be an ignition point, a phase transition, or a culmination. It does not come with a label ahead of time.
The Four Horsemen
The Four Horsemen are my way of reading the cycle. They are not a mechanical trading formula, and all four do not need to arrive on matching horses at the same moment. I am watching the headlines to see how often these forces appear and how strongly they reinforce one another.
I owe the framing itself to Owen A. Lamont (SVP / portfolio manager at Acadian; former Chicago / Harvard / Yale / Princeton finance professor). His Owenomics “Four Horsemen of the Bubble Apocalypse”—overvaluation, bubble beliefs, equity issuance, and investor inflows—is the inspiration this site’s dials build on. Start with No, we are not in a bubble yet, then Bubble beliefs, We are Not in an AI Bubble, The whirlwind is upon us, and IPOs of doom (Sep 2026). Also covered in Fortune (June 12, 2026).
Are prices unreasonable?
In May, Anthropic said run-rate revenue had crossed $47 billion while announcing a financing round that valued it at $965 billion. That private mark was already pricing an astonishing future into the present.
Update 10/31: Anthropic's public S-1 (Oct 26) shows $16.29B of H1 2026 revenue and Net Revenue Retention of about 900%. It also says 5–7% of nine-month revenue came from customers violating its usage policy. CAPE Multpl 40.13 (Oct 30, 2026). The 10-year Treasury closed at 4.94% on Oct 30 (Treasury official daily par curve).
Update 11/11: Anthropic's IPO range ($70–78, roadshow launched Nov 10) values it at up to about $1.40T fully diluted, against run-rate revenue of about $81B at Oct 31 (S-1/A). CAPE Multpl dipped to 38.91 on Nov 6 and was back at 40.24 on Nov 11, 2026. The 10-year Treasury closed at 4.98% on Nov 10 (Treasury official daily par curve).
A high multiple alone still does not prove a bubble. It does tell us the horse is already riding.
3 of 3 · flip for the Aug 4 thesis draft
Does everyone know it is expensive—and buy anyway?
This is where an investment thesis mutates into greater-fool behavior. “I know it is not worth that, but Joe Schmo is making money” is not fundamental analysis. It is a bet on the next buyer.
Right now the belief dial is mixed, not pure FOMO. Sam Altman called an OpenAI IPO in 2026 “ill-advised” on safety grounds; Dario Amodei is talking about pacing the frontier. Breakingviews meanwhile rhymes AI “neoclouds” with the 1999–2000 altnets—data-center upstarts, vendor finance, GPU leverage. Update 10/24: The altnet rhyme is getting louder. CoreWeave, Nebius and IREN are each down 30%+ in October, and Meta, Nvidia and Oracle CDS all set records (Oracle 343bp).
Update 10/29: Credit is the skeptic in the room. CoreWeave’s 5-year CDS set a record 1,167bp close on Oct 22 (S&P Global) and was 1,041.4bp on Oct 27 (TokenPost, vendor not named). Oracle’s closed at 293bp on Oct 29 on Investing.com, whose 2026 high close was 334.25bp on Oct 20; CNBC’s 343bp record came from a different price source. On Oct 29, Broadcom (175.4bp) and SpaceX (266.5bp) closed at records on ICE Data Services. Across eight names with same-vendor readings (Investing.com and S&P Global), the median spread rose 22.4% from Sep 29–30 to Oct 26–29. Lenders charging more to hold AI risk is evidence against buying anyway.
Update 10/31: The doubt spread to the megas. Alphabet had its worst day ever (−13%) on negative free cash flow, Intel its worst since 1974 (−28.6%), and Amazon's trailing free cash flow hit −$42.4B. OpenAI is undercutting Anthropic on price (Replit got 30%+ off).
Update 11/1: The dial now uses method v2 (Yale survey, weighted bubble headlines, capped editorial) and reads 72.
Safety talk and FOMO can share a stage. That combination is still bubble psychology—just a more complicated version of it.
Are companies supplying shares into eager demand?
SpaceX already went public—June 12, 2026, roughly $1.77 trillion and an ~$85.7 billion raise. Dress rehearsal: done.
Anthropic filed its S-1 publicly on Oct 26 (Nasdaq: ANT). IFR (Oct 31) expects the roadshow the week of Nov 9 and a Nov 20 debut raising $75–100B. As of Oct 31 there was no price range on EDGAR, and bankers conceded the $2T target might prove challenging. OpenAI still says no 2026 IPO, and Wall Street executives expect Q1 2027 (NY Post, Oct 30). The 'final boss' date is still unset.
Update 10/31: OpenAI cut its round to about $20B at $1.25T post-money, and some investors trimmed allocations (FT/CNBC, Oct 30–31).
Update 11/11: Anthropic set terms. The roadshow launched Nov 10: 825M Class A shares at $70–78 on Nasdaq (ANT), a $57.75–64.35B offering, plus a $26B Class N private placement to seven suppliers (Google and Nvidia about $10B each, plus AMD, Broadcom, Micron, Microsoft and TSMC). That is up to about $1.40T fully diluted, not $2T. Pricing is expected Nov 18 and trading Nov 19 (CNBC sources; the company's release gives the range, not the dates). Run-rate revenue was about $81B at Oct 31 (S-1/A). OpenAI's Q3 revenue was about $10B on a roughly $50B run rate (CFO Friar, Nov 10), and its round is still open; no IPO date.
The megas are issuing hard. Meta launched a $95B equity raise on Oct 26 and priced $37.5B that night ($25B of common at $625 plus $12.5B of mandatory convertible). Microsoft sold $25B of bonds, its first U.S. deal since 2017. Below them the window is shut: Switch moved to 2027, Nscale slipped, and Oura and Firmus postponed or pulled.
Are new participants stampeding in?
Every bubble eventually needs buyers who were not already in the trade. This is the cab-driver-stock-tip stage—not because cab drivers are uniquely foolish, but because the story has escaped the world of specialists and become common social currency.
I am still not seeing that full stampede in my own life. Lipper keeps flipping: inflows of +$13.37B and +$5.07B in mid-October, then −$6.65B for the week through Oct 28, even as tech funds took in $4.54B. Robinhood's margin book hit a record $22.1B in Q3. Still no new-entrant stampede.
Flows are choppy, not a regime: redemptions, then inflows, then redemptions — with the money going to bonds and cash rather than into stocks.
My current tripwire
My plan today is to remain long for now and go substantially to cash roughly three or four weeks before OpenAI's actual IPO—not its filing, not a rumor, but a credible trading date.
I expect tremendous demand around that event. I would also expect prices could continue rising after the IPO. I am not trying to capture the tippy tippy top. That is too close and too risky for me. I would rather leave some upside on the table than pretend I can identify the exact final print.
This is a working theory, not a blood oath. The sequence could change. OpenAI could delay. The market could break first. The IPO could begin a new phase instead of ending one. What finally pops the bubble? Beats me. Bubbles don't send a calendar invitation.
My thesis is that the procession of enormous IPOs will pull more demand, more bubble belief, and more issuance into the same window. The Horsemen are how I will judge whether the headlines are confirming that thesis.
I am not basing the other side of this plan on a predetermined percentage decline. Maybe the drawdown is 50%. Maybe it is only 20%. Maybe this cycle releases some pressure and an even larger bubble forms afterward. A round number on an index will not tell me whether the dangerous part of the system has actually cleared.
What I will be looking for is evidence in the headlines that leverage is being unwound: collateral demands, forced portfolio sales, a major leveraged firm failing, or some other event showing that people who expected to wait are no longer being given that choice. The Horsemen help me read the buildup. Forced unwinds help me read the damage.
If I step aside, I will give myself no more than six months before buying back into the market. That is not a forecast that the decline will end in six months. The unwind could take a year or two, and I may buy back only to ride part of it down. The rule is a fail-safe against permanent exile from the market.
I am not planning to short the United States, the stock market, or the AI future. I am trying to participate in the boom, step aside from some portion of the damage if my read is right, and then force myself back into the bright future I believed in from the beginning.
What not to do
Charlie Munger liked to invert a problem. In his 1986 Harvard School commencement speech, he borrowed Johnny Carson's approach: instead of pretending to have a formula for happiness, identify the behaviors that guarantee misery.
I cannot tell you exactly how to navigate a stock-market bubble. I can tell you how to improve your odds of getting burned.
Use leverage
If you want to go broke, use leverage.
Leverage multiplies a decline, but the larger problem is that it can take away your ability to wait. In simplified arithmetic, 2x exposure against a position that falls 50% wipes out the investor's starting equity before financing costs.
Nobody forces you to sell when you are up. You get forced to sell after the position has moved against you, when patience would be most valuable.
Aschenbrenner is the current textbook example—not because his AI thesis was stupid, but because intelligence does not repeal arithmetic. His fund reportedly produced extraordinary gains using leverage. Then AI-linked stocks fell. Reporting indicates the fund suffered heavy losses, faced collateral demands, and sold or transferred most or all of its public-equity portfolio while retaining private investments.
You can be one of the smartest people in every room you enter. You can be directionally right about the most important technology in the world. Leverage can still wreck you.
Sell because you are afraid
The person who stays unleveraged and stays in the market is not necessarily burned by a temporary 50% decline. The person who watches a life savings fall by half and sells at the bottom out of fear has turned that decline into a permanent loss.
A stock-market crash will not prove that AI was fake. Do not confuse a collapse in price with the collapse of the underlying technological thesis.
Remain unafraid. Our future is bright.
Stay out forever
If you decide to step aside, decide in advance how you will return. Maybe you will sell too early. Maybe you will buy back too early. Maybe the market will never give you the crash you expected.
The amateur fail-safe is not elegant, but mine is simple: I am not going to miss this forever. The objective is not to win an argument with the market. It is to remain solvent and participate in the long-term future.
Go investigate for yourself
The Four Horsemen are not commandments. They are four doors into the research: valuation, bubble psychology, issuance, and inflows. Follow the sources. Read people who disagree. Decide whether the evidence fits the framework or breaks it.
Then send this to the friends you already argue with about this stuff.
You really think the biggest technological breakthrough in human history won't create a bubble?This is a personal market thesis and a record of my own thinking, not individualized financial advice. I can be wrong, and markets can remain irrational longer than any plan expects.