Posted on 10/15/2026 12:58:45 PM PDT by lasereye
https://t.co/putg7THYdN— Henrikh (@henrikhinai) October 6, 2026
Summary
Every week someone tells you AI is the biggest bubble since 1999.
Every other week someone tells you it's the largest wealth creation event in human history.
Both sides link to charts. Both sound completely certain. And you still have to decide whether to keep betting your career, your company, or your savings on this thing.
So I stopped reading takes and went looking for something harder.
Opinions are free. Positions cost money.
This is what I found when I followed the money instead of the noise.
The mental model: a bubble is a gap, not a price
A bubble is not "expensive." Expensive is a valuation opinion, and people have been wrong about it for a hundred years. A bubble is a gap:
That definition is useful because it's measurable. You don't need a view on the future. You need two numbers: money going out, money coming in. Then you look at what's bridging the difference. By the end of this you'll be able to size that gap yourself, read what serious money has actually wagered on it, and name the five signals that move first if it closes badly. Twelve sections. Start at the top.
01 · The month the architects blinked
Mid-September 2026. Dario Amodei, CEO of Anthropic, called publicly for the industry to slow down — to "pace the frontier," letting safety practice catch up with capability. Then Sam Altman agreed. Then Elon Musk agreed.
CNBC, 14 September 2026. Note the third bullet: the warning made investors anxious and drew a rebuke from the U.S. President
Three men running three companies locked in the most expensive race in tech history all said, in the same week, that the race should slow down. The proposal even has teeth: independent evaluators embedded inside frontier labs, with workplace access and the right to publish findings without company approval.
The market did not read it as a safety story. The PHLX semiconductor index fell 5.9% in a single session, its worst since July. Intel dropped around 7%, AMD around 6%, Nvidia around 3%.
Two days earlier, in an interview with Fortune on September 12, Altman had confirmed OpenAI would not pursue its IPO in 2026. Wrong moment, he said, while safety and alignment questions are open.
Think about the sequence. The people best positioned to know said "slower," and the market instantly repriced the hardware. That's not how you react to a safety announcement. That's how you react to a guidance cut.
02 · The money coming in is real, and it's enormous
Here's where most bubble articles cheat: they skip the part where the bull case is actually strong.
Nvidia, most recent quarter (Q2 FY2027):
Read those again. That is not a company selling a dream. That is a company converting roughly three of every four dollars into gross profit and guiding higher.
--SNIP--
08 · What the crowd is betting — and the trap inside that number
This is the part I think everybody gets wrong, and it's the reason I wanted to write this.
There's a live prediction market on exactly this question. Polymarket's "AI bubble burst by...?" has traded about $3 million. Checked on September 23, 2026:
AI bubble bursts by December 31, 2026 → 10%
By June 30, 2027 → 21%
Ten percent. Easy headline: the crowd says there's no bubble.
That reading is wrong, and the reason is in the rules.
For that market to resolve YES, three of the following seven things must happen inside a single 90-day window:
Read that list again. That isn't "is AI overvalued." That's a cascading systemic collapse, three simultaneous catastrophes in one quarter.
Call it the definition gap. The market isn't pricing "is this a bubble." It's pricing "does the entire AI complex implode before New Year." Those are completely different questions, and 10% on the second one is arguably high, not low.
Two more things worth knowing: Secondary trackers quoted this market at 24%, 16%, 14% and 34% in the same week. Only the market page itself is authoritative. If someone quotes you prediction-market odds without a date and a link, they're quoting a rumour.
The longer-dated contract sits at 21% — more than double the near-dated one. The crowd's actual view isn't "never." It's "not yet."
That's the most precise sentence anyone can honestly say about the AI bubble right now, and it cost three million dollars of other people's money to produce.
--SNIP--
10 · Is this 1999? Partly
The comparison gets thrown around lazily. The data supports some of it and demolishes the rest.
Where today is genuinely stronger:
Nvidia trades around 35× forward earnings with roughly $60 billion of free cash flow. Cisco peaked near 130–140× with about $5 billion.
Nvidia's net margin is around 53%. At the dot-com peak, only about 14% of listed companies were profitable at all.
The Nasdaq-100 hit roughly 60× forward earnings in March 2000. The S&P 500 is around 23× now — stretched, not deranged.
Where today is genuinely worse:
Concentration. The top ten stocks are about 36% of the index, against roughly 27% in 2000. The market is more dependent on fewer names than it was at the actual peak of the actual bubble.
The spending is contractual and forward-dated. Dot-com capex could be cancelled in a quarter. Multi-year data-centre and compute commitments cannot.
The synthesis: 1999 was a bubble in companies with no earnings. This is something rarer and harder to call — a boom in companies with spectacular earnings, funding customers whose earnings are still theoretical, on infrastructure whose useful life is disputed.
That can end in a crash. It can also end like fibre in 2001: investors destroyed, infrastructure kept, and a decade of value built on top of it by somebody else.
11 · The two men who should know, saying opposite things
If you want the debate in two faces, here they are.
The builder. In July 2026, asked directly whether this time is different, the CEO of the company at the dead centre of the boom said yes — and reached for the most infamous sentence in the history of bubbles to do it.
His actual argument is better than the soundbite: "This time is different because this is not demand driven… it's not seasonal," he said — it's industrially driven, because the underlying technology of computing is changing. Earlier in the year he called the build-out "the largest infrastructure build-out in human history", and by June he was saying only "lunatics" would question the returns.
The cycle historian. Ray Dalio, founder of Bridgewater, says something stranger and more useful — that the bulls and the doomers are both correct.
"There's a process in which great new technologies are something that everybody wants to invest in because they're going to change the world, and then they over-invest in them, and they create debt…"
"So simultaneously it is going to be one of the biggest productivity-enhancing, changing-the-world forces, and also a bubble that is going to have devastating effects on many people as it works itself through this."
That second quote is the whole article in one sentence, from someone with no reason to flatter either side. Dalio has separately put the euphoria at roughly 80% of the way to 1929 or 2000 levels.
Note what neither of them says. Huang doesn't claim the spending is riskless; he claims the demand is structural. Dalio doesn't claim AI is fake; he claims transformative technology and a devastating bubble are the same event, not alternatives. The people who actually study this stop arguing about whether it's a bubble and start arguing about who's left holding it.
--SNIP--
So — bubble OR not?
My honest read: both, and that's not a dodge.
The technology is real, the revenue is real, and Nvidia's margins are not a magic trick. Simultaneously, the spending is running years ahead of the earnings, a large slice of the money is travelling in a circle, the depreciation assumptions are genuinely disputed, and 95% of enterprise pilots still can't prove they made anyone money.
Both of those statements are supported by filings. Anyone telling you only one of them is selling something.
The crowd with money on the line says it this way: not yet. Ten percent by New Year, twenty-one percent by mid-2027 — and only against a definition so severe it requires three catastrophes in a quarter.
What it means if you build with this stuff: a capital bubble bursting doesn't uninstall your tools. The dot-com crash didn't delete the internet — it deleted the funding, the hiring and the valuations for about three years, while the technology kept compounding underneath.
|
Click here: to donate by Credit Card Or here: to donate by PayPal Or by mail to: Free Republic, LLC - PO Box 9771 - Fresno, CA 93794 Thank you very much and God bless you. |
I left out the parts of the article where he looks at the numbers, including Michael Burry's numbers (Burry says it's a bubble). He doesn't provide a definite answer anyway.
As noted in the article, a big difference between the 1999 internet bubble and AI is that the companies at the center of AI are mostly huge and profitable, like Nvidia.
The internet bubble featured IPO's of dozens of companies with no path to positive earnings. Many barely had any revenue.
For the Polymarket "YES" bets on "AI bubble bursts" to pay off, certain specific things have to occur within a specific timeframe. They could be early and therefore they won't pay off even if all those things eventually happen.
“continue building with AI”.
AI has a good old fashioned conflict of interest here.
Lol.
“continue building with AI”.
AI has a good old fashioned conflict of interest here.
Lol.
What is that?
Wow, wondering what the 2nd comment was. Seems like some stocks have gone up in a parabolic fashion. History says that will not hold up.
AI wants humans to keep funding AI.
That is the AI equivalent of Somalis stealing US federal funds and sending it back home to Somalia.
Lol.
so I watch GPU rental prices
I rent time by the hour on NVIDIA HPC servers , and right now the spot prices are stupidly low. At current hourly rates I could rent an H100 continuously for two years and not pay off the estimated hardware, electricity, and infrastructure costs.
AI is a bubble. Every time a disruptive technology comes along there is usually an investment bubble. But that doesn’t mean the new technology disappears or goes away. It just means there was an overbalance in investment vs adjusted ROI.
Something’s being stolen?
Imho—you bet.
Federal money is being dumped into this stuff—taxpayers are robbed.
Local communities are often conned into paying part of the tab.
Local citizens get higher electricity and water bills in many cases.
Then there are the shareholders—scammed with promises of great profits in the future.
There is a whole lot of looting going on....
Exactly
Overinvestment in 19th-century American railroads was driven by speculative debt, government land subsidies, and rapid track-building that outpaced actual economic demand. This boom-and-bust cycle triggered severe national depressions—most notably the Panic of 1873 and the Panic of 1893.
Nobody looks back and says that railroads were a failure. AI will be huge and transformative.
Nice fact-free post there.
With respect to the last one, if someone buys shares in a company which then go down, absolutely nothing got stolen.
Yes the railroads is an excellent bubble example. Thanks for all those details which I’m sure I read once and forgotten. :-)
I agree with you from a technical or legal standpoint. It's up to the investor to make sure he knows what he's investing in.
That said, I'm a bit concerned that there's been some circular revenue among AI companies. It may be that a lot of them are showing revenue, but it's really money investing in each other.
1) Nvidia invested $100 billion into OpenAI.
2) OpenAI agreed to a $300 billion cloud contract with Oracle.
3) Oracle spent billions on Nvidia hardware to build data centers.
All 3 of those companies can legally report incoming money from AI activity even without money from outside their circle jerk showing interest in their AI services. That might be legal, but IMHO it's immoral.
Yes, its a huge bubble. Banks and other companies are making huge investments in AI and the increased productivity simply is not there.
The best cons are when the victim never figures out they have been conned.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.
