A commodity is deflating at record speed while the debt financing it is being built at record speed. That, George Noble said, is the artificial intelligence trade.
Most of the bear case on AI is about valuation. His is about arithmetic: reverse-engineer the trillions of capital spending into the gross profits and then the revenues needed to justify them, and ask whether the return on capital can be there at all.
"I said, show me the cash flows. You can't."
Noble was Peter Lynch's foreign stock pick at Fidelity in early 1984 and ran the firm's first international fund. He is now managing partner of Noble Capital Advisors, and he was on the same program a couple of months ago making a milder version of this argument.
The full segment is covered here so you can skip it.
Here are the 6 arguments that matter.
👤 Guest: George Noble, Managing Partner of Noble Capital Advisors, who ran Fidelity's first foreign stock fund from 1984
🎙️ Hosts: Tom Keene and Paul Sweeney, who anchor Bloomberg Surveillance
🧩 Other segments: Dominic Konstam of Mizuho; Marina Zavolock, Chief European Equity Strategist at Morgan Stanley; and Sinjin Bowron of Beach Point Capital
📰 Published: 15 September 2026 on the Bloomberg Surveillance feed
🟣 Apple Podcasts | 🔗 Episode page | ⏱️ length not available
Key Takeaways
The AI build-out is a deflating commodity funded by debt accumulating at record speed
Every new model arrives, and then a Chinese lab does it for 95% less
His comparison is not the dot-com bubble alone but dot-com merged with the housing crisis
The test he keeps setting is cash flow, and he says nobody has met it
Reverse-engineer the capital spending into required gross profits and required revenues, and the return on capital does not appear
Being right about the technology did not save the companies last time
Internet traffic compounded for 25 years and Nortel and Global Crossing still went bankrupt
He has a name for the mechanism: narrative dominance, the successor to fiscal dominance
He compares the Treasury Secretary to the man who defended sterling in 1992, and says the market attacks a defended price
Asked what advice he would give him, his answer was to resign
1. Fidelity, 1984
Keene opened on Noble's own history rather than on markets. He and Mark Mobius, Keene said, took Sir John Templeton's approach to large-cap international value investing and made it dance.
The job came because his mentor had a gap to fill. Peter Lynch had insatiable curiosity and needed a foreign stock specialist, so in early 1984 they asked Noble whether he had ever been abroad.
"See, I've been to Tijuana. I've been to Montreal." That was his answer, and he got the job anyway.
He was sent overseas with a brief and no team. "So they sent me off to here, and they said, get your glove, because end of the year of 84, we're going to start our first foreign stock fund. I was it."
He credits the culture rather than the plan. Under Ned Johnson, "They gave you opportunity, threw you into the fire, and let's see what you got."
2. Bad to Worse on AI
Keene turned to the present and told Noble he is scathing on artificial intelligence, then asked what listeners holding it in their retirement plans should do into the autumn.
His opening was that nothing has improved since he was last on. "Things have gone from bad to worse."
The structural problem, stated in one sentence. "You have a commodity that's deflating at record speed with debt that's being built up at record speed. What could possibly go wrong?"
The deflation is continuous and visible. Every week another model arrives, and then, in his words, "the Chinese can do it for 95% off" — with talk now of the President banning Chinese AI.
His verdict is unchanged from the last appearance. "I think this is going to be the biggest misallocation of capital in history."
On whether it is worse than the dot-com bubble, he relayed a framing he read and admired: it is as if you took dot-com and merged it with the global financial crisis and its Ponzi housing finance. "history rhymes, it doesn't repeat."
3. Show Me the Cash Flows
Sweeney pushed back with the bull case as he understands it: AI is a genuinely new way of computing and processing information, requiring enormous compute and therefore enormous capital. He asked what part of that Noble is not buying.
Noble separated the technology from the security, and used his mentor's framing to do it. He uses AI himself and gets attacked on Substack and X for it, but "It doesn't mean it's a good investment." Peter Lynch, he said, always talked about the difference between the product and the stock.
The test is the reverse-engineered income statement. Take the trillions being spent on capital equipment, work out how much gross profit is needed to justify it, and then what revenues that implies. "Liars figure, but figures don't lie."
The binary outcome he sets up: either an acceptable return on all that capital, or "a commodity business, and there's going to be no margins."
The historical case he cites is a bank credit analyst's. Peter Barrett's observation, which Noble repeated approvingly, is that the bulls of 1999 were right about the internet — traffic compounded at about 43% a year for 25 years, going up by something like 25 million percent — and it "Didn't stop Nortel from going bankrupt and Global Crossing from going bankrupt. These are two different things."
4. Narrative Dominance
Keene asked whether it is easier or harder to avoid losing money now than it was in his Fidelity years, given the volume of financial noise.
Noble said harder, and coined a phrase for why. He is tired of hearing about fiscal dominance — "If I hear that term one more time, I'm going to put my fist through the wall."
His replacement term is narrative dominance. He tied it to the idea of a post-truth society: "We're in the post-truth stock market. It doesn't matter what reality is."
The mechanism he describes is instruction plus liquidity. Investors are told what to believe, and enough money is put in to make the telling work — "And we're going to put in enough liquidity, spike the drinks enough."
5. A Market of Stocks
Keene asked whether the largest technology companies can be separated or should be treated alike.
He does not treat them alike. They are all a little different.
One of them he will not name on air. "The electric car company that shall not be named uniquely stands out for being extraordinarily unattractive."
The scale of the short is what he stresses. "I've never seen a short at such scale, a trillion five." He put its affiliate SpaceX in the same category, calling both unbelievable companies and atrocious.
At the other end he named Google and moved on. Fine, he said, with everything else in between.
The conclusion is a stock-picker's. "So again, it's a market of stocks. Know what you own. Peter Lynch, please call your office."
On the economy, his call is explicit and it is not recession. Asked how worried he is given how much growth is AI-related, he said "My call has not been for recession. Ours has been for rotation, not recession." With trillions being spent and a 7% deficit, he asked, where is the recession going to come from? The problem he sees is a global increase in the cost of capital.
What he is watching for the turn is credit, not equity. He named the credit default swaps on Oracle as the sort of thing to watch for the moment the market says no more.
Sweeney supplied the earnings arithmetic. At least half of this year's S&P earnings growth, on what he hears from strategists, has been AI spending and margin related — which, he said, does not bode well for equities. Noble agreed, and answered with hockey: "Let's go to where the puck is going, not where it is right now."
What surprises him is how well the market has held. Given the surge in bond yields and oil prices, he said, anyone told two months ago that yields would be at five and crude where it is would have expected equities down a lot.
6. Bessent as Norman Lamont
Keene asked whether the yield move breaks something suddenly, as leverage did in August 1998, or ebbs into a slower cataclysm. Noble's answer went to Treasury policy.
The usual contraction mechanism is broken. Falling stock prices normally make corporate managers cut back, which produces the slowdown. With prices staying up, "that sort of information content from the pricing was missing" while fiscal spending keeps driving the economy.
He read the Treasury's buyback announcement as a sell signal. On 19 August, when the plan to buy bonds was announced, "I said, no, this is a sign to short bonds, not buy bonds."
The comparison he draws is to Britain in the early 1990s. "Scott Bessent is the modern-day equivalent of Norman Lamont."
He also disputes a piece of Bessent's reputation. "I find it really rich that Mr. Bessent, who, by the way, is wrongly credited for having been involved in the Soros-Druckenmiller takedown of the Bank of England, had nothing to do with it at all."
The lesson he takes from that episode is about defended prices. "And by that, I mean what got the Bank of England in trouble is when you defend an artificial price that's not justified by the fundamentals, the market attacks you." He says that is what is being attempted in the bond market now, and in the AI complex as well.
So his forecast is two attacks at once. "I think the market's going to continue to attack the bonds. I think yields are going to continue to go up. I also think the market's going to attack AI, and I expect the whole edifice to collapse."
On the Federal Reserve he claimed no edge. His crystal ball on Fed watching "is no better than anybody else's", and he does not spend time on it because he is not good at it and knows nobody who is.
He still described the trap facing the chair. If Kevin Warsh delivers a dovish hike the market reads it as the end and rates go up; if he signals two or three more hikes the market will not like that either. "So I think he's in a thankless position."
The last word was the advice he would give the Treasury Secretary. Asked what Bessent should do, Noble's answer was one word: "Resign."
Bonus Insights
Keene's introduction is a piece of publishing history in itself. He mentioned Justin Baer's book about Fidelity, said he read it cover to cover, and noted with approval which internal subjects the book left out.
Noble takes the criticism for using the product he is short. People attack him on Substack and X for using AI, which is exactly the product-versus-stock distinction he says they are missing.
He is careful about which of his views are original. The dot-com-plus-financial-crisis framing and the Nortel comparison are both credited to other people by name before he endorses them.
Noble's bottom line is that the AI build-out and the Treasury's defense of the bond market are the same trade seen twice — an artificial price held up without fundamental support — and that the market will attack both, with yields continuing higher and the AI complex collapsing under capital spending whose cash flows nobody can produce.
Products, Companies & Tools Mentioned
Fidelity (Where Noble was handed the first foreign stock fund at the end of 1984, under Ned Johnson and alongside Peter Lynch)
Nortel and Global Crossing (The comparison he endorses: the internet bulls of 1999 were right about traffic and these companies still went bankrupt)
Oracle (The credit default swaps he watches as the signal that the market has stopped funding the AI build-out)
Google (The one large technology company he was willing to call fine)
SpaceX (Grouped with the electric car company he declines to name; he called both atrocious as investments)
Watch the full episode:
If this was worth your time, send it to someone who has to have a view on this.
Get the latest market chatter as it happens: