Dan Niles is running a book with a lot of shorts in it, because the median drawdown into a midterm election is about 10% and this year has more going against it than seasonality.
The standard reply to a 5% 10-year Treasury yield is that stocks have lived with rising yields before. His answer is that each new problem is another straw on the same camel, and that the one that has been carrying the market since 2022 — artificial intelligence spending — now has questions attached to it.
"So for me it's the market keeps pitching. But you don't have to take a swing."
Niles founded Niles Investment Management and runs money there. He published the same warning in late July, and before that called an 8% "speed bump" at the end of June that turned into a 29% fall in semiconductors inside five weeks.
The full segment is covered here so you can skip it.
Here are the 5 calls that matter.
👤 Guest: Dan Niles, Founder and Portfolio Manager of Niles Investment Management
🎙️ Hosts: Carl Quintanilla and Sara Eisen, who anchor CNBC's Squawk on the Street
🧩 Other segments: Marc Casper, Chairman and CEO of Thermo Fisher Scientific; John Collins, global co-head of M&A at Morgan Stanley; Matt Hougan, CIO of Bitwise Asset Management; and CBOE's Oliver Renick on gold options
📰 Published: 15 September 2026 on the Squawk on the Street feed
🟣 Apple Podcasts | 🔗 Episode page | ⏱️ 8 min
Key Takeaways
A 6% 10-year Treasury yield is not an outlandish target, on the arithmetic he uses
6% deficits as a share of GDP, $40 trillion of debt against $33 trillion of output, and hyperscalers now issuing debt alongside the government
His three rules all point the same way: don't fight the Fed, the bond market or midterm seasonality
The median drawdown from end-July to 9 November in a midterm year is about 10%, double a normal year
He is carrying a lot of shorts and is content to stand still
Meta is the long, at 17 times next year against peers in the low 20s
The three things it was missing — a competitive model, an API and an agent — all arrived inside a fortnight
Slowing the pace of frontier training is a capital-spending cut whatever else it is
Security spending rises, but it does not need the same number of GPUs as training does
The states are the part of the AI trade he says got worse in three weeks
Power access is the acknowledged bottleneck, and Pennsylvania and Texas are both slowing data-center approvals
1. A 6% 10-Year
Quintanilla opened on rates, with the 10-year at five and a Deutsche Bank note that day raising the question of a terminal destination of six. He asked whether that was too much.
Niles said no, and started from the level rather than the move. "Rates historically, if you look at where the 10-year is, they're pretty low."
His case is the supply of debt and who is issuing it. The United States is running 6% deficits as a share of GDP, which he called the highest outside a major war, with "40 trillion in debt against 33 trillion in GDP".
The new competitor for that money is the technology industry. "You've got hyperscalers now competing with the government to issue debt 6%."
The policy options are narrower than they were. With debt-to-GDP where it is, there is not the same capacity to intervene to hold rates down, and on his reading the current composition of the Federal Reserve does not want to: "They let the market kind of dictate where rates want to go."
The anchor desk had set the same problem up before he came on, noting a heavy auction calendar and questions about who buys it, with big technology companies selling debt into the same market.
2. Straws on the Camel
Quintanilla's follow-up was the bullish version: stocks and yields have risen together in famous eras before, so maybe equities do not mind.
Niles allowed the possibility and then declined it. Anything is possible, he said, "But you can also look at it and say, hey, there's more straws being put on this camel's back."
The straw he is most focused on is the AI trade, which has powered the market since the end of 2022 "And we had this thing called ChatGPT show up."
His framework is three rules, and all three say the same thing. Don't fight the Fed — "It looks like we're going to go into a hiking cycle." Don't fight the bond market, where the conversation is now about 6%. And don't fight seasonality.
The seasonal number is the one that does the work. "The median drawdown is about 10% from end of July through November 9th" in a run-up to midterms, which he said is double the figure in non-midterm years.
The conclusion is inaction rather than a position. "So for me it's the market keeps pitching. But you don't have to take a swing. And I don't mind just not taking a swing right now."
3. Short, Except Meta
Asked directly how he is positioned, Niles said he has been writing about this since late July and is worried about a 10% drawdown between now and the midterms.
"So I've got a lot of shorts on."
Meta is the exception, and he is long it. He thinks the sentiment can keep re-rating, and that the company has cleared its obstacles.
The legal overhang went first, with the lawsuit against the attorneys general settled.
The monetization gap closed inside a fortnight. Meta is doubling capital spending, and unlike Google, Amazon or Microsoft it had no public cloud and no API to sell the output through. Then Meta Spark 1.3 arrived and ranks well at low cost, an API followed two weeks ago, and last week an AI agent launched into a base of "3.6 billion daily active users".
The valuation is the last leg. Meta trades below the S&P and well below peers in the low 20 times, at "17 times calendar 27" on his numbers.
None of that makes him want to be long the tape. He compared the setup to the end of June, when he published a call for an 8% speed bump: "And that turned into one hell of a speed bump." Between 22 June and 29 July, semiconductors fell 29%.
His read now is that the list of problems is longer. "And now I feel like you've got even more issues now than you had back at the end of June." Quintanilla noted that the SOX is still down by a fifth from its June high.
4. Pacing Is a Capex Cut
Eisen pushed back on the idea that an AI sell-off is uniformly bad, arguing that the non-AI winners — software, crypto names — are where the opportunity shows up, and that investors should be more discerning about winners and losers.
Niles agreed with the framing and then sharpened it. Being discerning is exactly right, he said, and has been for a couple of months.
But he does not think the aggregate effect is neutral. If the industry slows the pace of advancement — what the labs are calling pacing — "if you are cutting back at the pace you're training these new models, by definition, that's going to slow down capex on that piece of it."
Security is the offsetting spend, and it is not the same spend. More money will go to security and related work, "But that's not going to require the same number of GPUs as if you're going to go ahead and train."
So the selectivity call has two sides. Security looks increasingly like a good place to be; demand for training hardware probably is not where it was, given what Sam Altman, Anthropic, OpenAI and Elon Musk have said about slowing the pace.
Quintanilla's own question was whether pacing flows through to capital spending at all, which is where Niles' answer started.
5. The States Turned
The second half of the AI argument was about power, and it is the part Niles said has deteriorated in a matter of weeks.
He named the contradiction directly. The industry says access to power is the gating item for expansion, and states are now moving to slow the ability of data centers to get power: "You can't have it both ways."
He named Pennsylvania and Texas — Texas being the one that surprised him — as places where approvals are being slowed.
Eisen disputed the generalization. "I don't know, I think there are enough states that are welcoming it."
Niles conceded the point and kept the direction. There are welcoming states, but "There's a lot more issues with Pennsylvania, Texas, other states that are slowing this stuff down that on the margin, it's certainly not better than it was three weeks ago."
Quintanilla listed the other side of the ledger — West Virginia, Virginia, Wyoming, Alaska — and said it is not clear to him that the case for reduced data-center or capital spending has been made.
The other thing keeping him cautious is oil. Niles said he does not expect a resolution to high oil prices through the midterms and potentially longer, and that high oil and high bond yields bite with a lag: "It's not day one. It's three six months out."
Bonus Insights
He treats his own published calls as the benchmark. Both the late-July warning and the end-of-June speed-bump note were cited by date, with the outcome attached rather than left implied.
The disagreement with Eisen was conceded rather than argued. When she said enough states are welcoming data centers, he agreed there are, and narrowed his claim to the change at the margin over three weeks — which is a different claim from the one she objected to.
Quintanilla's framing did most of the work in the segment, twice handing Niles the bullish version of the argument (stocks can rise with yields, states are still welcoming) before he answered.
Niles' bottom line is that he would rather hold cash and shorts than fight a Federal Reserve that is starting to hike, a bond market discussing 6%, and a midterm season whose median drawdown is 10% — with Meta at 17 times as the one long he is willing to carry through it.
Products, Companies & Tools Mentioned
Meta (His long: the lawsuit settled, Meta Spark 1.3 ranking well at low cost, an API two weeks ago and an AI agent last week into 3.6 billion daily active users, at 17 times calendar 2027)
Google, Amazon and Microsoft (The peers that already had a public cloud and an API to monetize AI spending through, which is what Meta was missing)
OpenAI and Anthropic (Named alongside Sam Altman and Elon Musk as the voices calling to slow the pace of advancement — which, on his reading, is a cut to training capex)
Deutsche Bank (The note that morning raising the question of a 6% terminal destination for the 10-year)
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: