# When Everyone Agrees Something Is Doomed — Notes on Tom Lee and Dan Ives

> Listening notes from The Compound and Friends, 2026-09-18: how market narratives get repriced, the interests behind the AI safety debate, and how to turn 'when does the bull market end' into conditions you can actually check. Educational, not investment advice.

Published: 2026-09-18
Locale: en
Tags: market narratives, AI capex, investment judgment, earnings, tech stocks


![An empty old theater, two chairs on stage under a single warm spotlight, rows of seats receding into the dark](/covers/compound-2026-09-18-tom-lee-and-dan-ives-explain-everything-cover.png)

> The world says Lord Mengchang could attract men of talent, and that talent flocked to him, and that in the end he escaped the tiger-and-wolf state of Qin by their strength. Alas! Mengchang was merely the chief of cock-crowers and dog-thieves. How does that amount to attracting talent?
>
> —— Wang Anshi, "On Reading the Biography of Lord Mengchang" (Northern Song, c. 11th century, translated by the author)

Wang Anshi's essay runs about eighty characters and does one thing from start to finish: it takes "everybody says so" apart. The world called Mengchang a magnet for talent — three thousand retainers. Wang asks: the two who actually saved him were a man who could imitate a rooster and a man who could crawl through a dog hole. Is that talent?

I kept coming back to that passage while listening to The Compound and Friends from September 18, 2026. It was recorded live, Josh Brown and Michael Batnick with Tom Lee and Dan Ives, a full house. Over two hours the thing that stayed with me was not any bullish argument. It was the two of them counting up how many consensus views had been overturned in a single year.

## What the episode is about

At recording time the market was pricing a 25 basis point hike at roughly 90% odds. Over the previous weekend, Dario Amodei of Anthropic had published a piece saying he had seen things in his own labs that worried him, calling for outside observers and a slower pace of frontier development. Semiconductors were 19% off their highs.

From there the conversation went out to the scale of the capex wave, the politics of AI regulation, robotics and "physical AI" as a structural question for the economy, and finally a run through individual names. Tom Lee co-founded Fundstrat, which was celebrating its twelfth anniversary that day. Dan Ives left Wedbush in July and co-founded an AI-focused merchant bank with Yorkville. Both are bullish, and their reasons are two different structures underneath — which is the part I found most interesting.

## The main points

**A year of bad news, and the index is 2.6% off its high.** The host read the list out loud: the Strait of Hormuz closed, oil spiking, tariffs, sticky inflation, the ten-year at 5%, the Fed hiking, a frozen housing market, rumors of a slowdown in frontier models. Each one was called a black swan at the time. The index is 2.6% off its all-time high.

![On the left, a tall stacked list of eight bad-news events; on the right, a single index bar rising almost to the top, with only a hairline gap left between its top and the all-time high, so the list looks long while the gap looks tiny.](/figures/one-year-of-bad-news-versus-a-small-gap-en.svg)

**The index is up 10%, earnings are up 25%, so the market got cheaper.** Tom Lee used that arithmetic to answer the "it has run too far" view. The sentence he added matters more than the numbers: these companies put up those figures after absorbing that whole list because management has been in permanent crisis mode cutting costs, rather than sitting in the boardroom saying everything is fine.

**Narratives reprice faster than fundamentals.** In March, software-as-a-service was declared dead. Salesforce, ServiceNow and the rest were in drawdowns of 20% to 60%, and not one of them had missed a quarter or shown any change in the business. Cybersecurity went through the same cycle: Anthropic announced a security offering in April, and people at the conference were saying the industry was finished. Six months later CrowdStrike, Fortinet and Palo Alto are each up more than 100% on the year. Dan Ives called the March story a bad fairy tale.

**The two of them read the safety letter differently.** Tom Lee said industries always outgrow their regulation, that AI has no self-regulatory body at all, and that self-regulating now — while communities are protesting data centers — is a smart move. He drew the wireless parallel: people feared phone radiation causing brain cancer, transmit power went from 1 watt to 25 milliwatts, and the industry survived. Dan Ives put it more bluntly: you reach the penthouse and then you stop the elevator so nobody else comes up. He pointed out that the frontier labs' lead is large, and a slowdown lets the software companies and the startups close the gap.

![A cutaway of a tower whose top floor is lit, with a bar sealing the elevator shaft partway up, so every upward arrow from the lower floors stops right below that bar.](/figures/stop-the-elevator-at-the-top-floor-en.svg)

**Two things end a bull market, and both men named conditions.** One is a genuine bubble — the day everyone decides there is only one AI model worth using and capex goes to zero. The other is monetary policy; Tom Lee said the Fed is the reason bull markets end about 80% of the time. On debt bubbles he offered a test: bubbles come from money so cheap that bad projects get funded, and right now the cost of money for data centers is rising, so the market has already applied its own brake.

**For robots to grow the economy, they have to become consumers.** This was my favorite piece of reasoning in the episode. Tom Lee framed the economy as two levers, labor and capital. Add robots as a third productive unit that itself consumes and pays tax, and you can grow without inflation. Then he supplied his own failure condition: if robots only replace human jobs, the economy shrinks and the doomers were right. He pushed the same question in both directions.

**Consumer discretionary has lagged for years because of how the category is built.** The host asked a sharp question: equal-weight consumer discretionary divided by the equal-weight S&P keeps making new lows, so does the sector still matter? Tom Lee's answer sidestepped the business cycle. Discretionary is the catch-all — a company gets sorted into industrials, tech, healthcare or staples first, and whatever is left lands in discretionary. iPhone spending counts as tech, a Costco trip counts as staples. What is shrinking is the residual bucket.

![Two panels side by side: on the left, a one-way arrow from a person to a robot sits above a small circle; on the right, a closed loop of robot, output and consumption sits above a much larger circle, and the two circles differ sharply in size.](/figures/robots-replace-versus-robots-consume-en.svg)

## Going further

### "Earnings are great. Why isn't my stock moving?"

This is the question I have been asked most over the past few years, and the one I struggled with longest. The arithmetic in the episode gives a clean way in: the change in price equals the change in earnings times the multiple the market will pay. Earnings up 25%, price up 10% — the 15% in between is multiple compression.

![Two bars of clearly different heights stand side by side, earnings on the left and price on the right, with the shorter price bar's gap to the taller one marked by a bracket labelled multiple compression.](/figures/earnings-up-price-up-less-multiple-compression-en.svg)

Compression itself splits two ways. In one, the market is discounting the future: it thinks the earnings won't hold, that this is a cyclical peak. Semis down 19% while Nvidia reports quarter after quarter is the case Dan Ives puts in that bucket, and he noted Nvidia trades at 16 times while Costco trades at 50 and Walmart at 48. In the other, earnings quality is the problem, and the market is reacting to something the statements have not shown yet.

Telling them apart happens outside those two numbers. What I now ask is: if this is a cyclical peak, which way should inventories and orders be moving? If it is earnings quality, the cash flow statement and the income statement should start diverging. When neither is moving and only the price is, the issue lives in the multiple, and multiples are a product of sentiment and flows. They come back.

Tom Lee added one thing: plenty of fund managers hold zero or underweight AI exposure because they believed it was a bubble. That money is still outside. I can't verify the claim, but it reminds me that price reflects the views of those already in the market, not the views of everyone.

### "The news says AI will wipe out this industry. Should I sell?"

March's software selloff and April's cybersecurity selloff were the live version of this question. I panicked too. I thought about selling. Looking back, there are a few ways to sort it out, none of which require predicting anything:

First: has any company's reported numbers reflected it? In March the answer was no — nobody missed, nobody guided down. Price down 60% with zero change in the financials means the narrative did the falling.

![A flat grey line shows earnings holding steady, while a blue line plunges in March and climbs back above its starting level over the next six months, so the two lines diverge and then cross on the same timeline.](/figures/narrative-falls-while-numbers-stay-flat-en.svg)

Second: can you write down a timeline for the replacement? Swapping out a CRM means data migration, process rewrites, compliance review, retraining. Those things have real durations. A displacement thesis with no timeline usually stops at "technically possible."

Third: whose revenue goes up? If software companies get replaced, where does the money go, who books it, and when does it show up in whose statements. When that chain won't connect, the narrative is financially empty.

Dan Ives' read became the later consensus: software companies sit downstream of AI as beneficiaries. One limit worth adding — he conceded that Adobe and Intuit face structural issues of their own. So "the narrative killed the wrong name" does not extend to every name. The check runs company by company.

### "So when does the bull market actually end?"

The part of the episode I valued most was that both men stated failure conditions: capex collapsing, the Fed deciding the economy is overheating and acting on it, and politics voting data center projects down. Those are observable events rather than feelings.

After that episode I did one thing: I wrote those three on a page, and after each one I left a line for "where would I see this." Capex comes from the large operators' disclosures and guidance. The Fed comes from the wording of the post-meeting statement. Data centers come from local council votes and local news. Writing it out, I noticed I had no regular source at all for the third one, which means I am blind to it.

![Three horizontal connections sit in a row: the first two run as solid lines from an exit condition to a concrete source of observation, while the third breaks off midway and ends in an empty dashed box.](/figures/three-exit-conditions-two-have-eyes-en.svg)

The value here isn't forecasting accuracy. When the bull market does end, I'll have a list to check against instead of deciding with my pulse in the moment. Dan Ives said in the episode that this is 1997 rather than 1999. I wouldn't lean on that as evidence, but it does remind me that wherever someone sets the clock, their own criteria get set to match. What I want is the list.

## Worth looking at

- The Compound and Friends, episode of September 18, 2026, hosted by Josh Brown and Michael Batnick with guests Tom Lee and Dan Ives
- Dario Amodei's September 2026 public letter and Sam Altman's response on X
- The Fed's post-meeting statement and press conference from this meeting
- Quarterly reports and guidance from Salesforce, ServiceNow, CrowdStrike, Fortinet and Palo Alto Networks
- Capex disclosures and next-year guidance in the major cloud operators' filings

## The one thing to take with you

**When bad news arrives, ask first which number it changed.**

March's software selloff and April's cybersecurity selloff showed me something: at the moment of the sharpest falls, no number had been changed. What changed was our picture of the future, and a picture has no units. It can go from 100 to 40 over a weekend and from 40 to 200 over six months.

Here is something I tried, and I'd like to pass it on — it has nothing to do with stocks. Take the thing that has cost you sleep lately: a value on a health report, a sentence from your manager in a meeting, a shift in your kid's grades. On a sheet of paper, make two columns. Left column: the numbers that have already changed — the value printed on the report, the announcement the company actually issued, the score on the sheet. Right column: what my imagination changed — the consequences I inferred, what I think comes next. Fold the paper away and set a reminder on your phone for three months out.

Three months later, take the paper out and check it. The first time I did this, the left column had one line and the right column had eleven, and three months later zero of the eleven had happened. That sheet was worth more to me than any argument, because it was in my own handwriting.

![The left column holds a single block while the right column stacks eleven blocks far higher, and an arrow on the right points to a box labelled three months later containing a large zero.](/figures/one-line-left-eleven-lines-right-en.svg)