> ## Content Index > Fetch the complete content index at: https://longtalkshort.com/llms.txt > Use this file to discover other available public pages before exploring further. # The Wrong Bubble: Josh Brown on Why AI's Real Risk Is Earnings, Not Prices - URL: https://longtalkshort.com/josh-brown-wrong-bubble-ai-earnings-not-valuations/ - Published: 2026-11-02T16:13:46.000Z - Updated: 2026-11-02T16:13:46.000Z - Description: Josh Brown tells Prof G Markets that if AI is a bubble, it's the most pathetic he's seen: prices already bake in doubt. Why he says the risk is in earnings, why some losses aren't losses, and the bear case he does take seriously, checked. - Author: Truman - Tags: Economics, Josh Brown, Scott Galloway, Prof G Markets **If you're waiting for the AI bubble to pop, you probably picture prices: stocks bid up far beyond what the businesses earn.** Josh Brown, co-founder and CEO of the advisory firm Ritholtz Wealth Management, tells Scott Galloway and Ed Elson on *Prof G Markets* that you're watching the wrong number: > If it is a bubble, it is the most pathetic bubble I have ever seen in my life The market, in his view, already prices in doubt. The risk worth watching is whether the profits last. *In short: Brown separates a valuation bubble, which he says the numbers don't show, from an earnings bubble, which he says is the only case a bear can credibly make. He also argues that big accounting losses at AI labs aren't the same as money lost, that a moat matters more than the growth rate, and that the real bear case is companies learning to ration their AI spending.* ## 1\. Not every loss is a loss The episode opens on Anthropic's S-1 filing. Brown's point is about the word "loss". **The $80.7 billion headline loss is mostly an accounting effect tied to the company's financing, he says.** The number fund managers watch is the operating loss, about $693 million for the first half of 2026, down from $2.1 billion in the first half of 2025. A real loss, he says, is unsold Bud Light after the 2023 boycott: product nobody will buy. **Spending ahead of demand is a bet.** "In year three, the goal is lose money, but make it count for something." ## 2\. The growth rate isn't the risk. The moat is Elson notes that Anthropic's growth slowed sharply in the third quarter. Brown shrugs: nobody grows 16-fold twice. What would worry him is how easily "the next model that's 10% more powerful" could take Anthropic's customers. Galloway floats a bold idea: Anthropic could be the world's most valuable company within three or four years. Brown pushes back. **A software lab's moat, as Elson puts it, is people, culture and intellectual property, and those are harder to defend.** So what is the danger? ## 3\. The bear case Brown takes seriously Brown's real worry is mundane. Big companies told staff to "go do some AI" and token use ran out of control. He cites Uber, which said it blew through its 2026 AI budget within months. What comes next, in his telling, is "rationalizing" rather than rationing: **AI budgets set by department, and software layers that send only the hardest 10% of a job to a premium model** ("a Mercedes-Benz") and the rest to cheap or open models ("an ox cart"). That's bad news for anyone selling only the most expensive frontier model. Elson adds two worries: hundreds of billions of dollars in compute commitments not yet on Anthropic's income statement, and demand that leans on startups funded by venture capital. ![Left: a large bar labelled $80.7 billion headline loss beside a small bar labelled $693 million operating loss, with a note that one is mostly accounting. Right: an AI workload box splitting into about 10% of tasks for a premium model and the rest for cheaper models, labelled rationalize, not ration. No people are shown.](https://storage.ghost.io/c/d7/93/d793216c-cf13-441f-bdae-4f3aef998c0e/content/images/2026/11/fig1-9.png) Left: Brown's distinction between an accounting loss and an operating loss in Anthropic's filing. Right: his bear case, where companies route most AI work to cheaper models. Our drawing, not to scale. Made for Long Talk Short. ## 4\. Why he calls it an anti-bubble Brown says investors are "factoring in growth skepticism every single day the market has opened". His evidence: analysts expect tech earnings to grow 65.6% in the quarter he cites, or 24.9% without chipmakers, yet **almost no semiconductor stocks are hitting 52-week highs.** He calls SanDisk, Micron and Western Digital "the three cheapest stocks in the market", says Micron trades at about five times earnings, and puts Nvidia at 15 times forward earnings, which he believes is the cheapest it has been on that measure. So, he concludes, if there is a bubble anywhere, it must be in the year's earnings. Elson agrees multiples have shrunk but still expects a correction, starting with heavily indebted data-center firms. Brown calls that bullish. ## 5\. Hard times can help the giants Brown's most counterintuitive point is that **high rates, high energy costs and tangled regulation can be good for the stock market's leaders.** The S&P 500 isn't "business", he says; it's "500 Michael Jordans". Compliance costs and trade headaches crush average firms, but the biggest ones can afford the experts, gain share and turn the burden into a moat. The biggest tech firms are "basically nation states with ticker symbols". His evidence is a 15-year chart of large companies beating small ones. Galloway notes that about 10% of Americans own most of the stock market; Brown answers that the build-out means construction jobs. ![Left: two gauges, one labelled valuation bubble reading low and one labelled earnings bubble with a question mark, with Brown's chip-stock figures beneath. Right: a tall column labelled largest firms beside a short column labelled average firm, under arrows for regulation, energy costs and high rates. No people are shown.](https://storage.ghost.io/c/d7/93/d793216c-cf13-441f-bdae-4f3aef998c0e/content/images/2026/11/fig2-9.png) Left: Brown's split between a valuation bubble and an earnings bubble. Right: his argument that economic headwinds become moats for the largest firms. Our grouping of his points. Made for Long Talk Short. ## Weighing it: what holds up **Framing:** The bubble remark is a small slice of a long episode that mostly covers Anthropic's filing and Netflix. The claim is Brown's view, not a tested forecast. **Holds up:** Nvidia did trade near 15 times analysts' estimates for its fiscal 2028 earnings in October 2026\. Micron was around five to seven times forward earnings, and SanDisk's PEG ratio (price-to-earnings divided by growth) was about 0.10 in one third-party table, near Brown's 0.08\. **Uber's chief technology officer did say in April 2026 that the company had used up its 2026 AI budget**, driven by AI coding tools. By August 2026, Uber told Axios its AI spending had stabilized despite heavier use, thanks to routing tasks to cheaper models, caching and session caps: Brown's "rationalizing" playing out at one company. **Partly holds:** On nearer-term estimates, some sources put Nvidia nearer 25 times, so "15" depends on which year you use. **Corrections:** **Low price-to-earnings ratios aren't proof of no bubble:** memory chipmakers often look cheapest when their earnings peak, and Micron reported a loss in 2023\. That's the earnings bubble Brown concedes. Elson's "80% of spending from 1% of companies" comes from Ramp's corporate-card data, a proxy for spending by Ramp's own customers, not the labs' audited revenue. Brown puts the S&P 500's entry bar at about $25 billion to $30 billion in market value; the published threshold is $22.7 billion. Galloway cites *The Economist* on 100,000 jobs destroyed by AI; its estimate was roughly 200,000 layoffs blamed on AI. **Not checked:** the 65.6% and 24.9% earnings-growth estimates, the 52-week-high readings, the "89% chance" of a year-end high and the anecdotes about Anthropic employees. **Disclosures:** We worked from YouTube's automatic captions, so quotes may contain small errors. **Ritholtz Wealth's public filing of its stock holdings for mid-2026 lists Nvidia among its ten largest positions**, plus Micron, Western Digital, Alphabet, Amazon and Microsoft. The show carries paid ads, including one for a fund that sells stakes in private tech firms. **Bottom line:** the distinction between a price bubble and a profit bubble is useful. Cheap-looking multiples on cyclical chip stocks are weaker evidence than he suggests. **Related reading:** [AI Isn't the Bubble. The Debt Behind It Might Be.](https://longtalkshort.com/ai-bubble-first-domino-debt-grand-angle/) · [All-In on Anthropic's filing and who pays for the AI build-out](https://longtalkshort.com/all-in-anthropic-s1-google-500-billion-who-pays-for-ai/) · [Grand Angle on copper and the AI build-out](https://longtalkshort.com/grand-angle-not-running-out-running-late-copper-ai-time/) ## Long Talk Short - Brown says the market is pricing in doubt, so if there's a bubble, it's in earnings, not valuations. - A big accounting loss isn't the same as product nobody will buy; spending ahead of demand is a bet. - Growth always slows. Whether customers can easily switch to a rival matters more. - The bear case he takes seriously is companies sending most AI work to cheaper models. **The one thing to remember:** before you call a bubble, ask which number is inflated, the price or the profit the price assumes. *This summary is general information, not investment advice. Do your own research or speak to a qualified adviser before making investment decisions.* Source: Ed Elson and Scott Galloway with Josh Brown, "Josh Brown: If This Is A Bubble, It's Pathetic," *Prof G Markets*, on YouTube. This is an independent summary and is not affiliated with Prof G Media, Josh Brown or Ritholtz Wealth Management.