# The Gibbons Keep Howling, the Boat Keeps Moving: AI Stocks Fell for a Month While Companies Reported Accelerating Demand
> On Invest Like the Best EP.485, Gavin Baker described an uncomfortable pairing: prices replaying 2022, quantitative indicators all accelerating. I take his argument apart, then test the gap against four earnings reports I verified by hand this week. Educational content, not investment advice.
Published: 2026-08-12
Locale: en
Tags: podcast, invest-like-the-best, AI infrastructure, research methods, education
> *Leaving Baidi at dawn among rose-tinted clouds,*
> *a thousand li to Jiangling, home in a single day.*
> *On both banks the gibbons howl without pause —*
> *the light boat has already passed ten thousand mountains.*
> —— Li Bai, "Departing Baidi at Dawn" (Tang dynasty, 759; translation mine)
## What this piece is about
Between July and early August, a good number of AI-related stocks fell forty to sixty percent from their highs. On Invest Like the Best EP.485 (August 4, 2026), Gavin Baker of Atreides Management opened with a line that stuck with me:
> "I would describe July as 2022 in a month."
Then came the part that actually matters. He asked around Silicon Valley and could not find a single quantitative indicator that had turned down. GPU availability, GPU rental prices, memory spot prices, token growth — all still accelerating. The stock prices were performing a bear market. The fundamentals were performing something else entirely.
I want to do two things here: take his argument apart to see what he actually uses to judge, then test the gap he describes against four earnings reports I checked by hand this past week. The short version: the gap is real, but its shape doesn't quite match the word "panic."
## One: he can't find the bullet
Baker raised a detail that technicians would care about. Every major selloff in recent memory had a named fear — recession in 2022, then DeepSeek, then tariffs. This time, apart from the credit market, the stated reasons all sound a bit absurd, and yet prices kept falling. He quoted the old line: the bullet that hits you is the one you never see.
So he didn't close the case. He kept returning to humility, including a line I wrote down:
> "The three most important words in investing aren't margin of safety, but I don't know."
Making a strong call while leaving the falsification conditions on the table is, to me, a marker of whether a speaker can be trusted. His conditions are concrete: hyperscaler operating cash flow stops accelerating; GPU prices contract in a sustained way; GPUs become easy to get; the labs' combined revenue stalls. Any one of those, and his whole argument goes back for review.
## Two: "a token is a token" — the accounting of the open-source scare
The segment most worth copying down was how he handled the fear that open-source models will destroy AI profitability.
> "A token is a token. And you need the exact same amount of compute to make a token, all else equal."
His dissection runs at the accounting level. Open-source models eat the margins of the frontier-model layer, swapping ninety-percent-margin tokens for thirty-percent-margin ones. But the compute layer (chips, memory, power) sees no drop in demand; with price elasticity it may see more. The panic hit the entire supply chain. Worked through the ledger, only one layer takes the damage, and another may benefit.
Breaking a scare down to accounting line items is worth more than any bullish or bearish conclusion, because the same bad news can mean opposite things to different layers. It's the same instinct as bottleneck-hunting in supply chains: don't ask whether the industry is fine, ask whether this layer is fine.
## Three: his boldest claim — the hyperscalers are under-earning
Baker spent time on something few people discuss: the gap between contract prices and spot prices for compute.
Most installed capacity was locked in years ago at low contract prices. Spot prices, meanwhile, have been climbing. His example: a startup that rented a B200 cluster about seven months ago at roughly $2.50 per GPU-hour would pay close to $4 for the same thing now, a fifty-to-sixty percent rise by his account. As contracts roll over and reprice, hyperscaler operating cash flow gets revised upward. He thinks Microsoft, Amazon, and Google are collectively under-earning right now: today's reported profitability is the version suppressed by old contracts.
He also offered a cash-flow figure: Microsoft, Meta, and Amazon's combined operating cash flow growth accelerated from 28% to 32% this quarter (28% to 35% adjusted for one-offs). That's his citation, not my recalculation — I'm presenting it as his evidence, not mine.
Credit is the one worry he called real: real rates up, spreads wider, some big names' bonds and CDS reflecting doubt. He didn't wave it away. He did arithmetic instead: if repriced cash flows can fund the buildout, the debt the market fears may never be needed. He was clear this is an estimate, not a fact.
## Four: my own comparison table
After absorbing his framework, I did one thing: laid out the four earnings reports I had personally verified this week and looked for the price-fundamentals gap. All figures below come from public filings and closing prices, dates noted.
**The gap group — the better the report, the colder the market:**
- **Leidos (LDOS)**: The August 4 report showed the highest quarterly revenue in company history ($4.56B, up 7%, organic up 4%), full-year outlook raised (organic growth ~7%, adjusted EBITDA growth ~19%), funded backlog up 44% year over year. The stock, as of the August 11 close, sat roughly thirty percent below its 52-week high.
- **Cadence (CDNS)**: Q2 revenue up 24.2%, full-year guidance raised by $125M in a single quarter, record backlog of $8.1B. I ran one calculation myself from public filings and closing prices: over roughly five months, trailing-twelve-month EPS rose about 24% while the stock rose about 18%. Earnings accelerating, multiple contracting (as of the August 11 close).
- **Adobe (ADBE)**: The mid-June report showed record revenue ($6.62B, up 12.7%, with the 10-Q stating the recent acquisition was immaterial — nearly all organic), and AI-related annual recurring revenue tripled in a year, passing $500M. The stock is down about thirty percent over a year. The press at the time called it "perfect numbers, cold market reaction."
These three are what Baker's gap looks like up close: numbers up, pricing down.
**The control group — same market, different temper:**
- **Sea (SE)**: Reported before the open on August 11 — revenue up 48%, well above consensus, but EPS missed for the fifth straight quarter. By the panic script this should fall. Instead the stock approached a six-month high, because management reaffirmed full-year profit targets and growth guidance.
Put all four on the same table and I get a reading that doesn't quite say "panic": **the market isn't fleeing indiscriminately — it has moved its anchor from current numbers to future certainty.** It pays a premium for SE because the next step is spelled out. It withholds credit from LDOS, CDNS, and ADBE because their good numbers answer "now," and the market currently only wants to hear "will you still be here in three years."
This doesn't contradict Baker; it's the other face of the same thing. He says the fundamentals haven't broken. My addition: the market hasn't lost its mind either — it changed rulers. The gap isn't caused by bad information. The unit of account changed. Whether the new ruler measures correctly is the judgment each of us has to make alone.
## Five: the falsification conditions, left on the table
House rule here: every judgment ships with the conditions that would prove it wrong.
Baker's version is listed above: cash flow decelerates, GPUs get cheaper, GPUs get easy to buy, lab revenue stalls — any one of the four.
My version: if over the next two quarters the "good numbers, bad price" group starts cutting — outlooks softening, order growth slowing — then the market's new ruler was right and I underestimated what it smelled. If instead they keep posting accelerating numbers while the pricing refuses to follow, the gap will grow too wide even for people holding the new ruler to ignore. Either way there's a date to check answers: next earnings season.
Whether the gibbons stop howling, I don't know. What I'm more sure of: write your judgments in a falsifiable shape, and the boat will know whether it's actually moving.
---
## Sources
- Invest Like the Best, EP.485, "Gavin Baker - AI Market Jitters" (2026-08-04). This piece paraphrases the argument's structure with short quotations; for the full content, listen to the original episode.
- Leidos Q2 2026 results (2026-08-04); Cadence Q2 2026 results; Adobe FY2026 Q2 results and 10-Q (2026-06-11); Sea Limited Q2 2026 results (2026-08-11).
- Prices and multiples are public market data as of the 2026-08-11 close; calculations marked "I ran myself" are the author's arithmetic on public data.
## Disclaimer
This article is educational content about research methods, not investment advice. Companies are mentioned solely to illustrate a research framework, not as recommendations. Third-party views quoted here (including Gavin Baker's interview remarks) are their own opinions. Investing carries risk; make your own decisions and own the outcomes.