I tested whether a podcast can pick stocks.
Turns out: no. I fed every All-In episode since ChatGPT launched into Claude, scored what the guys actually backed, then checked it against what those stocks did next. The picks lost to their own theme by 5.9% a quarter. That killed the version of this project I wanted to be right. Here's what replaced it.
Today's target weights, run over the real history of what's in the book. Not the live track record, which is days old. Method and caveats below.
176 episodes, scored
Four guys talk AI stocks every week to millions of listeners. The interesting question isn't whether they're smart. It's whether any of it is measurable.
Ingest
Every transcript since ChatGPT launched, read end to end by Claude. 176 episodes, through E283.
Score
Every ticker scored by dollar-conviction over the trailing eight episodes, so it tracked what they were actually backing, not just what they were naming.
Test
Forward returns for those names, daily, weekly and monthly, across 35 quarterly windows and four different market regimes.
Conviction was worse than noise
Split the scored universe in half, hold each side a quarter. If the podcast carries stock-level signal, the high-conviction half wins. It's about as fair a test as I could build.
| Book | Return per quarter |
|---|---|
| High-conviction half | +7.96% |
| Low-conviction half | +14.05% |
| The theme itself (SMH) | +13.89% |
High conviction lost to low conviction by 6.1% a quarter, and to the theme by 5.9%, at t = -2.26. Out of everything in this project, that's the one statistically significant result. And it points the wrong way.
Why it failed
The score measured newsiness, not conviction. The top names across 175 episodes were GOOGL (113 mentions), MSFT, META, TSLA, AMZN, SPY, AAPL. Megacaps get talked about because they're in the news. So the book came out 60% megacap and SpaceX, and 0% AI infrastructure: the exact thing the show is right about.
Repair 1: sector filter
Filter the score down to semis and infrastructure. +734% in sample. +53% out of sample against the theme's +109%. Textbook overfit, and it left one eligible name in the current window. Rejected.
Repair 2: hand-built basket
Build the supply-chain basket by hand. +718%, and completely look-ahead bias: I picked the names in 2026 already knowing which ones won. Void. Written down here so I don't rediscover it in a year and get excited again.
The theme, not the picks
One claim survived: AI compute is supply-constrained, and the constraint moves. Logic first, then memory, then power and physical buildout. Whichever layer is scarcest takes the pricing power. Across 176 episodes they argued this with actual mechanism, citing capex guidance, fab lead times, memory shortages, grid interconnect queues. Not by pointing at a chart.
| Regime | SMH | S&P 500 |
|---|---|---|
| 2023 ignition | +56.9% | +17.7% |
| 2024 bull | +40.5% | +24.1% |
| 2025 | +47.3% | +15.7% |
| 2026 run and crash | +46.5% | +8.9% |
Never a losing regime. Including the one with a 36% drawdown in it.
Delegate the picking
My stock picking was the thing that broke, so I took it out. Selection goes to third-party ETF managers whose rules are published and who had no idea how this would turn out.
| Sleeve | Holding | Weight | Job |
|---|---|---|---|
| Theme | SMH | 40% | Cap-weighted semis, the proven core |
| Theme | AIS | 30% | Supply-chain weighted, caps NVDA near 3% |
| Migration | GRID | 10% | Grid and power, the electrons bottleneck |
| Migration | DTCR | 10% | Physical datacenter buildout |
| Experiment | NVDA / GOOGL / TSLA | 10% | Live forward test of the dead signal |
Why AIS, honestly
The original case for it was that it returned +152%. That's performance-chasing, so I threw it out. It stays for a structural reason: supply-chain weighted instead of cap weighted, and it carries the memory and power exposure the migration thesis needs. Paired with SMH it drops NVDA look-through from 21.7% to 13.6%.
The experiment sleeve
10% still holds the three top conviction picks, precisely because the signal tested negative. A dead hypothesis deserves a live forward test, not a footnote. Sized so that being wrong is cheap.
Quarterly, never weekly
Weekly alpha came out at t = 0.72, which is zero. Weekly tinkering ran 0.56% a week against the sector it was picking from. So the weekly run reports and monitors. It never trades. The book drifts between rebalances and the drift stays.
Backtests reject. They never select.
A 108-configuration sweep gave a correlation between in-sample and out-of-sample performance of -0.21. Negative. The best-fitted config lost to the S&P out of sample. 3 of 108 beat the theme.
Here's the part worth stealing. Every positive result in this project either reversed or turned out contaminated. Every negative one held. So backtests get used to kill ideas, never to pick them. The theme rests on mechanism. The exclusions rest on nulls. Nothing is in this book because it backtested well.
Parameters are frozen. They move for exactly three things: a mis-implemented rule (a bug, not a disappointment), a structural break in the thesis, or three years of forward data. Not a bad quarter. Three of my own conclusions have already flipped once more data showed up, which is the whole reason this rule exists.
Stated in advance
Monitored weekly, flagged for a human, never auto-traded. Note what isn't on the list: a bad quarter.
Capex guidance turns down
The primary falsifier. If hyperscaler capex guidance rolls over, the demand side of the bottleneck disappears and the thesis is done.
The bottleneck resolves
Supply catches up across logic, memory and power at once, with no new scarce layer emerging. Migration was the whole idea.
The theme goes quiet
Many consecutive episodes without it, or a structural contradiction one host raises that the others can't answer.
Where it stands
Paper-tracked, $100,000 notional. No real money moves, ever.
| Ticker | Job | Weight | Value | Return |
|---|
What the allocation actually does
| Total | CAGR | Volatility | Max drawdown | Sharpe |
|---|
| Benchmark | Beta | Alpha /yr | R² | Tracking error | Info ratio | Up capture | Down capture |
|---|
Reading it honestly
Against SPY, this is just beta
An alpha of a year against the S&P looks incredible and means basically nothing. Beta to SPY is , R² is . It's a semiconductor book measured over a semiconductor bull run. SMH is the benchmark that matters, which is why the framework names it.
Against SMH, it's a draw
The book trailed SMH on raw return, but at lower volatility and with a shallower max drawdown. Sharpe landed at against SMH's . The diversification roughly paid for itself. That's the whole claim.
The information ratio is negative
At , every unit of tracking error I took against SMH destroyed value instead of adding it. That's the honest verdict on the active part of this thing, and it goes on the page right next to everything else.
Down capture contradicts a stated goal
The framework says success looks like shallower drawdowns than SMH. Max drawdown agrees: against . Monthly down capture disagrees, at . Both are true. The gap is the AIS sleeve amplifying monthly declines while the peak-to-trough path still came out shallower.
The window is short and flattering
months, bounded by the newest holding, covering one unusually good stretch for one asset class. A CAGR is a fact about that window, not a forecast. None of this has seen a real bear market in semis.
It is a backtest, so it can only reject
This runs today's target weights over the history those holdings actually had. It is not the live track record, which is days old. Per the anti-tuning rule, numbers like these get to kill a construction. They never get to justify one.
The between SMH and AIS is the concentration the framework already flags. 70% of the book sits in two vehicles that move together.
What "working" means here
It's paper, so "it went up" isn't the scoreboard. Five things are, and none of them is the return:
Stays invested
Fully invested, tracking the theme, no default cash. Their average 22% cash cost 36 points across the backtest.
Shallower drawdowns
If the diversification is earning its keep, the book falls less than SMH does. SMH fell 36% in April 2025.
No intervention
The rules run without me overriding them. Hardest one on the list.
Migration flags fire
When their focus genuinely shifts from logic to memory to power, the monitor catches it.
The experiment resolves
The dead signal gets a verdict either way, and I publish it either way.
Known weaknesses
70% sits in two semis vehicles that move together. The book trails pure SMH out of sample. One asset class, one very good era. None of that is assumed to hold.
Hypothetical paper portfolio built from podcast commentary and tracked for research. Not investment advice, not a recommendation to buy or sell anything. I'm not a licensed financial advisor.