By Lynn Räbsamen, CFA | Chief Markets Officer, WealthAi | Advisory Board Member, CFA Institute | Author, Artificial Stupelligence
A follow-up to The Q2 Results Are In. Original article: When ChatGPT and Claude Pick AI Stocks, Who Wins?
In January I gave two AI models the same $100,000 and the same brief. Pick the AI stocks that will beat the market over the next 1-3 years. Then buy and hold. No rebalancing, no rescue, no second chances.
For two quarters, Claude made it look easy. It won the first quarter outright, held its lead through the second, and did it all with less volatility than a ninety-one-stock index. In April I warned that a rotation back toward software, or a stumble in one big holding, could compress that lead quickly.
In July, both happened at once.
The Quarter the Streak Broke
Here is the third-quarter scorecard.
| Claude | ChatGPT | ETF Benchmark | |
|---|---|---|---|
| Q3 2026 (Jul–Sep) | -2.71% | +7.41% | -0.69% |
For the first time since this experiment began, ChatGPT won a quarter. And for the first time, Claude lost one. Claude’s portfolio actually shrank over the three months, from $154,530 down to $150,343, while ChatGPT climbed from $121,372 to $130,368.
After nine months of trailing, ChatGPT finally had the quarter it was waiting for.
The portfolio that spent half a year looking vindicated spent July looking mortal.
But the Year Still Belongs to Claude
Before anyone rewrites the story, look at the full nine months.
| Claude | ChatGPT | ETF Benchmark | |
|---|---|---|---|
| End Balance (Sep) | $150,343 | $130,368 | $128,117 |
| 9-Month Return | +50.3% | +30.4% | +28.1% |
| Standard Deviation | 37.0% | 30.2% | 38.2% |
| Max Drawdown | -12.4% | -9.8% | -12.5% |
| Sharpe Ratio | 1.56 | 1.19 | 0.94 |
Claude is still up more than fifty percent on the year and still ahead of both its rival and the index by a comfortable margin. One bad quarter does not undo two good ones.
But the gap is closing. At the half, Claude led ChatGPT by roughly $33,000. Now it is closer to twenty. The blowout has become a contest.
What Actually Happened in July
The entire quarter turns on a single month.
All year, the market had rewarded the physical layer of artificial intelligence, the chips, memory, power and cooling, and punished the software and cloud names that were supposed to profit from AI without building it. That was the trade. Claude was on the right side of it, and ChatGPT was not.
In July, the trade reversed, violently.
The infrastructure names that had carried Claude all year became the quarter’s worst casualties. Marvell fell 37 percent in a single month. Micron fell 29 percent. Vertiv fell 28 percent. ASML, AMD and Taiwan Semiconductor each dropped in the double digits. Investors had finally begun asking whether the hundreds of billions pouring into data centres would ever earn an adequate return, and they asked all at once.
At the same time, the names ChatGPT leaned on came roaring back. Microsoft rose 25 percent. Alibaba rose 27 percent. Amazon rose 14 percent. The money rotating out of infrastructure had to land somewhere, and it landed on the mega-cap software and cloud platforms that had lagged for six months.
The three stocks that made Claude win the first half were the three worst performers of July. ChatGPT never owned a single one of them.
That absence had cost ChatGPT for six straight months. In the seventh, it was the reason it won. Claude held Microsoft and Amazon too, which softened the blow, but with Vertiv and Taiwan Semi each sitting near thirteen percent of the portfolio and both falling hard, concentration finally cut the other way. Claude dropped more than 12 percent in July alone, then spent August and September clawing part of it back.
Concentration giveth, and concentration taketh away.
The Fund That Found Out the Hard Way
If you want to see how much worse this could have gone, look at what happened to the most famous believer in exactly Claude’s thesis.
Situational Awareness, the hedge fund founded by former OpenAI researcher Leopold Aschenbrenner, had built its entire book on a single idea: AI needs physical infrastructure, and the companies supplying it are undervalued. It is, more or less, Claude’s thesis, written in capital letters and underlined twice. Through June, the fund had reportedly returned an extraordinary 439 percent, and had swollen to around 45 billion dollars in assets.
Then July arrived, and the same stocks that bruised Claude destroyed Aschenbrenner.
The difference was leverage. The fund was reportedly running at roughly four-to-one. When its largest infrastructure positions, names like Micron, SanDisk, Nebius and CoreWeave, each fell more than 35 percent in the month, four-times leverage turned a painful drawdown into a fatal one. Margin calls followed. By the end of July the fund had sold its entire public portfolio to Citadel in a forced liquidation, its assets collapsing from 45 billion to roughly 10 billion in about 20 trading days. One of the largest single-month losses in hedge-fund history, reportedly around 67 percent.
Same thesis. Same month. Same stocks. One portfolio took a flesh wound. The other did not survive the quarter.
Here is the part worth sitting with. Aschenbrenner may well have been right. The fund was still up on the year even after the wipeout, and the long-term case for AI infrastructure did not change in July. What killed the fund was not the idea. It was the leverage wrapped around it. A correct thesis, expressed with borrowed money and no margin for error, became a margin call.
Claude’s portfolio fell in July for precisely the same reason Aschenbrenner’s did. It simply was not leveraged, so a bad month stayed a bad month instead of becoming the end of the story.
What This Tells Us at the Three-Quarter Mark
Two lessons, and they pull gently against each other.
The first is that Claude’s early dominance was never free. It came from concentration, and concentration is a loan the market eventually calls. For six months it paid Claude handsomely. In July it demanded some of that back. The risk was always there, politely invisible, right up until it wasn’t.
The second is that structure is destiny. Claude and Situational Awareness bet on the same thing and met the same storm. One is still up fifty percent on the year and merely annoyed. The other is a cautionary tale told at conferences. The only material difference between a flesh wound and a fatality, this quarter, was leverage.
And underneath both sits the quiet fact that has defined this experiment from the start: nobody has touched either portfolio since January. No one trimmed Vertiv before July. No one rotated into Microsoft in time. What we are watching is still the long tail of a single decision, made once, by two machines, playing out through whatever the market throws at it.
The Closing Line
For two quarters the story was how well a concentrated bet could do. This quarter it was the price of that same concentration. Both are true. Both were always going to be true. The market simply chose Q3 to send the invoice.
Claude still leads. But the lead is no longer a victory lap. It is a race again, and that makes the next three months the most interesting yet.
Q4 and the full-year verdict come in January. Until then, one revision to the house motto. Buy the pipes, not the hype. But whatever you buy, don’t buy it on margin.
This article was partially drafted with AI, and reviewed and edited by a human.
For more insights about what AI can or cannot do, check out my book “Artificial Stupelligence: The Hilarious Truth About AI”.
Subscribe here to be the first to receive my insights.
