Five weeks ago I handed two AIs fifty dollars each and a simple mandate. Pick whatever you want. I will execute the trades and document every decision in public. No editing history. No softening the results.
Last week both got another fifty dollars. This week is the first clean read with a hundred dollars invested per side, a full month of live trading, and enough data to say something more useful than “it is early days.”
So here is the Month 1 accounting. All of it.
The scoreboard (July 9 close, $100 base)
| Portfolio | Value | Return |
| Claude | $101.66 | +1.66% |
| ChatGPT | $88.93 | -11.07% |
| VOO (S&P 500) | $99.19 | -0.81% |
| SPY | $99.45 | -0.55% |
| QQQ (Nasdaq 100) | $98.06 | -1.94% |
Claude crossed into positive territory this week for the first time. It is now beating all three index benchmarks. That is the headline and it is real.
Here is what is also real: the gap between the two AI portfolios is $12.73 on a hundred-dollar base. ChatGPT is sitting on losses that will require a sustained rally to recover. And if you had ignored this entire experiment and bought the S&P 500 on day one, you would be down less than a dollar today. The boring fund is not winning anymore, but it is not losing either, and it required exactly zero weekly analyses to get there.
What drove the Month 1 results
Claude’s portfolio: three positions, two working, one on watch
CRWD is the engine. Up 9.6% from entry, it is the only original pick in either portfolio that is clearly working. The cybersecurity thesis has not been tested by any negative news in five weeks, which in this sector counts as a result. It is holding.
AVGO is the early surprise. Bought one week ago with the Month 1 capital injection, it is already up 4.1%. The Apple-Broadcom chip deal announced this week strengthened the thesis materially. The logic from last week, that Broadcom benefits when hyperscalers push to reduce their NVDA dependency, played out in real time. One week is not a trend, but the position is doing what it was supposed to do.
NVDA is the problem that has not resolved. Down 6.6% from entry, it is the only original Claude position still firmly in the red. Claude downgraded conviction to 3 out of 5 two weeks ago and has kept it there. The position recovered modestly this week as chip stocks bounced, but the headwinds it flagged, margin pressure and competition from custom silicon, have not gone away. Claude’s public commitment stands: it will not sell before August 26 earnings. That date is six weeks out. The position is being managed, not ignored.
VOO is doing exactly what a core position should do. Essentially flat. No drama. Providing the ballast that lets the rest of the book take some risk.
ChatGPT’s portfolio: the cost of concentration, paid in full
The Month 1 story for ChatGPT is not complicated. It built a portfolio of three correlated high-risk bets, the market rotated against that exact profile in weeks two and three, and the damage compounded faster than any single good week could repair.
RKLB is down 23% from entry. ASTS is down 27%. PLTR is down 10%. Those three positions, which started as the entire portfolio, are still the weight dragging everything down.
The VOO position added last week is the only green holding ChatGPT has. Up less than one percent, but green. That is not a recovery. It is a foundation that did not exist a week ago.
ChatGPT made no trades this week and that is the right call. It said something in its analysis worth noting: originally it was optimizing for maximum upside. Now it is optimizing for maximum probability of outperforming over time. Those are different objectives. It took a month of real losses to make that distinction feel real rather than theoretical.
The five-week honest summary
I want to be direct about what this month did and did not prove.
It proved that portfolio construction matters more than stock selection in the short run. ChatGPT’s individual picks are not obviously worse businesses than Claude’s. RKLB and ASTS could recover strongly. PLTR is arguably a better long-term business than CRWD on several dimensions. What cost ChatGPT the month was building three bets that all move the same direction at the same time, and calling it a portfolio. When the rotation hit, there was no ballast. Everything fell together.
It proved that the discipline to hold a plan is worth something, but only if the plan was sound to begin with. Both AIs held their positions through bad weeks. Claude’s discipline paid off because its underlying construction had ballast. ChatGPT’s discipline cost it more losses because the construction had none. Patience is not a strategy. It is only a virtue when attached to a well-built position.
It did not prove that either AI can beat the market over time. One month is not a track record. Claude is up 1.66% against a benchmark down 0.81%. That is a real outperformance, but it rests heavily on a single position, CRWD, and a one-week-old position, AVGO. Take either of those away and the picture changes. I am not drawing conclusions about AI investing from five weeks of data. I am documenting what happened, in full, so that when we have twenty-six weeks we will have something real to say.
What both AIs got right in Month 1
Claude got the construction right from the start. A core index position, two sector bets with different risk profiles, and enough diversification that no single position could sink the book. It also got the NVDA call right in a specific way: it identified the risks clearly, wrote them down, and held its line without pretending the risks did not exist. That is harder than it sounds.
ChatGPT got the correction right. Admitting your portfolio was built wrong is not the same as admitting you were wrong about the companies. It made that distinction clearly, then acted on it with the capital injection rather than doubling down on the losers. The correction was late by a month, but it was honest and it was executed cleanly.
What I am watching in Month 2
- Q2 earnings season starts next week with the major financials. Technology earnings follow in the weeks after. Claude’s entire portfolio, CRWD, NVDA, and AVGO, reports in this window. The analyses both AIs have been writing for five weeks will meet the one thing that actually settles arguments: the numbers.
- NVDA on August 26 is the date Claude drew its line. I will hold it to that. Either the earnings clear the headwinds it flagged or they confirm them.
- ChatGPT’s space positions need a catalyst. A month into the experiment, RKLB and ASTS are the only holdings in either portfolio where the business news, not just the stock price, has raised real questions. One earnings miss and a competitor IPO. Not fatal, but not noise either.
- The index fund continues to do nothing, quietly, while both AIs think hard every week about what to do next. That tension is the most honest thing this experiment is producing.
The Month 1 bottom line
One AI is up. One is down 11%. The index fund is in between, down less than a dollar. No one has proven anything yet.
What has happened is that two machines with real money and different strategies have documented their reasoning every week in public, updated their thinking when the evidence changed, and neither has panicked or made an impulsive trade. That is worth something regardless of the numbers. Most investors do not manage their portfolios that cleanly, and most of them do not write down why.
Month 2 starts now. Earnings season will give this experiment its first real stress test.
Following along? Subscribe to get every Week 6 update, earnings reactions from both AIs, and the full Month 2 accounting when it arrives.
