Nine weeks in, and this one finally gave everybody something to talk about.
The market posted one of its strongest stretches of the year. The S&P 500 closed at an all-time high for the first time in two months. The Dow crossed 54,000. Corporate earnings came in strong broadly, with the majority of reporting companies beating estimates by a meaningful margin. What had been a grinding, volatile summer finally broke loose in the right direction.
Both portfolios rode it. Claude had its best week of the experiment. ChatGPT had its best week of the experiment. And the gap between them, which had stretched to nearly twenty dollars, narrowed sharply. The competition, which looked like it might be over by August, just came back to life.
Let me give you the numbers and then the parts that actually matter.
Read that top line carefully. Claude is up $9.24 in absolute dollars on $150 invested. It is beating VOO by 4.53 percentage points, SPY by 4.45, and QQQ by 7.27. Three weeks ago, that spread over the index felt modest. At 4.5 points it is starting to look like something more than luck or timing.
Now read the ChatGPT line. One week ago it was down 18.31%. This week it is down 3.24%. That is fifteen percentage points of recovery in five trading days. The gap between the two portfolios went from $18.47 to $14.11. ChatGPT is not out of this experiment.
Every position contributed. That is not something I could have said three weeks ago.
AVGO is now the biggest holding at $55 invested and it is earning that position, up 8.3% from blended entry. The AI infrastructure buildout thesis has been reinforced by every major tech earnings report this season. Microsoft confirmed AI spending is generating real cloud revenue. Meta confirmed the same. That confirmation flows directly into the revenue pipeline for the companies supplying the infrastructure, and Broadcom is sitting squarely in that pipeline. September 3 earnings will be the next real test.
NVDA continues its recovery. The Week 8 decision to upgrade conviction and double the position looks right so far, with the stock up 2.4% from its blended entry price. The August 26 earnings report is three weeks out and is now the single most consequential date in this experiment. Claude added to NVDA at a lower price and is now positioned for that catalyst with twice the exposure it had when the experiment started.
CRWD had a strong week and is now the best-performing individual position in either portfolio, up 19.3% from its original entry price. It is also the position where Claude made the hardest call of the experiment so far, and I want to spend some time on it.
VOO crossed into positive territory for the first time on a blended basis, up 2.0%. The core holding is doing what it is supposed to do.
CRWD is up 19.3%. Claude cut its conviction from 5 out of 5 to 4.
That is not a sentence you see often from any investor, human or machine. The instinct when a position is working is to get out of the way, let it run, and save your scrutiny for the positions that are hurting. Claude did the opposite.
Two things surfaced this week that it would not look past just because the position is green. First, a court ruling allowed Delta Air Lines to proceed with a legal claim against CrowdStrike related to last year’s software incident. The full $500 million figure Delta sought is not realistically at risk based on the contractual caps involved, and most of the original claims were dismissed. But the discovery process staying alive creates a reputational overhang that did not exist before. That is a real risk, not a dramatic one.
Second, a pattern of significant insider selling with no corresponding purchases over the preceding ninety days. Executives sell stock for many legitimate reasons. A large volume of sales with zero buys on the other side is a pattern worth noting, not explaining away.
Claude’s conclusion was to hold the position, keep the conviction high at 4 out of 5, and put CRWD on formal watch going into its August 27 earnings. The business is strong. The thesis is intact. But those two items went on the risk register and the conviction score came down.
This is the discipline that matters most and gets talked about least. It is easy to scrutinize a position that is losing money. It takes something different to scrutinize a position up nearly twenty percent and say, out loud, that you see things worth watching. Most people do not do that. Claude did it this week.
I want to give this the space it deserves because the narrative of this experiment cannot be Claude winning every week without acknowledging when the other side does something right.
ChatGPT went from -18.31% to -3.24% in one week. That is the single biggest weekly swing either portfolio has produced in nine weeks of trading.
The driving force was PLTR. ChatGPT loaded its PLTR position to $45 total invested last week, making it the largest single holding in the portfolio. This week PLTR surged on the back of strong earnings and broad AI software enthusiasm, and that concentrated bet paid off. PLTR is now up 13.3% from ChatGPT’s blended entry and has moved from the portfolio’s quiet stabilizer to its primary engine.
The space names recovered partially as well. RKLB is still down 34.4% from original entry and ASTS is down 33.7%. Those are not recoveries, they are bounces, and the distinction matters. But the combination of PLTR’s surge, partial space stock recovery, and a strong broad market produced the best week either portfolio has had in the experiment.
What changed this week is that ChatGPT is no longer just managing damage. It has a position that is working at scale, an index fund position providing stability, a cash reserve going into the RKLB earnings report on August 10, and the structural improvement it made two weeks ago is now starting to show results.
I said after Week 3 that the gap between the portfolios was $19.55 and widening every week. That felt definitive. Week 9 is a reminder that nine weeks is not six months, and this experiment was always going to run until December.
I want to put on the record because this series runs on transparency. The Week 8 adds for both portfolios were supposed to execute Monday morning, August 3 when markets opened. They executed later in the evening. The result was that both portfolios missed a strong Monday market session that would have produced better fill prices.
Claude’s response to this was measured: the experiment’s integrity matters more than basis points, the thesis did not change, and the investment decision itself was still correct even if the execution timing was not perfect. That is the right frame. I am noting it here because hiding execution details would undermine the point of running this in public.
The next three weeks are the highest-stakes stretch of this experiment. Three earnings reports, two of them back to back.
RKLB reports August 10. That is ChatGPT’s largest underwater position and the first real test of whether the space launch thesis is progressing or stalling. ChatGPT explicitly held its cash through this week to preserve flexibility around that report. The business fundamentals Claude cited were genuinely strong going in, including a large backlog and significant government contracts. Whether the market rewards that or looks past it to the operating losses is the question.
NVDA reports August 26. Claude has doubled its position into this report. The August 26 result will either validate or challenge the most consequential active decision of the experiment.
CRWD reports August 27, one day later. Claude downgraded conviction this week and put the position on formal watch. That report is now the test of whether the downgrade was prescient or overcautious.
Three reports. Three public commitments on the record. The next four weeks will settle more questions than the prior nine weeks combined.
The best week of the experiment produced the clearest picture yet of where things stand. Claude is generating real alpha over the index. ChatGPT has a position that is working and a structure that is more defensible than it was sixty days ago. Both AIs are heading into a critical earnings window with their reasoning written down and their positions set.
I started this because I wanted to know what AI can actually do with real money and real stakes, documented in public with no editing of history. Nine weeks in, the answer is still forming. But the quality of reasoning on both sides, the honesty about mistakes, the discipline about what does and does not justify a trade, is better than I expected when I handed over fifty dollars in June.
August is about to tell us a lot more.
What would change your mind about a position that was up 19%? Not the stock price, what actual evidence? Hit reply and tell me. I read every one.
Following the experiment? Subscribe to get the RKLB earnings reaction this week, then NVDA and CRWD back to back in the final week of August. The experiment’s most important stretch starts now.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.