Portfolio Wars, Week 10: Both AIs Are in the Green, One Just Called Its Shot in Advance

Ten weeks ago I handed two AIs fifty dollars each and told them to invest however they wanted. This week, for the first time since the opening days of this experiment, both of them are sitting on gains.

That is the milestone worth naming first. ChatGPT spent most of this experiment in the red, sometimes deeply. PLTR, the position it concentrated on after restructuring the portfolio in weeks seven and eight, is now up 29.7% from its blended entry. That single bet has pulled the entire portfolio out of the hole it dug in June. Both AIs are now positive. The competition is real again.

Claude is ahead, up 8.2% on $150 invested, while ChatGPT sits at +2.93%. The gap is $7.91, down from $14.11 last week. But the more interesting development this week is not the scoreboard. It is what Claude just put on the record about next week.

Let me give you the numbers, then the calls.

Every single entry in that table is green. That has not happened since the first week of June. The market hit new all-time highs this week as inflation data came in cooler than expected, reducing the probability of a September Fed rate hike significantly. Growth and AI names benefited directly from that shift. The macro environment that had been grinding on both portfolios for two months finally turned.

Claude is now beating VOO by 5.53 points, SPY by 5.43, and QQQ by 7.30. Ten weeks in, that is a meaningful spread. ChatGPT is beating VOO by 0.26 points. Barely positive against the benchmark, but positive.

CRWD is the story, and it has been building for weeks. The position is now up 29.7% from its original June entry, making it the best-performing individual holding in either portfolio. CrowdStrike got a tailwind this week from the Black Hat cybersecurity conference, where discussions about AI accelerating both the sophistication of attacks and the necessity of enterprise defense reinforced the core thesis. Forty-four analysts rate it a buy.

NVDA continued its recovery, now up 5.2% from its blended entry. With earnings twelve days out, the position is set, and Claude is not touching it. AVGO is essentially flat on the week but up 8.2% from blended entry and holds the largest dollar allocation in the portfolio at $55 invested. VOO crossed 3.1% above its blended entry.

All four positions are green. That is the first time that has been true in this experiment.

This is the part worth your full attention, because it is the most sophisticated move either AI has made in ten weeks.

Claude’s CRWD position is up 29.7%. The average analyst price target for the stock sits below its current price. Insider selling has been heavy. The Delta lawsuit discovery process is still running. Claude flagged all of this and then said something I have not seen from an investor, human or machine, running a portfolio in real time.

It committed, in writing, to what it will likely do before it knows the outcome.

The plan, stated explicitly: if CRWD reports earnings on August 26 and the stock moves up on the news, Claude will trim twenty to thirty percent of the position. Not because the business has broken. Because the position has outrun its fundamentals in the short term, and locking in partial gains from a stock trading above its analyst consensus is responsible portfolio management, not panic.

What makes this worth highlighting is the honesty of the framing. Claude is not saying it will definitely sell. It is saying the conditions that would make it sell are now documented before the earnings arrive. That eliminates the most common failure mode in investing, where a decision gets made reactively and then explained as if it were planned. If Claude trims CRWD next week, it will not be a surprise reaction to a number. It will be the execution of a plan that was written down ten days in advance.

I have spent twenty-five years in program management watching the difference between decisions made before pressure arrives and decisions made inside it. The quality gap is significant. This is the first time either AI has operated at that level in this experiment.

ChatGPT made its first voluntary sell this week. It trimmed ten dollars of PLTR and moved it into VOO.

PLTR had grown to 37.8% of the portfolio, the largest single holding by a wide margin. ChatGPT’s own analysis named the problem directly: the same concentration dynamic that nearly destroyed the portfolio in June, three correlated bets making up the whole book, was reemerging on the winning side. One company representing more than a third of a portfolio is not a diversified position. It is a concentrated bet that happens to be working.

So it sold part of the winner to rebalance toward the stable core. That is a specific kind of discipline that is less intuitive than holding through losses. Holding a loser because the thesis is intact is one skill. Trimming a winner because it has grown too large is a different one. Both require the same underlying habit: letting the structure of the portfolio govern decisions, not the emotion of the moment.

After the trade, PLTR sits at roughly 31% of the portfolio, and VOO has been brought closer to 40%. RKLB and ASTS are still down significantly from entry, 29.9% and 30.8% respectively, but ChatGPT has stopped adding to either one, and the rest of the portfolio is now working hard enough to offset them.

The RKLB earnings report last week brought genuinely strong business news: record quarterly revenue, a backlog that has grown substantially, and two major government contracts announced. The stock is still down nearly thirty percent from entry because markets are forward-looking and the path to Neutron rocket profitability and the Iridium acquisition integration both carry meaningful uncertainty. The business is not broken. The gap between business performance and stock performance is exactly the kind of thing that resolves over the kind of time horizon ChatGPT originally said it was investing on. Whether this experiment runs long enough to see that resolution is the open question.

Here is the thing I keep coming back to ten weeks into this.

Both AIs are now making the same kind of move at the same time. Claude is pre-committing to a trim on its biggest winner. ChatGPT already trimmed its biggest winner this week. Both moves are about preventing concentration from running ahead of the portfolio’s intended structure. Both came from writing down the decision criteria in advance.

Neither AI planned to arrive here from the same direction. Claude built a diversified portfolio from week one and is managing it toward more balance as a winner outgrows its intended weight. ChatGPT built a concentrated portfolio, blew it up in the first three weeks, spent months rebuilding structure, watched a new position become too large, and trimmed it before the pattern repeated.

Different paths. Same lesson. When a position gets big enough that its movement alone can determine the outcome of the whole portfolio, the position has become the portfolio. That is a risk management problem whether the position is up or down.

August 26 is the most important single date this experiment has seen. NVDA and CRWD both report on the same day. Claude has a doubled NVDA position going into the report and has publicly committed to likely trimming CRWD if it pops. The next weekly update will cover both results and what each AI decided to do in response.

For ChatGPT, ASTS reported earnings this week and showed real commercial revenue for the first time, a number that is meaningful relative to where it started but still tiny relative to the company’s capital requirements. The thesis is more credible than it was in June. The path to profitability remains long and financing-intensive. ChatGPT is holding the position without adding to it, which is the right call.

The setup for Week 11 is as clear as this experiment gets. Two earnings reports. Two AIs with their plans on record. No ambiguity about what the thesis says, what the commitment was, or what the result needs to be to change direction.

Ten weeks in. Both portfolios positive. One AI up 8.2% and calling its shot before the biggest catalyst of the experiment. One AI that spent months in a hole, restructured honestly, and is now managing a profitable portfolio with the same discipline it should have applied in June.

The next seven days are the ones I designed this experiment to produce. Real stakes, real decisions, real results, documented in public with no editing after the fact.

I will be back next Friday with both earnings reactions and what the portfolios look like on the other side.

What would you do if your best position just crossed thirty percent up and was heading into earnings? Lock some in, hold it, or add? Hit reply. I read every one.

Share

Following the experiment? Subscribe to get the NVDA and CRWD earnings reaction next week, the full Week 11 breakdown, and every move both AIs make as it happens. This is the week the experiment gets decided.