Portfolio Wars, Week 8: The Worst Market Day in Months, a Microsoft Surge, and $50 More to Invest
The Dow had its worst session since April on Wednesday. By Thursday it was a different story. Claude came out essentially flat. Then both AIs had to decide where to put new money.
The easy weeks are not the ones that teach you anything.
Week 8 packed more market movement into five days than any stretch of this experiment. The Dow fell more than a thousand points Wednesday as the Federal Reserve held rates steady and bond yields surged. The Nasdaq entered correction territory, down more than ten percent from its June peak. The VIX, the market’s fear gauge, jumped sharply. By any measure it was the kind of week that tests whether a portfolio is built correctly or just got lucky in a calm market.
Then Thursday arrived. Microsoft reported earnings that confirmed AI spending is translating into real cloud revenue growth, and the Nasdaq had one of its best single sessions of the year. Semiconductor stocks surged. The whole tape reversed.
When the dust settled on Friday, Claude’s portfolio had gone from $102.47 to $100.16. Down just over two dollars across the biggest market swing of the experiment. QQQ, the Nasdaq index fund, finished the week down nearly seven percent. Claude lost two dollars.
That is the construction story. And then both AIs were told they had fifty dollars of new money to deploy into it when the markets open on Monday.
The scoreboard (July 30 close)
Claude’s headline return looks modest after three weeks of stronger numbers. The context matters. Every benchmark fell this week, two of them meaningfully. Claude absorbed one of the worst market days since last spring, a Thursday reversal, and a volatile Friday close, and came out with its original hundred dollars essentially intact.
QQQ is now down nearly seven percent from where this experiment started. Claude is up a fraction. That spread of more than seven points came not from picking the right stocks in a rising market, but from staying above water in a falling one. Those are different skills.
ChatGPT is now down 18.31%, its deepest loss in the experiment. RKLB fell further. ASTS fell further. PLTR and VOO held, but they are not large enough positions to offset the damage above them.
Why the week went the way it did
Wednesday’s selloff had a specific cause. The Federal Reserve held interest rates steady, which would normally be neutral news. What rattled the market was the internal dissent: three Fed members voted for a rate hike rather than a hold. That signal, that some inside the Fed think inflation is not under control, sent bond yields higher and hit growth stocks hard. Higher rates compress the valuations on exactly the kind of high-multiple technology names both portfolios hold.
Then Thursday flipped the script. Microsoft reported that its AI business had grown dramatically year over year, that cloud revenue was accelerating, and that enterprise customers were paying for AI tools at a rate that exceeded most expectations. That story answered the question that has been hanging over the tech sector for most of this experiment: is AI spending actually generating returns for the companies deploying it, or is it a bet that has not yet paid off? Microsoft said it is paying off. The market responded accordingly.
For Claude’s portfolio specifically, the Microsoft print matters beyond just Thursday’s bounce. The AI infrastructure buildout that drives revenue for chip suppliers and cybersecurity firms got a concrete data point this week confirming the demand is real. That does not change the near-term volatility. It does reinforce the thesis.
What Claude is doing with the $50
Claude upgraded its NVDA conviction this week from 3 out of 5 to 4. Its reasoning: Microsoft’s AI revenue confirmation was the demand signal it had been waiting for, and NVDA, still below its original entry price after eight weeks, now looks like the cheapest of the major AI names on a forward earnings basis. The correction handed it a better entry point than it had in June.
It will split the fifty dollars evenly. Twenty-five into NVDA, doubling that position. Twenty-five into AVGO, nearly doubling that one too.
Here is the updated portfolio after the injection:
The logic is worth spelling out because it is a specific kind of decision that is easy to get wrong. Claude is not adding to NVDA and AVGO because the stocks went down and averaging down feels like discipline. It is adding because the thesis got stronger this week, not weaker, and the prices are lower than they were. Those are different reasons that happen to produce the same action. Getting that distinction right matters.
What ChatGPT is doing with the $50
ChatGPT made a different call entirely, and I think it is the right one given where its portfolio stands.
It’s putting twenty-five dollars into VOO, twenty into PLTR, and keeping five in cash. It explicitly will not add to RKLB or ASTS, the two positions down more than forty percent.
Its reasoning was stated simply: finding exceptional companies is only half the challenge. The other half is building a portfolio that can survive long enough for those companies to prove you right. It has watched RKLB and ASTS absorb losses for eight weeks without a catalyst that would change the math. Adding capital to them now would not be conviction. It would be hope disguised as a strategy.
So the AI that built the most aggressive portfolio in week one is now reinforcing its most stable positions and leaving its speculative bets alone. That is not the same as admitting those bets were wrong. It is admitting that good ideas need a structure around them that can hold while they develop.
Updated ChatGPT portfolio after injection:
PLTR and VOO will now represent two thirds of the portfolio. RKLB and ASTS together will represent less than a quarter of it. That is a more defensible structure than what existed eight weeks ago.
The capital allocation decision is the real AI test
I want to step back for a moment because this week is the most instructive the experiment has produced, and not just because of the market volatility.
Both AIs just made consequential allocation decisions with real money under real uncertainty. They had the same information, the same market conditions, and the same fifty dollars. They reached opposite conclusions about where to put it, and both conclusions are defensible on their own terms.
Claude concentrated more heavily into its highest-conviction names at depressed prices going into a known earnings catalyst. That is a specific bet: that the thesis is right, the timing is reasonable, and the correction created an opportunity. It could work very well or it could add to losses if August earnings disappoint.
ChatGPT reinforced its stable core and left its speculative positions alone. That is also a specific bet: that the portfolio needs to survive first and grow second, and that the speculative names need operational progress before they deserve more capital.
This is what good capital allocation actually looks like, not in a textbook, but in a live portfolio with real stakes and a deadline. When I help organizations think through AI implementations, the hardest conversation is almost never about the technology. It is about where to put the next dollar when results are mixed and the future is uncertain. Both of these AIs just demonstrated that conversation clearly, in public, with their reasoning written down.
What I am watching
The next four weeks are the sharpest in this experiment. NVDA reports August 26. CRWD reports August 27. Claude has now doubled its NVDA position going into that date, which means those earnings carry twice the weight they did last week.
The question I will be watching is not just whether the numbers beat estimates. It is whether the thesis that drove Claude’s original purchase and this week’s addition is confirmed or challenged. There is a difference between a stock that goes up on earnings and a thesis that turns out to be correct. I want to know which one we get.
For ChatGPT, the watch is the same as it has been: RKLB and ASTS need operational news, not just price recovery, before the conviction numbers justify adding capital. That has not changed. The portfolio structure around those positions just got stronger.
The honest read on Week 8
The most volatile week of the experiment. Both portfolios took hits. Claude came out essentially flat while the Nasdaq dropped seven percent. Then both AIs made their second major capital allocation decision under pressure, and the contrast between the two approaches could not be sharper.
One is concentrating into its conviction going into earnings. One is fortifying the core and waiting on its speculative positions to earn more capital. Both approaches are coherent. Neither is obviously right yet.
August 26 and 27 will tell us a lot. I will be here for both.
What would you have done with the fifty dollars? Added to your conviction positions or fortified the core? Hit reply and tell me. I read every one.
Following the experiment? Subscribe to get the NVDA and CRWD earnings reactions, the Week 9 update, and every move both AIs make as it happens. The next four weeks are the ones that matter.