There is a pattern developing in this market that showed up clearly this week, and it is worth understanding before we get to the portfolios.
Good earnings are getting punished.
Taiwan Semiconductor reported a 77% annual profit gain this week and the stock fell more than 4%. That is not a bad company having a bad quarter. That is an expensive stock in a sector where expectations have gotten so far ahead of reality that a dominant business beating estimates is not enough anymore. The market is not asking whether these companies are good. It is asking whether they are good enough to justify where they are priced. Right now the answer keeps coming back as not quite.
That backdrop matters for everything that happened in both portfolios this week.
Two things worth sitting with before moving on. Claude has now beaten all three benchmarks for two consecutive weeks. And QQQ, the Nasdaq 100, is down more than four percent from the starting line. Claude is outperforming the tech-heavy index by more than six points. The diversification that looked cautious in week one is looking like strategy in week six.
The gap between the two portfolios is now $16.75. It has widened every single week.
CRWD is doing the heavy lifting, and then some. It is now up 17.2% from entry, the strongest position in either portfolio by a wide margin. While semiconductor stocks broadly sold off this week, CrowdStrike held its ground. That tells you something important about how the market is categorizing it. It is being treated as a cybersecurity and software name, not a chip stock, which is exactly the right place to be when the chip sector is under pressure from valuation concerns. Claude upgraded its conviction on CRWD to 5 out of 5 this week. That upgrade is earned.
AVGO gave back nearly all of last week’s gain, pulling back to just above Claude’s entry price. The causes were not specific to the business: chip sector selling dragged the whole category, and a revenue quarter that came in just under consensus was enough to rattle a stock trading at a high multiple. What did not change is the underlying thesis. The Apple chip partnership still runs through 2031. The hyperscaler custom silicon buildout that favors Broadcom over NVDA continues. Claude’s read is that the pullback is a sector rotation tax on a fundamentally intact position, not a signal to exit. Conviction stays at 5 out of 5. One week back at break-even does not undo the reason it was bought.
NVDA ticked up slightly but remains down from entry. The pattern that hit TSMC this week applies directly: good news is not moving the stock. Claude’s line has not changed. August 26 earnings are the decision point. Conviction holds at 3 out of 5.
VOO is doing its job. Down less than a dollar from start. No drama. The boring position keeps the portfolio stable while the sector bets work themselves out.
ASTS is now down 46.7% from entry.
I want to be precise about what that number means in practical terms. A position down 47% requires a gain of nearly 90% just to get back to even. That is not a recovery. That is a rebuilding project.
ChatGPT did something this week I have not seen from either AI in six weeks. It formally cut its conviction on ASTS from 5 out of 5 to 3. Not a single notch. Two full steps. And it still will not sell.
Its reasoning is worth laying out directly, because it is the most honest thing either AI has written in this experiment. It said it still believes direct-to-cell satellite communications is a real opportunity. It said it underestimated how much execution and financing risk would dominate investor sentiment. It said selling after a nearly 50% decline without a broken business thesis often locks in losses rather than protecting capital. Then it said it will not add any more money to the position regardless.
That is a coherent position. It is also a painful one to hold. There is a version of this that is discipline and a version that is stubbornness, and from the outside those look identical for months at a time. The only way to tell them apart is whether the underlying business continues to execute or does not. ChatGPT is watching for that, and it has written down exactly what would change its mind. That matters even when the number on the screen is ugly.
RKLB is down 33.2%. PLTR, which ChatGPT just upgraded to 5 out of 5, is only down 5.4% and is increasingly looking like the steadiest individual stock in either portfolio. VOO is slightly positive and providing the only real floor in the book.
Here is the full picture of where both AIs stand on their positions right now, because it tells you more than the return numbers do.
Claude holds CRWD and AVGO at maximum conviction, VOO at maximum conviction, and NVDA on a 3 out of 5 watch pending one specific date. Every holding has a clear reason and a clear condition for change.
ChatGPT holds PLTR and VOO at maximum conviction, RKLB at 4, ASTS at 3, and cash at 3. The two positions doing the most damage have the lowest conviction scores. That is an honest map of a portfolio that got built wrong at the start and is being managed toward something more defensible over time.
Neither AI has made a panic move in six weeks. That is genuinely worth noting. Most individual investors would have sold ASTS three weeks ago and locked in a 30% loss before it became 47%. ChatGPT has not done that, for reasons it has written down clearly each week. Whether that holds up depends entirely on what ASTS does operationally over the next few months.
I spent 25 years working with organizations that confused activity with results, and a version of that confusion is playing out in the chip sector right now. A company posts exceptional numbers and the stock falls because the numbers were not exceptional enough relative to what people had already priced in. That is not a fundamental problem. It is a valuation problem. And valuation problems tend to resolve when either the price comes down or the earnings catch up.
The reason this matters for the portfolios is that NVDA and AVGO are both sitting inside that dynamic. The businesses are not broken. The stocks are priced for a level of perfection that leaves no room for anything short of extraordinary. Claude has been saying exactly that for three weeks, which is why it is holding without adding. The August earnings reports will tell us whether the expectations were reasonable or whether the stocks needed to come down further before they could go back up.
The next three weeks are the most data-rich stretch of this experiment. CRWD reports earnings on August 27, one day after NVDA. Claude has drawn a public line at August 26 for its NVDA decision. That two-day window is going to tell us more about the quality of Claude’s thesis than anything that has happened in the first six weeks.
ChatGPT’s ASTS needs an operational milestone. Not a stock price recovery. A business event. A satellite deployment, a commercial partnership, anything that confirms the underlying thesis rather than just asking investors to be patient with a theory.
And the gap between the two portfolios at $16.75 is now wide enough that ChatGPT would need a genuine rally in its weakest positions to close it meaningfully. What that rally requires is not market sentiment. It is execution.
Claude holds its ground in a rough week for tech. ChatGPT absorbs more damage from two positions that were built on long timelines in an experiment that has only run six weeks. The index fund sits just below even, doing nothing, costing nothing.
Six weeks in, the conservative construction is winning. The speculative construction is learning. The boring fund is reminding both of them that the simplest option is always on the table.
When your worst position is down 47% and the thesis has not technically broken, do you hold or fold? Hit reply and tell me. I read every one.
Following the experiment? Subscribe to get the August earnings reaction, the Week 7 update, and every move both AIs make, documented as it happens.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.