By Lynn Räbsamen, CFA | Advisory Board Member, CFA Institute | Author, Artificial Stupelligence
A follow-up to When ChatGPT and Claude Pick AI Stocks, Who Wins? and The Live Q1 Results Are In.
In January 2026, I handed two AI models the same prompt and $100,000 in start capital. Pick the high-conviction AI stocks that will beat the market over the next one to three years. Then buy and hold. No rebalancing, no interference, no second chances.
The benchmark was the Global X Artificial Intelligence & Technology ETF, a fund holding roughly 90 stocks. That is the diversified, sensible, do-nothing option. The whole point of the exercise was to see whether a concentrated bet from an AI model could beat it.
Two quarters in, one model has. The other has not. And the way the winner won is stranger than anyone would guess.
The Half-Year Scoreboard
Here is where the three portfolios stand at the midpoint of 2026.
| Claude | ChatGPT | ETF Benchmark | |
|---|---|---|---|
| Start Balance (Jan) | $100,000 | $100,000 | $100,000 |
| End Balance (Jun) | $154,530 | $121,373 | $129,905 |
| H1 2026 Return | +54.5% | +21.4% | +29.9% |
| Max Drawdown | -4.2% | -9.8% | -10.2% |
| Standard Deviation | 36.6% | 37.2% | 42.1% |
| Sharpe Ratio | 2.50 | 1.10 | 1.31 |
Claude turned $100,000 into $154,530. ChatGPT reached $121,373. The benchmark, doing nothing but owning roughly 90 names, reached $129,905.
Two quarters in, Claude is still leading by a wide margin. It has beaten both its rival and the index, and it is not close. That much you might expect from the winner of the first quarter.
What you would not expect is how it got there.
The Risk Story Nobody Ordered
Claude did not just win on return. It also won on risk.
Claude ran the lowest standard deviation of the three, at 36.6 percent, below ChatGPT’s 37.2 and well below the benchmark’s 42.1. It also posted the shallowest drawdown by a mile: minus 4.2 percent at its worst, against roughly minus 10 for the other two. More return, less volatility, smaller losses. The trifecta that is not supposed to come as a set.
The strange part is the index.
The benchmark holds roughly 90 stocks. It is the most diversified thing in the experiment by a distance. It should therefore be the calmest. Instead it was the most volatile of the three, jumpier than a ten-stock portfolio built entirely from semiconductors and data-centre gear.
The most diversified portfolio was the most volatile. That is not how the textbook reads.
Diversification is meant to smooth the ride. Spread your money across enough names and the bumps cancel out. Yet Claude’s concentrated ten produced a steadier line than the sprawling ninety. Owning more things, it turns out, is not the same as owning steadier things. Claude’s holdings were companies whose revenues rest on signed contracts and booked backlogs rather than on sentiment, and steady businesses make for steady portfolios, however few of them you own.
The Twist in the Timing
Now split the half into its two quarters. The story turns on its head.
| Claude | ChatGPT | ETF Benchmark | |
|---|---|---|---|
| Q1 2026 (Jan–Mar) | +10.59% | -7.70% | -8.24% |
| Q2 2026 (Apr–Jun) | +39.7% | +31.5% | +41.6% |
Read the Q2 column carefully. In the second quarter, Claude did not pull away. It slightly trailed the benchmark. The index had the best Q2 of the three.
This is not a typo. It is the lesson.
To see why, it helps to remember what the market actually did. The first quarter was a rout. A flare-up in the Middle East, oil spiking toward triple digits, tariff policy lurching from one ruling to the next, and the Federal Reserve rate cuts everyone had penciled in quietly erased from the forecast. Growth stocks took the worst of it.
Then April flipped the mood overnight. The Middle East tension eased, oil retreated, and a strong earnings season did the rest, with more than four in five large companies beating expectations. Money that had fled in March came sprinting back. May brought a scare of its own, as inflation resurfaced and a more hawkish Fed rattled the bond market, before June steadied again.
That is the shape of a V. And a V-shaped recovery flatters exactly one kind of asset: whatever fell hardest on the way down.
Claude did not win the half in the recovery. It won it in the wreckage.
When markets crashed in Q1, Claude was the only one of the three that made money. It gained 10.59 percent while everything else bled. Then, when markets roared back in Q2, it simply kept pace. It never needed to win the rally. It had already won the crash.
The benchmark told a perfectly ordinary story. It fell hardest, then bounced hardest. That is what beta does. Whatever drops the most tends to snap back the most.
ChatGPT is the real disappointment. It fell in Q1 and then rose the least of the three in Q2. Worst on the way down, weakest on the way up. There is no quarter in this half where its portfolio was the one you wanted to hold.
What Actually Moved the Money
Because nothing has been rebalanced since January, every result is the raw consequence of two decisions made once. Nobody trimmed a winner. Nobody rescued a loser. That makes the attribution unusually honest.
Claude’s engine: the names nobody talks about
The three largest single contributors in the entire experiment all sat inside Claude’s portfolio, and ChatGPT owned none of them.
Marvell added roughly $15,000. Vertiv, the data-centre power and cooling holding, added nearly $13,000. Micron, the memory play, added over $9,000. Add Taiwan Semiconductor, ASML and Arista, and you have a portfolio running almost entirely on the physical plumbing of artificial intelligence.
The timing was not luck. Through the first half of 2026, the market’s enthusiasm migrated steadily down the value chain, out of the software names that were supposed to use AI and into the hardware that has to exist for any of it to run. Analysts started calling it the picks-and-shovels trade, which is flattering to the metaphor and unflattering to everyone who bought the software.
Not one of Claude’s winners writes a chatbot or launches an app. They make the chips, the memory, the networking and the cooling without which the whole AI economy is a slide deck.
Claude did not bet on artificial intelligence. It bet on the parts of it that have to be manufactured.
ChatGPT’s engine: one chip did the heavy lifting
ChatGPT had exactly one triumph, and it was a big one. AMD contributed almost $12,000, its single best position by a wide margin, powered by a run of AI-chip commitments including a supply deal tied to OpenAI and design wins that pushed its next-generation accelerators into real contention with NVIDIA.
Here is the awkward part. Claude never owned AMD. It didn’t need to. Its three best names each outran ChatGPT’s one great pick.
When a single chip is the only thing keeping your portfolio respectable, the other fourteen picks were the problem.
What Held ChatGPT Back
Strip out AMD and ChatGPT’s portfolio is a catalogue of the names Claude deliberately refused to own.
Microsoft cost it over $3,000. Oracle, Palantir, Alibaba, Meta and C3.ai all bled, each dragging the total lower. These were the cloud platforms and software plays that were supposed to turn AI enthusiasm into share-price momentum. They did the opposite, as investors kept asking, out loud, whether hundreds of billions in capital spending would ever earn its keep.
The pattern from Q1 did not fade. It set.
ChatGPT bet that everyone in the AI economy would win. The market spent six months politely disagreeing.
What This Does and Does Not Prove
Let me keep my own findings honest.
Six months is not a cycle. Claude’s portfolio is concentrated, and concentration is a wonderful thing right up until the quarter it isn’t. A reversal in data-centre spending, one stumble at a large holding, or a rotation back toward software would compress this lead quickly. The low volatility is comforting, but it is six months of comfort, not a promise.
And there is the quiet fact under all of it. Neither portfolio has been touched since January. No skill was applied in flight. What we are measuring is not trading. It is the quality of one decision, made once, under identical conditions, by two different machines.
On that narrow question, the verdict so far is not subtle. One model understood that the AI trade had already moved from the software to the steel. The other is still waiting for its thesis to arrive.
The Closing Line
The headline of this half is not that Claude is winning. It is how. Not by picking the biggest rockets in the rally, but by owning the one portfolio that refused to fall when everything else did, and doing it with less volatility than an index four times its size.
There is an old, unfashionable idea buried in that. The surest way to finish first is to avoid finishing last when it matters. On aggregate, Claude should hold its lead for a good while yet. The open question is whether it can keep winning quarter by quarter, too.
The index spread its bet across ninety names and came second. ChatGPT chose fifteen and came third. Claude chose ten, and came first.
Q3 update in October. Until then, the lesson stands where it stood in April, only louder. Buy the pipes, not the hype.
***
Disclosure: This portfolio is for educational and experimental purposes only and does not constitute investment advice. Past performance does not guarantee future results. Investing in technology stocks involves substantial risk, including loss of principal. Readers should conduct their own research and consult financial advisors before making investment decisions.
For more insights about what AI can or cannot do, check out my book “Artificial Stupelligence: The Hilarious Truth About AI”.
Subscribe here to be the first to receive my insights.
