A follow-up to When ChatGPT and Claude Pick AI Stocks, Who Wins?
In January 2026, I published an experiment. Two AI models. One identical prompt. Two radically different portfolios. I promised a quarterly update.
Here it is.
And the first quarter did not disappoint.
A Quick Recap for New Readers
Back in January, I gave ChatGPT and Claude the same brief: build a high-conviction AI stock portfolio. Same constraints, same starting capital, no peeking at each other’s answers.
ChatGPT built a 15-stock portfolio spread across the full AI value chain — hyperscalers, cloud platforms, semiconductors, software. Classic diversification. Every major player in the ecosystem got a seat at the table.
Claude took a different view entirely. It built a concentrated 10-stock portfolio focused on what it called “irreplaceable bottlenecks” — the physical infrastructure that AI cannot exist without. Power management, advanced semiconductor manufacturing, high-speed networking, memory. It excluded cloud platforms completely. No Alphabet. No Meta. No Oracle. Just pipes, cables, and cooling systems.
Before the experiment went live, I ran a three-year backtest (January 2023 – March 2026) as a reference point. Claude’s portfolio returned 78.45% annualised. ChatGPT’s returned 63.96%. The benchmark ETF returned 29.80%. Interesting — but backtests are backtests. The real question was always what would happen with live market conditions.
Q1 2026 answered that question loudly.
The First Live Quarter: +10.59% vs. -7.70%
| Claude | ChatGPT | ETF Benchmark | |
|---|---|---|---|
| Q1 2026 Return | +10.59% | -7.70% | -8.24% |
| Direction vs. Market | Opposite | In line | In line |
| Backtest CAGR (3yr, for reference) | 78.45% | 63.96% | 29.80% |
Backtest period: January 2023 – March 2026. Live period began January 2026.
Let’s start with the headline.
From January through March 2026, Claude’s portfolio returned +10.59%. ChatGPT’s portfolio returned -7.70%. The Global X AI & Technology benchmark ETF returned -8.24%.
That’s an 18-percentage-point gap between the two portfolios in a single quarter. ChatGPT performed in line with the benchmark. Claude went the other direction entirely.
To be clear about what this means in context: Q1 2026 was not a quiet quarter. It was a bruising one. The S&P 500 fell 4.3% — its worst quarter since 2022. The Nasdaq dropped 5.8%. Tariff chaos, oil price spikes, and evaporating expectations of Fed rate cuts all converged simultaneously. This was the kind of quarter designed to expose the weaknesses in a portfolio.
Claude’s weakness, apparently, was not exposed.
What Actually Happened in Q1 2026
Three shocks hit the market in quick succession.
First, the U.S.-Iran conflict escalated sharply, pushing oil toward $100 a barrel and briefly closing the Strait of Hormuz. Energy stocks surged 38%. Almost everything else didn’t.
Second, tariff policy turned chaotic. A Supreme Court ruling invalidated a tranche of existing duties mid-quarter. Trump pivoted to new authority and reimposed tariffs at 10%, then 15%. Tech supply chains, most of which run through Asia, had no idea which rules applied to them on any given week.
Third, inflation expectations reset. The Fed cuts that markets had spent six months anticipating disappeared from the forecast entirely. The Atlanta Fed’s GDP tracker briefly went negative. The word “stagflation” started appearing in places it hadn’t appeared in years.
It was, in short, exactly the kind of environment where diversified exposure to growth stocks tends to hurt.
ChatGPT’s portfolio felt all of it. Claude’s portfolio, concentrated in physical infrastructure, largely didn’t.
The Stocks That Made the Difference
Claude’s outperformance wasn’t random. It came from specific holdings that were structurally positioned to benefit from the very conditions that punished the broader market.
Vertiv Holdings (VRT) — Claude’s top position at 12%
Vertiv surged roughly 58% in Q1 2026. The company provides power management and thermal cooling solutions for data centres — the unglamorous plumbing that keeps server racks from melting. In Q4 2025, it reported orders up 252% year-over-year. Its backlog swelled to $15 billion. In March, it was added to the S&P 500.
Every hyperscaler on earth is spending hundreds of billions on AI infrastructure. Every data centre that gets built needs power delivery and cooling. When oil prices spike and energy costs become front-page news, the companies managing power efficiency in data centres become more valuable, not less. Vertiv had a position in Claude’s portfolio at 12% concentration. ChatGPT did not hold it at all.
Marvell Technology (MRVL)
Marvell gained 16.5% in the quarter. On the final day of March, NVIDIA invested $2 billion directly into the company — integrating it into its NVLink Fusion ecosystem for custom chips, networking, and silicon photonics. The company reported data centre revenue up 47% year-over-year. Claude held it. The thesis played out.
TSMC (TSM)
TSMC climbed 12–14%, posting record revenue up 35% year-over-year and guiding for $52–56 billion in capital expenditure to ramp 2nm and 3nm chip manufacturing. When every AI company in the world is racing to produce the most advanced chips possible, and one company is the only one capable of building them at scale, that company tends to do well regardless of macroeconomic noise. Claude owned it. ChatGPT owned it too — but with less concentration and more dilution from positions that moved the other way.
The Stocks That Hurt ChatGPT
The casualties in ChatGPT’s portfolio were, almost without exception, the companies Claude had specifically chosen not to own.
C3. ai collapsed 37%. The enterprise AI software company reported revenue of $53 million against guidance of $72–80 million, cut 26% of its workforce, and revised its full-year outlook downward. ChatGPT held it as a speculative AI software play. It turned out to be exactly the kind of speculative AI software play that Q1 2026 was designed to punish.
Oracle dropped 23%. Its aggressive capital expenditure push sent free cash flow negative. Long-term debt crossed $100 billion. Reports surfaced of delays to the Stargate Texas project. Securities class action lawsuits were filed. ChatGPT had allocated meaningfully to Oracle as a cloud infrastructure play. It was, on paper, a reasonable thesis. Q1 did not agree.
Alphabet fell 2–3% — modest by comparison, but still a drag. Even the best-performing hyperscalers spent the quarter defending their capex decisions to sceptical investors who were starting to wonder whether $600–700 billion in annual AI infrastructure spending would ever generate adequate returns.
The pattern was clear. ChatGPT had bet that everyone in the AI ecosystem would benefit from the AI boom. The market, in Q1 2026, decided that was too optimistic.
Why Claude’s Thesis Held Up
The core of Claude’s portfolio logic — articulated back in January — was that the most durable way to invest in AI is to own what AI physically cannot exist without.
Not the companies using AI. Not even the companies building AI applications. The companies providing power, cooling, custom silicon, and high-speed networking to the data centres where AI actually lives.
This thesis had a specific Q1 advantage. When markets get nervous about AI valuations — when investors start questioning whether hyperscaler capex will generate returns, when software stocks sell off, when the macro environment deteriorates — infrastructure companies tend to decouple. Their revenues are driven by capital expenditure commitments already made, contracts already signed, backlogs already booked. Vertiv’s $15 billion backlog doesn’t evaporate because Oracle’s stock price drops.
Claude understood something that took the market a quarter to price in: the AI gold rush was always going to benefit the pick-and-shovel sellers more reliably than the gold seekers. When the rush gets turbulent, the shovel sellers keep getting paid.
One Quarter Is Not a Conclusion
To be fair — and this matters — one quarter of live performance proves very little on its own.
Claude’s portfolio carries a higher maximum drawdown than ChatGPT’s, even over the full backtest period. Concentrated positions cut both ways. A quarter in which infrastructure stocks underperform, or in which a single large holding disappoints, would look very different. The Vertiv thesis is compelling right now. It may look less compelling in Q2 if data centre spending slows or energy costs stabilise.
What Q1 2026 does confirm is that Claude’s portfolio construction had a genuine, identifiable logic — and that logic was rewarded in the first live quarter of the experiment.
Whether it continues to be rewarded is what the next three quarterly updates are for.
The experiment continues. Q2 update coming in July.
Until then — buy the pipes, not the hype.
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.
