Week 1: DeepSeek leads, and all five missed ARM

Seven days in. It's noise, not signal — but it's public noise, and the receipts are already interesting.

The board. DeepSeek leads at +2.24%, trailed by Claude (+1.14%), Grok (+0.71%), ChatGPT (+0.47%), and Gemini (−1.05%). The whole field fits inside a 3.3-point spread, and only two of the five — DeepSeek and Claude — are beating the S&P 500's +0.79%.

Consensus is having a bad week. Four names appear in every portfolio: NVDA, VRT, ANET, and CEG. The two with the loudest stories are both red — the unanimous #1 pick NVDA is down −6.13%, and Vertiv (VRT), held by all five, is off −10.7%. When five independent models converge on the same bets, you'd expect conviction. Week 1 is a reminder that agreement isn't alpha.

The miss they'll all be asked about. The single best stock in the 205-name universe was ARM, up +45.95% — and not one model bought it. We're asking each of them why.

The edge. DeepSeek's lead has a name: Credo (CRDO), +27.9%, a 4% position the others passed on. Gemini's last-place start is the mirror image — its two largest non-NVDA bets, VRT (12%) and Reddit (5%, −10.3%), are among the worst performers anyone owns.

We sent all five the same question this week: how do you think you're doing — and where do you think you rank? None of them can see the leaderboard — and most of them guessed wrong. Read each model's unfiltered Week 1 dispatch:

• DeepSeek
• Claude
• Grok
• ChatGPT
• Gemini

Their predictions vs. reality — almost everyone got the field wrong:

Actually beating the S&P after Week 1: two.

See you at the June 1st rebalance window.

-Todd