An experimental stock portfolio managed entirely by Grok, xAI's large language model, has turned an initial $100,000 stake into $148,485 in under nine months — a 48.48% return that leaves the S&P 500's 14.10% gain over the same stretch far behind, according to Finbold. Had the same $100,000 simply tracked the index, it would be worth $114,101 today, putting Grok's outperformance at more than 34 percentage points.
The portfolio is entered in Rallies Arena, a public experiment launched in late November 2025 that gives competing AI models — including OpenAI's ChatGPT and Anthropic's Claude — identical starting capital and lets each manage its own live, tracked portfolio independently.
A Concentrated, High-Conviction Bet
Grok's results come from a portfolio concentrated in just three names. Micron Technology (NASDAQ: MU) makes up about 65% of the holdings at $95,800, with Salesforce (NYSE: CRM) at $26,900 and ServiceNow (NYSE: NOW) at $24,200 rounding out the rest. That concentration cuts both ways: the portfolio's Sharpe ratio of 2.19 points to strong risk-adjusted returns, but a maximum drawdown of 13.5% shows the approach hasn't been without stretches of real volatility.
Performance tracked the S&P 500 closely in the competition's early months before diverging sharply in April and May, then extending further after a rally in late May pushed the portfolio's value above $150,000. Returns have moderated somewhat since, but Grok has held its lead into August.
Grok Isn't the Arena's Top Performer
Framed against the rest of the field, Grok's run looks less like a runaway win and more like a solid middle finish. ChatGPT's Rallies Arena portfolio has returned 72.4% since launch — turning its own $100,000 into roughly $172,400 — by concentrating into a small number of high-conviction names, including mid-cap semiconductor firm Credo Technology, during a tech downturn earlier in the competition. Claude and other competing models have posted gains in the same broad 20-50% range as Grok's.
Related: Anthropic's Revenue Soars 14x to $11.5B as Run Rate Tops $47B
Paper Gains, Real Questions
None of this is live trading. Every Rallies Arena portfolio is a simulated, paper-money exercise, and the AI models face none of the real-world constraints — slippage, liquidity limits, market impact — that would come with actually deploying capital into a mid-cap stock like Credo or a concentrated Micron position. Still, the experiment has become a closely watched proxy for how differently frontier AI models approach risk and conviction when given identical inputs, and Grok's three-stock, high-conviction style stands in visible contrast to how its rivals have chosen to play the same nine months.
