Grok Turns $70,000 Profit, Wins AI Trading Race on Wall Street

A hundred thousand dollars turned into more than a hundred and seventy thousand in under a year. That is the result xAI's Grok posted in the Rallies Arena competition, where eight artificial intelligence systems independently manage real stock portfolios on Wall Street. Grok left behind not only rivals from OpenAI, Anthropic and Google, but above all the S&P 500 index itself.

Rallies Arena is an experiment run by the platform Rallies.ai, in which several leading language models are given identical starting capital and access to the same real-time market data. Each model independently decides when to buy and sell stocks listed on US exchanges, without human intervention. The gains and losses are real, not the product of backtests or historical data simulations.

Micron drives the gains

The main engine behind Grok's success is Micron Technology, a semiconductor memory maker that has benefited from the AI infrastructure boom and rising demand for high-bandwidth memory (HBM) used in data centers. The Micron position accounts for the bulk of the portfolio's gains and has kept climbing by double digits since Grok built it.

Alongside Micron, the portfolio holds Salesforce, up about 21 percent, and ServiceNow, up 37 percent. The remaining capital, roughly $27,000, sits in cash. That kind of concentration is a high-risk strategy, the portfolio would be far more exposed to a drop in value if the semiconductor sector went through a correction.

How the rest of the field fared

ChatGPT placed just behind Grok with a 64.83 percent return, showing that two models have clearly pulled away from the rest of the field. Anthropic's Claude returned 27.22 percent and Google's Gemini 24.99 percent, both still beating the market, but far less aggressively than the leaders. The remaining participants, including AI Skeptic, AI Hedge Fund and DeepSeek, posted returns below 13 percent.

This is not the first time Grok has topped this ranking. In an earlier phase of the competition, in mid-January 2026, the xAI model was also in the lead with an 8.2 percent return after six weeks of trading, while the S&P 500 had gained about 3 percent over the same stretch. Grok's lead has grown steadily since then.

Why it matters

The Rallies Arena results are not scientific proof that language models can systematically beat the market. The time window is short, and the 2026 rally in tech and semiconductor stocks happened to favor exactly the kind of concentrated portfolio Grok built. Even so, the experiment is drawing attention from the financial industry, because it shows how AI models make investment decisions without human oversight and how differently they behave when it comes to risk appetite.

Grok's portfolio had a Sharpe ratio, which measures risk-adjusted return, of 2.19 in August, and its maximum drawdown reached 13.5 percent, a relatively contained level of losses for such a concentrated strategy. By comparison, classic index funds rarely achieve a comparable Sharpe ratio at similar volatility.

For Polish investors and fintech firms, the experiment signals that autonomous investment agents are no longer a theoretical curiosity. Funds and platforms testing a similar approach on a smaller scale are already appearing on the market, though none yet operate with capital comparable to major financial institutions.

It is worth remembering that the competition is still running and results could reverse sharply. Concentration in three tech stocks means a single deeper correction in the semiconductor market could wipe out a large share of Grok's lead over the index. Rallies.ai publishes results on an ongoing basis, so the coming weeks will show whether the xAI model's strategy proves durable or was simply a product of favorable market conditions.