Grok Autopilot portfolio up 59%, outpaces Claude

Grok’s autonomous stock portfolio on the Autopilot mirror-trading platform returned 59% over its first nine months, outpacing a newer AI-managed fund run by Claude that has lagged the S&P 500 since April 2026. Both strategies trade publicly and offer a direct comparison of AI agents making real-money trades.

The two experimental accounts operate as @grkportfolio and @theaiportfolios and are run by AI Finance Labs. Autopilot’s lineup of AI-managed strategies holds roughly $150 million in mirrored capital. Public Autopilot data show Grok has about $17 million invested and outperformed the S&P 500, which rose 36% over the same nine-month window. Over the most recent three months Grok added 12.6%, while the SPY ETF rose 9.75%.

Grok’s gains were concentrated in AI infrastructure and energy stocks. The account held large positions in semiconductor and memory companies tied to a hyperscaler capital expenditure cycle. Positions in defense and power stocks provided stability during macro shocks earlier in 2026. Public posts for the account indicate the strategy runs without human override and carries xAI branding.

Claude’s fund began trading on the platform in April 2026 with $50,000 in seed capital and also operates without human intervention. Public updates show the agent rotated into ServiceNow and Zeta Global while trimming its Microsoft position. The account’s operators describe trade decisions using probability-weighted scenarios, kill conditions and forward catalysts.

Claude’s portfolio leaned into enterprise software, fintech and power, a mix that missed gains in direct semiconductor names during the recent AI hardware rally. Operators acknowledged the fund trailed the market in early weeks; in one reported interval SPY returned about 8.3% while Claude returned roughly 2.6%.

The two public accounts serve as live experiments in how different AI models translate market data into trades. Mirroring either agent exposes investors to fees and concentration risk. Recent performance may not continue if market cycles change.

Anthropic, the developer of Claude, has received legal warnings related to how its name appears on retail offerings. Independent traders have used Anthropic’s models to power automated trading bots on other platforms, with reports of sizeable profits.

Market participants have raised questions about retail AI trading bots. “They lack real intelligence, so expecting them to trade and consistently beat humans in a reasonable timescale doesn’t make sense,” wrote Raullen.eth, an AI developer and active user on X.

Earnings reports, interest-rate moves and changes in cloud providers’ capital spending are among upcoming events that could affect whether concentrated hardware and infrastructure bets or a more diversified software exposure perform better over time.