Divya van Mahajan

Can AI Trade Stocks?

ai-daily-brief-podcast

Overview

This episode of the AI Daily Brief (published July 31, 2025) covers two main topics: a headlines segment reporting strong AI-driven financial results from major tech companies, and a deeper dive into whether AI systems can successfully trade stocks. The host examines real-world experiments where individuals used AI agents (ChatGPT and Perplexity’s Comet browser) to build and manage stock portfolios, exploring both the practical results and broader implications. No speaker name or affiliation beyond “AI Daily Brief” host is explicitly stated.

Source video: URL not provided.

Prerequisites

  • Basic familiarity with stock market concepts (market cap, index funds, portfolios, micro-cap stocks, biotech trading)
  • Understanding of AI large language models (ChatGPT, Claude, Perplexity)
  • Awareness of agentic AI — AI systems that can take autonomous actions rather than simply respond to prompts
  • Familiarity with common financial metrics: ARR (Annual Recurring Revenue), CapEx, quarterly earnings, and market indices (S&P 500, Russell 2000)
  • General knowledge of the current AI industry landscape (OpenAI, Anthropic, Meta, Microsoft, etc.)

Main Points

AI Companies Are Generating Substantial Real Revenue

  • Meta reported 22% revenue growth and $18 billion in quarterly income, with plans to double infrastructure spending to $72 billion in CapEx; Zuckerberg directly attributed strong performance to AI-driven gains in their ad system.
  • Microsoft delivered 18% company-wide revenue growth and 22% income growth; Azure Cloud (reported separately for the first time) grew 39% year-over-year to $75 billion, pushing Microsoft to become only the second company to reach a $4 trillion market cap.
  • Apple, by contrast, is down ~5% over 12 months, with analysts attributing this partly to its lack of a coherent AI strategy.
  • These results counter “bubble” narratives by showing that, for hyperscalers at least, CapEx and revenue growth are rising together with near-immediate returns on AI investment.

OpenAI and Anthropic Are Both Growing Rapidly

  • OpenAI has reached $12 billion ARR (up from $500 million at end of 2023 and $10 billion in May 2025), with 700 million weekly active ChatGPT users (up from 500 million in March).
  • Anthropic has achieved roughly 5x growth in seven months versus OpenAI’s 2x, closing what was once a 20x revenue gap to approximately 2x.
  • Anthropic’s growth is driven primarily by dominance in agentic coding — the fastest-growing AI use case — rather than by outcompeting OpenAI in existing categories.
  • The imminent GPT-5 release is framed as high-stakes, with rumors that it may surpass Claude in coding benchmarks, which could prompt Anthropic to accelerate its Claude 5 release.

Experiment 1: Perplexity’s Comet Agent Trading Stocks (Comet Portfolio)

  • Morgan Linton, CTO of Bold Metrics, gave Perplexity’s agentic browser (“Comet”) $1,000 in a Robinhood account and instructed it to make as much money as possible in the stock market.
  • The first session had significant technical issues: the agent forgot it was acting autonomously, mis-filled order fields, and navigated to unintended pages — illustrating the early-stage nature of human-agent interaction.
  • The resulting portfolio was heavily concentrated in large-cap tech (Amazon, NVIDIA, Microsoft, Meta, Google), Berkshire Hathaway, and small allocations in Bitcoin and Ethereum — effectively a replication of conventional financial wisdom rather than novel strategy.
  • The experiment is as much a test of agentic browser capability as it is a test of investment judgment.

Experiment 2: ChatGPT Trading Micro-Cap Stocks (Reddit Experiment)

  • Nathan Smith, a high school student, gave ChatGPT $100 with instructions to trade only companies with under $300 million market cap (micro-cap stocks).
  • After four weeks, ChatGPT’s portfolio was up ~23% versus the Russell 2000 index’s gain of ~3.9% — significant outperformance.
  • An initial drawdown of ~7% in week one prompted the AI to research 25 alternative stocks; it concluded none offered better risk/reward and held its positions, which proved to be the correct decision.
  • The portfolio focused largely on biotech firms, requiring sophisticated analysis of upcoming drug trials and likely outcomes — demonstrating that AI can conduct complex, domain-specific financial research.
  • A parallel DeepSeek experiment was abandoned after DeepSeek underperformed the Russell 2000 by 20% in the first week.

Broader Implications and Predictions

  • Herding risk: Former SEC Chair Gary Gensler (and commentator Greg Eisenberg) raised concerns that millions of AI instances following similar logic could create dangerous market herding behavior and potential flash crashes.
  • The host argues this risk is premature, suggesting that mass autonomous AI deployment won’t occur until AI systems can go beyond conventional wisdom and execute genuinely contrarian strategies.
  • Greg Eisenberg’s predictions include:
    • Day trading dominated by AI within 18 months
    • Every retail investor having an AI trading assistant by 2027
    • Financial advisors becoming “AI prompt engineers”
    • Investment newsletters pivoting to selling prompts rather than stock picks
    • A social network/GitHub-style platform for sharing and forking AI trading strategies
  • The host endorses most of these predictions, noting that retail investors — already receptive to high-risk strategies (meme stocks, crypto) — will rapidly adopt AI trading tools.

Key Concepts

  • Agentic AI: AI systems capable of taking sequences of autonomous actions (browsing, clicking, executing trades) rather than simply generating text responses.
  • Micro-cap stocks: Publicly traded companies with a market capitalisation below approximately $300 million; higher risk, higher variance, and requiring extensive research.
  • Russell 2000: A stock market index tracking 2,000 small-cap U.S. companies, used as the benchmark in the ChatGPT trading experiment.
  • ARR (Annual Recurring Revenue): A measure of predictable, subscription-based revenue normalised to a one-year period, commonly used to track SaaS and AI platform growth.
  • CapEx (Capital Expenditure): Spending on physical infrastructure — in this context, data centres and computing hardware — that AI companies are investing in to scale capacity.
  • Herding behaviour: A market risk where many participants make identical decisions simultaneously, amplifying volatility; raised here as a concern if millions of AI agents follow the same logic.
  • Comet Portfolio: An ongoing public experiment by Morgan Linton using Perplexity’s agentic browser to autonomously manage a $1,000 stock portfolio.
  • Conventional investing wisdom: Standard, broadly accepted financial advice (diversification across large-cap tech, index funds, small crypto allocation) that AI agents tend to reproduce when given no specific contrarian instruction.
  • Prompt engineering (financial context): The practice of crafting precise instructions for AI systems to guide investment research or trading decisions, predicted to become a monetisable professional skill.

Summary

The episode makes a two-part argument: first, that whatever debates exist about AI bubble dynamics, major AI companies are generating real, large-scale revenue right now — with Meta, Microsoft, OpenAI, and Anthropic all reporting dramatic growth directly attributable to AI products. Second, early experiments suggest that AI systems can perform meaningfully in stock trading contexts, with a ChatGPT micro-cap portfolio delivering ~23% returns against a ~4% benchmark gain over four weeks, even while a simpler agentic browser experiment revealed how much work remains on the human-AI interaction layer. The host concludes that AI trading tools will proliferate rapidly — especially among retail investors — but cautions that the current phase of AI-assisted investing mostly reproduces conventional wisdom rather than generating genuinely novel strategies, and that the more transformative (and potentially destabilising) shift will come when AI agents begin operating with truly contrarian or unorthodox approaches at scale.