AI Trading Is Exploding — But the Bots Are Losing Money

The broader algorithmic trading market is also expanding rapidly. Multiple research firms estimate the global algorithmic trading market at approximately $20–$28 billion in 2026, though definitions vary . The TradeAlgo annual report projects it will surpass $32.8 billion in 2026, growing at a 13–15% CAGR . Other estimates range from $20.23 billion (Mordor Intelligence) to $27.17 billion (Research and Markets) depending on market boundaries. AI-driven systems now handle nearly 89% of global trading volume by 2025, with the broader market projected to reach $35 billion by 2030 .

The Evidence Gap: Bloomberg's AI Trading Experiment

Despite the hype, the actual trading performance of leading AI models has been underwhelming. Bloomberg ran 32 AI trading contests where frontier models from OpenAI, Anthropic, Google, and xAI were each given $10,000 and two weeks to trade U.S. tech stocks . The results were brutal: the overall portfolio lost roughly a third of its capital. Across all 32 sets of results, an AI model finished in profit only six times.

The systems traded too much, made wildly different decisions when given identical instructions, and consistently failed to generate positive returns . The best performer was Grok 4.20 (xAI), which delivered the best result when it was aware of rival performance, placing only 158 trades compared to others — under the same prompt, Alibaba's Qwen traded 1,418 times . But across the board, the verdict was blunt: most AI trading bots lose money .

Risks and Regulatory Warnings

Herding behavior and systemic risk: The European Systemic Risk Board (ESRB) warned in a December 2025 report that AI can create systemic financial risks through common exposures, interconnectedness, and AI-generated misinformation leading to widespread market mispricing . The U.S. Federal Reserve conducted experiments showing that while AI agents made more "rational" decisions than humans in isolation, increased reliance on AI-powered investment advice could still amplify herding dynamics . Academic research confirms that as trading algorithms learn from past trading history, they start to emulate — and occasionally magnify — the behavioral biases of human investors, leading to "algorithmic herding" .

Flawed and unaccountable AI recommendations: A June 2026 letter from U.S. Representatives to brokerage platforms highlighted that firms offering AI agentic trading tools explicitly disclaim accuracy, completeness, or suitability of AI outputs, and warn that agentic trading involves "significant risk" — with no control, supervision, or auditing of these agents . The letter notes that brokerages "may not guarantee the accuracy, completeness, or suitability of any agent output, nor do they control, supervise, monitor, recommend, or audit these agents" .

Regulatory warnings: The European Securities and Markets Authority (ESMA) issued a formal warning in March 2025 on using AI for investing, advising retail investors of the risks of relying on AI tools without understanding their limitations . A U.S. Senate report on hedge fund use of AI raised concerns about inadequate disclosures to clients and the potential for increased risks to market stability .

Behavioral biases in AI-assisted decisions: A July 2026 study found that young retail investors using AI tools exhibit herding mentality, overconfidence, and confirmation bias — and that higher trust in AI correlated with reduced investment discipline . Another study found that AI-informed portfolios show lower diversification and increased herding behavior during volatile periods .

Real-World Fallout

These concerns aren't hypothetical. In late July 2026, Situational Awareness, a high-profile AI-driven hedge fund, suffered a severe blow-up that required a 24-hour race to salvage its positions, as reported by Bloomberg . The Wall Street Journal documented a simultaneous "day the bots broke loose" event , while Bloomberg noted AI hedge fund collapses hitting amid tech earnings turmoil . These events underscore that flawed AI recommendations and herding behavior can translate into real, large-scale losses.

The Bottom Line

The democratization of AI-powered quantitative trading tools is real and accelerating. Retail investors now have access to capabilities that cost hedge funds millions to build just five years ago . But the evidence so far suggests that most AI trading systems — even frontier models from leading labs — consistently lose money when deployed in live markets. Meanwhile, regulators across the U.S. and Europe have issued formal warnings about systemic risks, flawed recommendations, and the lack of accountability in agentic trading tools. The technology may one day fulfill its promise, but for now, the gap between the hype and the results is stark.