You hardly ever get to see a certain area of a hedge fund office. There are too many monitors, not enough windows, and a stale espresso odor that seems to be ingrained in the carpet. The traders in movies don’t resemble the people inside. They appear to be mathematicians. That’s what they are. These are the experts—the scientists, engineers, and PhD holders in physics and statistics who have devoted years to creating the methodical models that trade billions of dollars every day. And despite everything you’ve read about white-collar jobs being replaced by AI, they are still very much in existence.
Over the past eighteen months, the discussion surrounding AI and hedge funds has taken a very realistic turn. A few years ago, it was impossible to attend a fintech conference without hearing someone predict that large language models would soon power entire portfolios. The same rooms are now filled with quant leads courteously outlining why that isn’t happening and most likely won’t be anytime soon. Amadeo Alentorn of Jupiter Asset Management stated bluntly at a recent Quant Strats conference in London that “human creativity” is what will keep quants ahead and that there is currently “too much hype” about what generative AI can truly do for investment management.
| Detail | Information |
|---|---|
| Industry | Quantitative hedge funds |
| Primary AI Tools in Use | Generative AI, LLMs, deep learning, RNNs, LSTMs |
| Dominant Use Cases | Research summarization, document analysis, code acceleration, alternative data processing |
| Real Edge Still Held By | Human quants, data scientists, risk managers |
| Notable Skeptic | Ken Griffin, Citadel founder |
| Jupiter Asset Management Lead | Amadeo Alentorn (Head of Systematic Equities) |
| H2O Asset Management CTO | Timothee Consigny |
| Morgan Stanley Quant Lead | Stephan Kessler |
| UBS Asset Management Quant Head | Matthias Uhl |
| Millennium Quant | Haoxue Wang |
| AIMA Survey Findings | Primary GenAI use = admin time savings & content generation |
| Key Back-Office Benefit | Compliance monitoring and document review |
| Front-Office Benefit | Alternative data parsing, bond prospectus analysis |
| Typical Latency Requirement | Microseconds for execution (colocated infrastructure) |
| Data Sources Used | Satellite imagery, credit card data, shipping data, sentiment, earnings calls |
| Common Infrastructure Layer | Proprietary quant stacks, Python, C++, custom ML frameworks |
| Biggest Limitation | LLMs can’t generate original alpha; they require direction |
| Industry Consensus | AI augments quants, doesn’t replace them |
| Allocator Concern | Governance, security, model risk |
| Notable Quote | “A language model can’t read your mind” — Haoxue Wang |
At the Robin Hood conference in New York in October, Ken Griffin, the founder of Citadel, which oversees over $65 billion and employs some of the world’s most advanced systematic strategies, made a similar statement. According to Bloomberg, Griffin informed the audience that generative AI “falls short” in identifying ideas that outperform the market. That’s a startling admission from a company that relies solely on computational advantage. Griffin would be the last person to publicly declare that AI would take the place of Citadel’s quants.
The real story behind these funds is both more fascinating and more dull than the headlines. Indeed, LLMs are now ubiquitous. At Morgan Stanley, they are analyzing bond prospectuses; according to Stephan Kessler, tasks that used to take days are now completed in minutes. They are assisting UBS teams in automating repetitive tasks. They are writing internal research notes, reviewing compliance logs for warning signs, and summarizing earnings calls. “Time savings on administrative tasks” and “content generation” for investor relations were the most common use cases for generative AI, according to an AIMA survey of hedge fund managers conducted last year. That’s helpful. Furthermore, the industry is not experiencing the revolution that was anticipated.
The quants are clear-eyed about the limits, which are real. The CTO of H2O Asset Management in Paris, Timothee Consigny, likened the current AI era to Formula 1 cars, saying that while everyone has access to extremely fast vehicles, very few people are proficient drivers. Even more blunt was Haoxue Wang, a quant at Izzy Englander’s Millennium: “A language model can’t read your mind.” More important than what the model was trained on is what you feed it. Furthermore, all of that feeding—feature engineering, data architecture, and domain intuition—remains essentially human labor.
Speaking with those who manage money professionally gives me the impression that the AI revolution is affecting hedge funds in the same way that all revolutions do: unevenly, partially, and with less fanfare than anticipated. The back office speeds up. The process of conducting research becomes cleaner. Half the time is spent writing code. However, people still possess the edge, insight, and intuition for what the market is mispricing at 11 a.m. on a Tuesday, which is what really generates alpha. Perhaps the most telling fact is that hiring continues even at companies that are making the biggest investments in AI. Not for quick engineers. For quants. The traditional type, who keeps a notepad filled with calculations and looks at every model output with suspicion.
