Hedge funds build in-house AI labs to gain edge

A number of large hedge funds are creating internal artificial intelligence research teams and labs and recruiting mathematicians and computer scientists after frontier models such as Moonshot’s Kimi K3 were released for public download. Firms say the availability of advanced public models has shifted attention to proprietary data, models and computing infrastructure.

Managers face three options: use third-party models, buy specialist services, or build internal research teams to develop custom large language models and systems. Millennium is forming a dedicated research unit. Morteon Capital Partners, based in Mexico City, is establishing a lab to draw regional talent. Les Finemore, CIO at Morteon, said, “The ambition is to build the premier AI research lab in Latin America. There is a huge amount of underappreciated and exceptional research talent.”

Industry participants say free access to high-end models reduces the gap in baseline reasoning ability. The remaining differences between firms are likely to come from proprietary datasets, custom model training, and the engineering stacks that link models to trading and risk systems. Two Sigma has built an internal factor analysis platform called Venn, which it uses for research and offers to some outside investors. Other firms are developing similar internal tools.

Building a research lab and hiring top researchers requires significant spending. One large firm’s proposed annual research budget is expected to approach $100 million. George Kailias, founder of AI-native hedge fund Aethon, noted that the expense puts full in-house development out of reach for many smaller managers and that large platforms are most able to justify the cost.

Some industry figures question how many funds will translate internal models into lasting performance gains. Joe O’Donnell, CEO of Canary Data, said he does not expect many funds to fully grasp the range of skills needed to produce a meaningful competitive advantage. Richard Craib, founder of AI-native fund Numerai, warned that the middle layer of investment systems-software, risk models and mathematical reasoning-must be recreated for an LLM-driven workflow to matter.

Allocators are increasing scrutiny of how managers integrate AI across trading, risk and research. Paul Zummo, head of hedge funds at a large asset manager, said persistent interest in quantitative funds is driven by their modelling approaches. Chawkat Nammour, a portfolio manager, said managers face pressure to keep experimenting with AI tools in order to demonstrate potential workflow improvements.

Some practitioners expect generative AI to be central to daily operations. Sid Ghatak, co-founder of Increase Alpha, described in-house generative systems as a likely core part of many firms’ workflows, with further advantage coming from a firm’s ability to extend those systems into proprietary models. Several allocators, including Marcus Storr at FERI, said they would not be comfortable with fully autonomous investment systems managing client capital.

The overlap between quantitative investing and AI research is increasing. Research teams have spun out from funds in some cases to form independent AI ventures. For large managers, the decision to build internal AI capability involves weighing heavy upfront budgets against the potential to improve trading performance, risk tools and product offerings over time.