As firms compete with frontier labs for elite researchers, building an in-house AI capability is becoming as much a hiring challenge as a technical one
Moonshot, the Chinese AI startup, released its Kimi K3 model to the world for public download last week. It is not just a new model; it signals that frontier AI is rapidly commoditising. If state-of-the-art reasoning models become abundant and free to access, the advantage shifts away from who utilises them and towards how firms can build their own proprietary infrastructure and data.
For funds trying to maximise their AI capabilities, they face a choice: use existing models through Anthropic or Moonshot, build proprietary tools, or launch their own research lab to develop advanced AI models.
Some of the largest platforms have followed this path. Bloomberg recently reported that Millennium was looking to launch its own bespoke AI research labs to expand the development and application of its current technologies, whilst Morteon Capital Partners (MCP), based in Mexico City, is exploiting the growing regional hotspots for AI talent, creating its own research lab as an extension to their core commodity fund offering. Les Finemore, CIO, 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.”
MCP is targeting researchers from some of the best universities and institutions to bolster this offering. Hedge funds are now not competing just with each other for talent, but with AI labs for elite mathematicians and computer scientists. A profile that would have been traditionally drawn to a quantitative investment firm, which has been at the forefront of machine learning development within the industry over the past decade.
Two Sigma exemplifies this; the firm recently created its own factor analysis platform called Venn, which is used both internally and by the firm’s investors. This innovation is garnering allocators’ attention; Paul Zummo, Head of Hedge Funds at JP Morgan Asset Management, told us earlier in the year the firm was consistently drawn to quant funds, citing their unique modelling approach that was able to navigate downturns.
However, despite quant funds’ attraction for some allocators. Richard Craib, founder of AI native hedge fund Numerai, believes the growing proliferation of AI will have a diminishing influence on their models. “For quant funds, its input is data and the output is returns. In the middle is software, risk modelling and mathematical reasoning. All of that middle can be executed by an LLM.”
—
Funds deciding whether to partner with a third-party vendor to grow their AI capabilities or develop research models in-house is becoming a key strategic dilemma. Liang Wengfeng demonstrated the close intertwinement between quantitative investment and the development of AI models after DeepSeek spun out from his hedge fund High-Flyer.
Yet, Craib believes that funds trying to emulate Wengfeng and develop research capabilities of a sufficient, scalable level will have to pass a significant bar. “Do you have the capability to build your own LLM? That is going to become a crucial barometer across the industry. If you don’t have that division in your business, or maybe only started recently, you are severely hamstrung.” Numerai runs its own LLMs, “using an agent to write code is easy; anyone can use Claude. The edge comes from building actual technology and almost becoming an AI lab yourself.”
To achieve this comes with considerable capital outlay; Millennium’s research lab venture is likely to cost $100m a year. For the largest platforms, expenditure in this direction is cost-efficient, as the marginal gains can result in multi-million dollar windfalls, but medium- to smaller-sized funds don’t have the resources to see tangible returns. George Kailias, founder of AI native hedge fund Aethon, notes, “The cost-benefit just isn’t there for the small platforms, but for the larger platforms, who have the capital resources, it’s going to become a default investment decision.”
Contrastingly, despite the vast capital the large platforms pledge to developing these proprietary tools, Joe O’Donnell, CEO of Canary Data, a bespoke AI provider for public market investors, believes very few funds will be able to create models that offer a material difference to performance. “Coming from a hedge fund background (O’Donnell ran the short portfolio at Tiger Global for 10 years), I don’t think these people fully understand the various functions needed to have a genuine edge. To create an edgeable model isn’t just prompt engineering; you need an intelligence layer and an emergent layer on top of that,” Kailias adds, “it’s not about tracking prices; it’s about the node framework; this is the key to creating adaptable and robust models.”
Funds able to demonstrate these capabilities will have another string to their bow when being assessed by allocators. In recent discussions, investors have repeatedly stressed that auditing a firm’s AI integration across their workflow process is a central part of a due diligence process. Chawkat Nammour, Portfolio Manager at Bainbridge Partners, notes there is increasing pressure for managers to continue to innovate: “Every hedge fund is experimenting with these tools and I am constantly looking for new ways AI can improve the workflow.” Meanwhile, Sid Ghatak, Co-founder of Increase Alpha, believes that in-house generative AI will be a prerequisite for funds. “Having generative AI as a central fulcrum in a firm’s workflow will be a necessity. Then the challenge is for funds: can you build upon that and create your own model? That’s where the lasting competitive edge will be.”
Despite the growing ubiquity of AI within funds’ workflow, there is still reticence among investors about it fully eliminating the role of human discretion within investment teams. Marcus Storr, Head of Alternative Investments at FERI, told us earlier in the year the firm would not be comfortable with running their clients’ money through hedge funds if it was done in a fully autonomous fashion. Similarly, Zummo added that funds obviously have to show AI capabilities, but the alpha it generates “will decay over the next couple of years”.
That decay is precisely why building an in-house AI research capability is about more than technological sophistication. It is a strategic bet on where future competitive advantage will lie and one that only the largest platforms can currently make. In doing so, the line between investing, quantitative research, and AI development continues to blur.
