By Fei Thompson, London
…
The debate around artificial intelligence in hedge funds is shifting. As adoption becomes more widespread, attention is moving away from how models generate signals and towards how capital is allocated, and how decisions are made under uncertainty.
A growing number of managers argue that the real edge no longer lies in predicting markets, but in structuring portfolios that can adapt dynamically to changing conditions.
One example is Alphacircle AI, which describes itself as an “AI-native” portfolio allocation system using deep reinforcement learning to dynamically allocate capital across US large-cap equities.
Rather than relying on predefined signals or factor models, the system treats portfolio construction itself as a learning problem, continuously adjusting allocations under explicit drawdown and risk constraints.
The firm reports returns of 36.2% in 2025, with a Sharpe ratio above 2 and maximum drawdown of around 8%, although these figures have not been independently verified. Structurally, the strategy is long-only, unlevered, and rebalances daily, with a stated focus on absolute returns and low market dependency.
While the approach reflects a broader shift towards AI-native portfolio construction, the lack of independently verified track record highlights the early-stage nature of many such strategies.
This shift is not limited to newer entrants.
At established firms, AI is increasingly being integrated into existing investment processes, not as a replacement for portfolio managers but as a way to augment decision-making.
Timothee Consigny, CTO and Head of GenAI Innovation, H2O Asset Management, said that the firm initially allowed its system to generate explicit trade suggestions, but later removed them after observing that portfolio managers were deferring too heavily to the model.
Rather than using GenAI to produce answers, the firm shifted its focus toward using it to help portfolio managers ask better questions, while keeping responsibility for investment decisions firmly with humans.
The change reflects a broader realisation across the industry: the challenge is not simply building better models, but ensuring that human decision-makers remain engaged in the process.
Rather than treating markets purely as a prediction problem, many firms are reframing them as a sequence of decisions made under uncertainty.
This approach aligns more closely with reinforcement learning frameworks, where the objective is not to forecast individual outcomes, but to optimise decisions over time given changing conditions and constraints.
At the same time, the rapid adoption of AI is raising questions about differentiation.
As more managers deploy similar tools and infrastructure, the technology itself is becoming less of a competitive advantage. Instead, the focus is shifting towards how effectively firms integrate AI into their broader investment processes, governance structures and risk frameworks.
This is particularly evident in how firms are managing the balance between automation and oversight.
While some strategies are marketed as “autonomous”, in practice most operate within tightly defined boundaries, with human accountability, auditability and the ability to intervene remaining central.
In many cases, execution is still handled by regulated brokers, ensuring that market interaction and risk controls remain within established frameworks.
For allocators, this evolution presents both opportunity and challenge.
AI-driven strategies offer the potential for more adaptive portfolio construction, reduced behavioural bias and improved risk management. However, they also introduce new questions around transparency, validation and governance, particularly where performance data is limited or not independently verified.
As AI becomes embedded across the industry, the distinction between “AI funds” and traditional quantitative strategies may begin to blur.
The more meaningful divide is likely to emerge elsewhere: between firms that treat AI as a tool to enhance existing processes, and those that attempt to rebuild the investment stack around it.
In that context, the question is no longer whether hedge funds will adopt AI, but how they will use it, and where, if anywhere, a durable edge can still be found.
