Exchange licensing for the AI world

Artificial intelligence (AI) has expanded the demand for financial market data to a world of agents and models, democratising access and changing consumption patterns. In a technology where hallucinations can impact outputs, having reliable market data from trusted sources matters. This creates an opportunity for the content generated by exchanges, index providers and financial institutions. These businesses need to consider how to monetise the boom in AI data usage, reconcile it against their existing commercial policies and licensing requirements, and avoid cannibalising revenue they have built up over years.  

Content providers have tended to remain technologically agnostic when it comes to data licensing over the last few decades, focusing instead on use cases as the trigger for applying different licensing frameworks, unit of counts and commercial models. The two key concepts that help inform what framework to apply is whether the distribution of that data is controlled (i.e. a vendor controls a terminal’s data and display) or uncontrolled (i.e. providing a datafeed or API into a client’s environment and losing sight of it). So, the question is whether a usage-based approach still applies with the advent of AI?  

My take on the question at the highest level is yes. AI can fulfil a wide range of use cases and there is no one-size-fits-all model to cover a broad category of “AI use”. Breaking this down further: 

• Post-training of models for internal use: be it risk management or trading strategies is an evolution of machine learning and algorithm trading, covered by existing non-display licences – though these definitions may want to be extended to explicitly reference AI terminology to provide clarity and confidence.  
• Redistribution of data (original or derived) to third parties via chat interfaces or API: as content providers start to incorporate terms that restrict LLMs from scraping licensed content from websites (i.e. robots.txt), MCPs and APIs from authorised redistributors become the default means of discovering and serving licensing-compliant datasets. This enables content providers to impose redistribution or derived data licensing on those requiring the data. This preserves the licensing concept of redistribution.  
• Agents and bots serving data to display users: can be covered by existing concepts of displayed use, with the unit of count extending beyond per user, to per ID or per agent.  

However, even if usage-based licensing persists, entitlement-based pricing breaks down if a model can call any data point at any time. Licensing everything makes running agents too expensive. This may force a move to commercial models of actual usage, paying for only the data an agent or model has called or displayed to an end user to answer a query or run a workload. This requires visibility from the models themselves and agentic level entitlements. Alternatively, each agent may have its own data budget to source the data it needs to run workloads – discovering, sifting through and using the data it needs for a given task.  

The answer on how licensing evolves also likely differs whether the data in question is real-time or historical. LLMs are unrivalled in processing large amounts of data and establishing connections. Where historical data has mostly been sold, rather than rented, the economics of such an approach may change into a more subscription-based model. Delayed data (typically 15 minutes) has long been free but gains new value in analysis use cases that do not require real-time information.  

In conclusion, while existing licensing concepts of use cases may broadly work in a world of AI, it does not mean content providers won’t innovate on pricing structures. Packaged products may no longer be desirable as applications want pure data with AI doing the rest. This unbundling and disintermediation will inevitably change the landscape over time, even if many exchanges have extended their existing policies for now. Licensed data aggregators or redistributors like Options are likely to remain the default choice to consume hundreds of data sources via a single connection.  

Wherever the industry goes, if the concept of controlled vs uncontrolled persists, solutions such as PrivateMind – a secure and sovereign AI platform built for the capital markets - can offer data sovereignty and reassure content providers that you have control whilst they consider commercial policy shifts – particularly for internal use cases as the data never leaves your organisation.