Pyth Network has expanded rapidly from its roots in blockchain-based market data into a broader financial data platform, bringing together more than 138 institutional publishers and thousands of price feeds across equities, FX, commodities, crypto and other asset classes. Its ecosystem now includes names ranging from Coinbase and Fidelity to Kalshi, Revolut, Polymarket and Hyperliquid.
To learn more about Pyth Network, join the Pyth Connect live Q3 quarterly update.
Behind that growth is a change in how financial data is being distributed and consumed. Market data has traditionally been associated with terminals and screens, but an increasing proportion now flows directly into trading systems, analytics platforms and other applications through APIs. At the same time, the expansion of 24/7 and always-on markets is creating demand for pricing that extends beyond conventional trading hours, while autonomous AI agents are emerging as a new class of machine-based data consumer.
Pyth sits at the intersection of these developments. Its infrastructure was originally designed for continuously operating blockchain markets and has since expanded across traditional and digital asset classes, including 24/7 pricing for selected equities and other instruments.
Agentic AI could push that transition further. As autonomous agents begin to identify, retrieve and analyse financial information independently, market data infrastructure will increasingly need to support machines making their own decisions about what data they require and when they require it.
From terminals to APIs
APIs have already changed the relationship between financial applications and market data. Rather than accessing information primarily through a terminal, firms can bring data directly into trading, research, risk and analytics workflows and combine it with their own systems and datasets.
An API-first model can also simplify the process of adding new data. Pyth’s own approach, for example, allows users to discover datasets through Pyth Terminal, subscribe and obtain an API key for direct consumption. The company has been developing its platform around the idea of providing access across multiple asset classes through a common interface.
Established market data infrastructure, however, has evolved to handle difficult requirements around entitlements, licensing, reliability, governance and control. Making data easier to access doesn’t remove those requirements. Autonomous consumption may make some of them more important.
Michael Cahill, Founder and CEO of Douro Labs and Contributor at Pyth Network, sees a fundamental change in the identity of the data consumer. “Market data stops being something humans read and becomes something machines independently act on,” he tells TradingTech Insight. “That means infrastructure must be real-time, machine-readable and verifiable, but also radically simpler to access.”
From applications to autonomous agents
Conventional applications generally make predefined requests for information according to rules established by their developers. An AI agent can potentially go further, determining what information it needs, finding an appropriate source, retrieving the data, analysing it and using the result as part of a wider task.
Research, market monitoring, portfolio analysis and risk management are obvious areas for this kind of interaction, with more automated trading workflows potentially following. The underlying data infrastructure then has to support systems that select and consume information dynamically rather than simply responding to predetermined requests.
“AI agents cannot navigate fragmented vendors, asset-specific integrations and complex licensing agreements every time they need a new dataset,” says Cahill. “The future needs to look much more like every asset class, one API, zero licensing fees: a single programmable layer where agents can access trusted data across equities, FX, commodities, crypto and beyond, with clear provenance and permissions built in.”
Pyth has already begun adapting its infrastructure for this type of consumption, including support for Model Context Protocol (MCP), which provides a standardised way for AI models and agents to interact with external data and tools.
Rethinking access and control
An agent capable of independently interrogating multiple datasets could generate vastly more data requests than a human terminal user or a conventional application. Existing commercial models don’t necessarily translate neatly into an environment where autonomous agents continuously query and combine large numbers of datasets.
“Existing licensing models were built around humans, terminals and conventional applications, not autonomous agents consuming thousands of datasets continuously,” says Cahill. “The industry still needs to determine whether AI data is priced by query, inference, token, action or subscription.”
Data provenance becomes equally important. An autonomous system acting on financial information needs to establish where the data originated, how recently it was updated and what permissions govern its use. Firms will also need controls determining which datasets agents can access and what they are permitted to do with the information.
Distribution, entitlements, licensing, metadata and governance have often been treated as distinct components of the market data environment. Machine-driven consumption puts pressure on those components to work together more closely.
Building for machine-native finance
Pyth was originally developed to provide real-time financial data to blockchain applications, where machine-to-machine consumption was a requirement from the outset. It is now applying elements of that architecture more broadly across traditional and digital markets.
APIs have already made programmatic consumption of market data commonplace. Agentic AI adds another level of scale and independence, with autonomous systems able to select and interrogate datasets according to the task they’re carrying out. Infrastructure built around human users, terminals and predefined application workflows will increasingly have to accommodate that different pattern of consumption.
Cahill expects the shift to require a more fundamental rethink of established models. “Blockchain forced market data to rethink distribution once; AI is going to force an even bigger rethink,” he says.
As financial AI moves from answering questions towards carrying out increasingly complex tasks, reliable data access becomes part of what determines what those systems can actually do. Market data infrastructure will increasingly need to accommodate consumers that don’t look at screens, don’t work fixed hours and may decide for themselves which information they need.
This article is sponsored by Pyth Network. Explore Pyth Pro here.
Subscribe to our newsletter
