Access Is Becoming Free. The Premium Is Moving to Trust.

The machine consumer breaks seat-based licensing. The real question is not whether licensing survives, but where the edge goes once the pipe is cheap.

Seat-based data licensing rested on two assumptions: that the consumer of the data is a human, and that the human reads it through a terminal or a platform. Both are breaking at once. Once you accept that the buyer of market data is turning into a machine, the survival question answers itself, and a harder one takes its place. When access to data costs almost nothing, what are you actually paying for?

The closing panel I moderated at Neudata's Traditional and Market Data Summit in London had the survival question in its title: "The machine consumer: can traditional licensing survive agentic workflows?" Within ten minutes the discussion had moved on to the harder one.

The room was data vendors, buy-side data teams, and the people who price this stuff for a living. On the panel, four practitioners across the seats that matter: Abhijeet Gaikwad (Founder and CIO, Agami Capital), Alexander Lokhov (Data Strategist, Jump Trading), Wai Chung Ip (Senior Quantitative Developer, Deutsche Bank), and Jahmal Nicholson (Discretionary Data Product Lead, Man Group). What follows is my read of where the argument landed, not a transcript.

The seat was always a proxy

A per-seat license was never really about the seat. It was a proxy for consumption. One human, one screen, a bounded amount of data a person can read in a day, priced accordingly. That proxy held for forty years because the assumption under it held: a person sat between the data and the decision.

The machine consumer removes the person. An agent reads faster, in parallel, and at a scale no per-seat model was ever priced for. Point one analyst's workflow at an agent and it pulls in a morning what a desk used to pull in a week. The unit the license counted, a human reading a screen, stops being the unit that consumes the data. So licensing does not die. It loses the thing it was measuring, and the industry moves to what it should have measured all along: consumption, per query, per token, per unit pulled. That shift is mechanical and already underway. It is also the least interesting part of the story.

Access is becoming free

Here is the part that matters. The cost of connecting an agent to a data source is collapsing.

For most of the last decade, integration was the tax. Every new dataset meant a new API, a new schema, weeks of engineering to wire it into anything useful. That friction was a moat in disguise. It kept small teams out and kept switching costs high. Model Context Protocol and the wave of agentic connectors now shipping change the arithmetic. FactSet put out an MCP server for its portfolio analytics; others are moving the same way. The connection that used to take a quarter takes an afternoon. Access is being solved in public.

Which is why the "AI-ready premium" some vendors are floating does not survive contact with a serious buyer. The pitch is that machine-readable, agent-friendly data should cost more. The best line of the session put the objection plainly: charging an AI premium is like charging extra to open the file in Excel. Machine-readability is table stakes now, not a feature. Nobody pays a premium for the pipe once the pipe is standard.

Clay Christensen named this pattern years ago, the Law of Conservation of Attractive Profits. When one layer of a value chain commoditizes, the profit does not vanish. It migrates to the adjacent layer that is still scarce. Access is commoditizing. So where does the profit go?

The edge moves upstream and downstream

Two directions, and both were in the room.

Upstream, to data nobody else has. If access is free, the only data worth a premium is data that is genuinely scarce: proprietary, exclusive, or assembled in a way that is hard to copy. Commodity data delivered through a beautiful agent-ready pipe is still commodity data. The pipe got cheaper. The content did not get more valuable by traveling through it.

Downstream, to judgment and trust. This is the one I would put money on. When everyone can pull the same data into the same models, the differentiator is what you wrap around it: the reasoning, the context, and above all whether you can trust the number that comes out. The phrase that stuck from the discussion was "decision-ready, not AI-ready." Nobody at a fund pays more because data is machine-readable. They pay for data they can act on without checking it three times.

There was a useful tension here too. One view held that consumption pricing and dataset slicing finally let a small team compete on cost, spreading one budget across ten datasets instead of two. The counter was that more data reaching the desk makes the filtering harder, not easier. Both are right, and together they make the point. Once access is cheap, the scarce skill is not getting the data. It is trusting it, and knowing what to ignore.

What trust actually costs

People underrate how expensive trust is, because getting the data was always the visible line item.

When I was running risk at a multi-strategy hedge fund, the data was never the hard part. We had Bloomberg, we had the risk vendors, we had internal books. The hard part was trusting the number. The vendor's exposure did not match the internal exposure, and someone had to find out why before the CIO acted on it. A stress number was only as good as the reconciliation behind it. Most of the real work in a risk function is not acquisition. It is reconciliation, provenance, and the unglamorous business of making three systems agree on one figure.

That work does not get cheaper when access gets cheaper. If anything it gets harder, because the agent can now pull ten times as much, ten times as fast, from ten times as many sources. The bottleneck moves from "can I get the data" to "can I trust what just landed on the desk." That is where the premium goes. Not to the pipe. To the layer that makes the output trustworthy.

Where this leaves a lean team

For a small fund this is good news, if you read it correctly.

The old world priced you out. Seat licenses, integration budgets, and a quant bench to wire it all together were the price of entry, and a five-person fund could not pay it. Consumption pricing and cheap access remove that barrier. The same team can now reach data that used to sit behind a six-figure enterprise agreement.

Do not mistake cheap access for an edge. Everyone gets the cheap access. The edge is in the trust layer, and that layer is buildable without a twenty-person team: agents that sit on top of the stack you already run, reconcile across your sources, and hand the desk a number you can act on. The moat is no longer that the data was hard to get. It is that your answer is one you can defend. That is the version we are building for lean funds, and I think the next few years of this reward the operators who see the shift early.

So, back to the question the panel opened with. Does traditional licensing survive the machine consumer? It gets repriced, and fast. But that was never the real question. If your moat was that the data was hard to get, that moat is draining. The one worth building now is the one that makes your data trustworthy faster than anyone else can. When the pipe is free, who do you trust to tell you what the number is?