Two numbers from this month's enterprise AI reporting sit oddly together. Per-token AI prices have fallen roughly 98% since 2023. Over the same period, enterprise AI bills have tripled.

Falling prices and rising bills are only a paradox until you look at the pricing model. For commodity trading firms weighing AI inside their CTRM (Commodity Trading and Risk Management) systems, that model deserves as much scrutiny as the technology itself.

Why bills rise while prices fall

The driver is volume. Agentic AI workflows run autonomously across multiple stages, repeatedly invoking models, retrieving information, evaluating outputs and triggering further tasks until the job is done. Each step consumes tokens. As Ashish Nadkarni, group vice president at IDC, told BizTech this month: “Once you fire off an agentic AI work stream, it's not going to stop till it accomplishes the outcome.” He added that most firms have no way to see how efficiently those work streams run.

Usage is growing faster than prices are falling, and the meter belongs to someone else, creating the AI cost paradox that many businesses are now suffering from.

The metered model meets trading data

For a trading firm, per-token pricing through a third-party API carries a second cost that never appears on the invoice. The API sits outside your infrastructure, so the contracts, positions and P&L the AI needs to be useful must travel to it. The commercial edge leaves the perimeter with them.

Metered AI on sensitive trading data means paying an unpredictable bill for the privilege of exposure. Both problems come from the same architectural decision about where the AI runs.

A different cost architecture

Euclid AI runs on client-owned infrastructure or inside Euclid's private Swiss data centre, connected directly to the CTRM. The cost is the infrastructure and electricity the client already controls. There is no per-token meter and no third-party API between your data and the intelligence working on it.

The numbers reflect that structure. Euclid runs at $5k to $50k a year, against $45k to $200k for legacy cloud-dependent CTRM platforms. The figure is bounded by design, because the resources doing the work belong to you.

The question for year two

Any vendor can quote a price for month one. The harder question is what an AI feature costs in year two, once agentic workflows are running against your live trading data at full volume, on a meter you do not control.

Fixed cost and private data come from the same decision. Where the AI runs decides what it costs, and who holds your edge while it works.

Request a private AI workflow audit.