Two research houses have examined AI in commodity trading this year and arrived at the same conclusion from opposite directions. The budgets are in place, but the data is the constraint.

Commodity trading has stopped debating whether to invest in technology, but the current numbers settle it.

What the spending looks like

In its June analysis of technology in commodity trading, BCG reports that tech-savvy trading firms now allocate 20 to 25% or more of their cost base to technology and reinvest 10% or more of gross margin into platforms and people. Those are software company ratios applied to physical trading businesses.

The consultancy also identifies what that spending is chasing. Trading performance increasingly depends on how quickly a firm can process unstructured data, allocate risk and trade across interconnected markets. The firms that convert raw information into positions fastest are the ones pulling ahead.

What the executives report

The spending side is one half of the picture. Earlier this year, HC Group and FT Longitude surveyed 131 senior executives across the industry for their report on AI in commodity trading. The respondents expect AI to add an average 3.34% to trading P&L by 2027. Asked what stands between them and that number, they ranked data quality, access and organisation as the top barrier to scaling AI.

Put the two findings together and the shape of the problem is clear. Firms are committing a quarter of their cost base to technology while their own executives report that the data feeding it is fragmented, inconsistent and hard to reach.

Why trading data stays fragmented

The reasons are structural. A trading firm's most valuable information arrives unstructured: contracts in dozens of formats, broker confirmations, counterparty emails, daily PDF price reports. It accumulates across desks, inboxes and spreadsheets, each with its own conventions. And the most sensitive of it, positions, P&L and counterparty terms, is precisely the material firms are least willing to hand to an external system for processing.

That last point is the trap. The standard vendor answer to fragmented data is to centralise it on an external AI platform. The data gets structured. The firm loses control of it. For information that constitutes the commercial edge, that is a poor exchange, and the survey respondents' caution reflects it.

Organising the data without giving it away

Euclid resolves the trade-off at the architecture level. Euclid AI runs inside the client's environment, deployed on client-owned infrastructure or inside Euclid's private Swiss data centre, connected directly to the CTRM. The model comes to the data. The data stays inside the perimeter.

Inside that environment, the unstructured data problem is exactly what Euclid AI works on today. It processes trading documents: drop a contract, invoice or confirmation into the integrated chat and the AI returns structured, populated fields for review, cutting an eight minute manual entry to under ten seconds. It ingests market data, parsing high, low and mean prices from daily PDF reports for automatic database updates. And it provides contextual platform support, answering user questions from the platform's technical documentation.

Each capability converts unstructured information into structured, usable platform data, which is the barrier the executives named, addressed where the data already lives.

The return on the spend

BCG's numbers describe an industry paying for an edge. The HC Group survey describes the constraint holding that edge back. Both point at the same asset. A firm's trading data is where the return on all that technology spend will come from, and it earns that return fastest when it can be organised, queried and put to work without leaving the firm's control.

Technology spend is a choice. Where your data goes to earn its return should be too.

Request a sovereign AI walkthrough.