•
2 minute read
Who owns the infrastructure your AI runs on? The question we left standing
Continuing our series of questions worth asking of any AI vendor demo, we chose one that might seem to have an easy answer, but is, in fact, surprisingly subtle.
Location is the easy answer, and the wrong one
Ask most vendors who owns their infrastructure and the answer arrives as a location. A data centre in Frankfurt. A product with “sovereign” in its name. A map pin, effectively, standing in for a legal question.
Location and ownership are not the same fact. A European data centre operated by a US-incorporated cloud provider is still subject to the CLOUD Act, which gives US authorities the right to demand data from American companies regardless of where that data physically sits. Microsoft's own legal team confirmed as much to French lawmakers under direct questioning: they could not guarantee European data would be protected from US government access. No data centre address changes that structure. Neither does a “sovereign cloud” label.
What ownership actually determines
Ownership is not a compliance checkbox. It determines a single, practical fact: who can a government compel, and through what legal route.
For a commodity trading firm, the data in question is not abstract. It is positions, counterparty terms, pricing logic built up over years, the contracts and confirmations that describe $12 trillion in annual trades across the industry. 67% of legacy CTRM data already sits inside US-based clouds. Every AI feature layered on top of that infrastructure inherits the same jurisdictional exposure the underlying platform already carries, whether the interface says “sovereign” or not.
The ownership question, asked properly, is this: if a foreign authority wanted this data, whose door would they knock on, and could that door legally be forced open regardless of where the servers physically stand?
Euclid's answer is a fact, not a location
Euclid's infrastructure is 100% client-owned. Deployment runs on the client's own servers or inside Euclid's private Swiss data centre, with no US hyperscaler in the ownership chain and no multi-tier corporate structure standing between a client's trading data and a foreign subpoena. The private AI layer, trained on 15+ years of proprietary commodity trading data, runs inside that same environment across five operational domains: Trading, Risk, Operations, Finance, Management.
The cost structure follows the same logic as the ownership structure: fixed at $5k to $50k a year, against $45k to $200k for legacy cloud-dependent CTRM, because ownership of the infrastructure removes the incentive to meter usage in the first place.
Three questions, three answers
Where does the model run: inside your environment, not a vendor's. Who owns the infrastructure: you do, or Euclid does on your behalf inside a private Swiss facility, never a US hyperscaler standing between the two. What happens to the cost when usage doubles: nothing, because the price was never tied to usage to begin with.
