Styxis
Company
Knowledge

A customer-owned data center still needs an operating-responsibility map

Separate physical ownership from responsibility for operating a sovereign cloud service.

Ownership is only one layer

A company can own a building while depending on another company to operate the systems inside it. That distinction becomes important when a sovereign cloud proposal is described as on-site. Before approving the architecture, ask who can administer each component, authorize access and approve a change. Truthound can check whether the resulting responsibility table is complete and internally connected; it cannot confer operational sovereignty.

A planned operating model

SAP and NTT announced planned cooperation on October 28, 2026, combining sovereign cloud and business AI capabilities with infrastructure and operations. An initial area is SAP Sovereign Cloud On-Site with NTT-managed infrastructure. The release describes intended cooperation and customer-controlled premises. It should not be read as a completed deployment or as a universal assignment of every operating responsibility.

The gap between two documents

The practical risk is a gap between two apparently sensible documents. A facilities list may identify the customer as owner, while the service design assumes a provider handles patching. Neither document may say who authorizes emergency access. If approval stops at the phrase customer-owned, the missing responsibility becomes visible only when somebody needs to act.

Name the operational functions

Make one row for each component and operational function. Record the physical operator, service operator, access approver and change owner, with references to the relevant service description. A role should point to a maintained responsibility register, rather than free text that merely looks familiar. An unresolved role is a design question, not an empty cell to fill with the nearest organization name.

Resolve references, not authority

For the component responsibility register prepared as a Polars DataFrame, Truthound Core Python SDK 3.3.4's ReferentialIntegrityValidator checks a supplied column against a reference column in provided reference data. It could flag an approver identifier that does not exist in the approved role register. Separate completeness rules are needed for missing assignments. The reference register must itself be authorized and current: matching a name cannot establish that the person has the contractual or legal authority claimed.

An existing dataframe alternative

A team with an established Pandera 0.35.0 workflow can retain its schema and custom membership checks. Pandera supports multiple dataframe backends, including Polars, so backend branding does not settle this choice. The reference set, predicate and update process still need explicit tests. Truthound's named reference check can make that particular rule easier to express, but that is an interface choice rather than evidence of stronger governance.

Review the scope of each role

Review the exceptions with operations and procurement together. Suppose two service components point to the same emergency approver but only one falls within that approver's mandate. Add the missing scope to the register instead of merely making both identifiers valid. The useful result is a responsibility decision that changes the architecture or agreement before an incident occurs.

Approval follows the service boundary

Approve the design only within the reviewed service scope and revision. Reopen it when a subcontractor, access procedure or component boundary changes. A clean responsibility extract can support that review, while actual access-control tests and the signed operating agreement supply different evidence. Keeping those evidence types separate avoids turning physical possession into a claim the design has not earned.

Evidence

Event checked
Subject checked
Modified