On 2 July 2026, Cognizant and Domyn announced a partnership to deliver AI systems into regulated European industries. The architecture they describe is layered and, for once, precisely stated. Domyn supplies the infrastructure and language-model layer, deployed on premises or in a private cloud inside the client’s own environment. Cognizant builds the application layer on top: smaller task-specific models, agents, domain execution. No data leaves the client’s controlled environment at any layer.
The go-to-market spans the UK and Ireland, DACH, Northern and Southern Europe, and the Middle East. Cognizant brings more than 60 AI patents and over 1,500 industry-specific agents to the arrangement.
Why the shape of the deal matters
Systems integrators have delivered regulated-industry AI for years, and the standard pattern was a hyperscaler underneath with the integrator building on top. That pattern has a specific consequence for a European client: the governance story rests on contractual assurances about a platform operated under another jurisdiction’s law, and the integrator’s leverage over that platform is limited to what its own vendor agreement provides.
Building the same practice on an infrastructure layer that sits inside the client’s environment inverts where the guarantees come from. Residency stops being a clause and becomes a property of the deployment. That a consultancy of Cognizant’s scale, roughly USD 20 billion in revenue, has organised its European regulated-industry practice this way is the signal worth reading. Integrators of that size do not choose delivery models for ideological reasons. They choose what regulated clients will sign.
The comparison becomes direct for Swiss buyers
The explicit DACH scope makes this concrete rather than abstract. A Swiss financial institution, cantonal body, or industrial group evaluating an AI programme in the second half of 2026 is choosing between two delivery models that are now both available at scale.
In the first, data transits infrastructure operated under a foreign legal regime, and governance is delegated to a vendor contract. In the second, data stays inside infrastructure the institution controls, and governance is enforced by how the system is built rather than by what the contract promises. Both can be delivered by credible partners. That is the change: sovereign deployment is no longer the option you take when the mainstream option is unavailable to you.
What to ask for
The practical takeaway for anyone running such an evaluation is that the layered architecture in this partnership is a reasonable template to hold suppliers to. Three questions separate a genuine version from a relabelled one.
Where does inference physically execute, and can that be demonstrated rather than asserted? Which party holds the logs of every model call, and under whose disclosure obligations do those logs sit? And if the application layer needs replacing in three years, does the infrastructure layer survive that change, or does the whole stack come out together?
Suppliers building on genuinely separable layers can answer all three. The answers are also the ones an auditor, and increasingly an insurer, will want on file.