The European enterprise technology sector is shifting its focus from foundational models to the operational systems required to deploy them safely. On June 15, 2026, Munich-based Celonis and Berlin-based deepset announced a joint venture to build a sovereign AI platform. The product is engineered specifically for mission-critical operations across defence, policing, cybersecurity, and critical infrastructure.
For technology leaders in the DACH region, the alliance represents a significant structural development. European organisations have spent the past two years navigating the tension between the mandate to adopt artificial intelligence and the strict data sovereignty requirements enforced by the EU AI Act. This partnership provides a domestic off-ramp. It allows highly regulated enterprises to deploy complex, agentic workflows without surrendering their data or infrastructure control to foreign hyperscalers.
The architecture of orchestration
The system integrates two distinct technological layers to address a primary failure mode of enterprise AI: hallucination caused by a lack of business context.
Celonis provides the operational ground truth through its Process Intelligence platform and Context Model. The system continuously extracts data from enterprise software to map exactly how a business functions, creating a factual record of workflows, bottlenecks, and data dependencies. By feeding this structural data into the AI platform, the system provides the boundaries necessary to keep autonomous agents grounded in actual business logic.
Deepset supplies the execution layer via its Haystack framework. This open-source orchestration tool is already utilised by institutions such as the European Commission to build, govern, and monitor multi-agent systems. When combined, the architecture allows an organisation to deploy agents that understand the precise mechanics of an internal process. These agents can execute tasks transparently, unifying fragmented documents, accelerating compliance investigations, and detecting operational risks before they escalate.
Neutralising vendor lock-in
The most critical element of the joint platform is its deployment model. The architecture is built to operate on the user's preferred infrastructure. It explicitly targets public sector entities and defence organisations that require absolute control over their models and data operations.
Milos Rusic, chief executive of deepset, noted that the collaboration aims to deliver a transparent European alternative for operational AI. He stated that organisations in mission-critical environments require systems they can fully trust, making auditability a core structural requirement rather than an optional compliance feature.
By neutralising the risk of cloud vendor lock-in, the platform appeals directly to German chief information officers. These leaders are currently trapped between boardroom demands for AI-driven modernisation and the legal reality of strict data privacy regimes. High-risk AI applications, as defined by the EU AI Act, mandate rigorous oversight mechanisms that are often difficult to guarantee when relying entirely on external cloud infrastructure.
The infrastructure burden
While the strategic logic of the partnership is clear, the operational reality of sovereign AI introduces a severe complication. The platform promises freedom from major US cloud providers, but achieving that freedom transfers the compute and maintenance burden directly onto the client.
Running multi-agent AI systems requires immense processing power, persistent memory management, and constant security patching. Deploying these architectures on sovereign European clouds or entirely on-premises infrastructure is an exceptionally heavy technical lift. It demands specialised hardware clusters, continuous model monitoring, and dedicated internal engineering teams to maintain system stability.
Many European enterprises and public sector bodies are currently ill-equipped to handle this level of infrastructure management. They lack the internal talent and the hardware procurement budgets required to match the reliability and scale offered by the major cloud providers. The appeal of total sovereignty is high, but the cost of maintaining it could limit initial adoption to only the largest or most heavily funded government agencies.
A maturing ecosystem
Despite the technical hurdles, the Celonis and deepset partnership demonstrates an important evolution in the European AI market. The narrative has moved past the attempt to build foundational models capable of rivaling the largest US developers. The focus is now firmly on the application and operational layers.
By combining deep process data with open-source agent orchestration, the two German companies are addressing a specific, high-value problem. They are providing the architecture required to make artificial intelligence functional and legally compliant for the most cautious sectors of the European economy. The success of the initiative will ultimately depend on whether those sectors have the technical capacity to run the platform they have been asking for.