Sovereign AI & Control AI
Introduction
Artificial intelligence is moving from experimentation into operational infrastructure.
Across Stockholm MLOps events, a clear pattern has emerged. Organizations are no longer primarily asking which model to use. Instead, they are asking where systems should run, who controls them, how they are governed, and whether they can be trusted under real-world operational constraints.
This shift affects healthcare, public sector, defense, manufacturing, financial services and enterprise environments alike. While the use cases differ, the operational questions are becoming remarkably similar.
Can organizations deploy AI without becoming dependent on a single provider? Can they move workloads between environments as requirements change? Can they maintain control over infrastructure, governance and data while still moving quickly enough to innovate?
These questions sit at the heart of sovereign AI and modern AI infrastructure.
But there is another question worth asking.
When organizations talk about sovereign AI, are they really looking for sovereignty?
Or are they looking for control?
For many organizations, sovereignty is not the objective itself. The objective is maintaining meaningful control over critical systems, infrastructure, data and decision-making. Sovereignty is often the mechanism through which that control is achieved.
The recurring signal from Stockholm MLOps is clear:
AI in production increasingly resembles critical infrastructure rather than traditional software.
Why This Matters
For years, sovereignty was discussed primarily through the lens of geography and regulation.
Where is the data stored?
Which country owns the infrastructure?
Which laws apply?
Those questions remain important, but they no longer tell the whole story.
Organizations increasingly discover that sovereignty is an operational capability. It influences infrastructure architecture, deployment strategy, governance models, procurement decisions, resilience planning and long-term flexibility.
A company may technically satisfy regulatory requirements while still lacking operational control. It may have access to powerful models while lacking the infrastructure required to deploy them safely. It may have governance policies in place while lacking visibility into how AI systems actually behave in production.
This distinction is becoming increasingly important.
Sovereignty addresses ownership, independence and strategic flexibility. Control addresses the practical ability to operate, govern, monitor and adapt AI systems once they are running in production.
In practice, many of the conversations taking place under the banner of sovereign AI are actually discussions about control.
Can workloads be moved between environments?
Can governance policies be enforced consistently?
Can organizations avoid being locked into a single provider?
Can they understand, audit and influence how AI systems behave?
Can they remain resilient as technology, regulations and business requirements evolve?
Viewed through this lens, sovereign AI becomes less about politics and more about operational capability.
The goal is not necessarily to own everything.
The goal is to retain meaningful control over the systems that matter.
This is why sovereign AI is increasingly becoming an operational discussion rather than a political one.
How Europe Defines Technological Sovereignty
One of the challenges with sovereign AI is that the term means different things to different people.
For some organizations it means data residency.
For others it means infrastructure ownership.
For others it means reducing dependency on foreign providers.
The European Commission increasingly frames technological sovereignty as Europe's ability to make its own technology choices, reduce critical dependencies and maintain control over strategically important digital capabilities. The concept spans semiconductors, cloud infrastructure, artificial intelligence, open-source technologies, compute capacity and digital resilience.
Importantly, the discussion is not centered on complete self-sufficiency.
The objective is not to isolate Europe from the rest of the world.
Instead, the goal is to ensure that Europe has the capability to develop, deploy and operate critical technologies while maintaining control over key infrastructure, reducing strategic dependencies and preserving long-term resilience.
This aligns closely with many of the conversations emerging across Stockholm MLOps.
The recurring question is not whether organizations should use cloud services, open-source models or global technology providers.
The question is whether they retain meaningful control over how critical AI systems operate, where workloads run, how data is handled and how governance is enforced.
Viewed through this lens, sovereign AI becomes less about politics and more about operational capability.
It becomes a question of infrastructure, deployment flexibility, governance, resilience and trust.
European Momentum Is Accelerating
Many of the infrastructure themes emerging across Stockholm MLOps are now appearing at the highest levels of European industrial strategy.
In June 2026, the European Commission announced a major expansion of Europe's AI infrastructure ambitions, including additional AI Factories, expanded compute capacity and broader support for organizations building AI systems across Europe.
What makes the announcement significant is that it focuses less on models and more on infrastructure.
The discussion centers on compute, operational capacity, deployment environments and long-term technological competitiveness. In many ways, it mirrors the conversations taking place across Stockholm MLOps events, where practitioners increasingly focus on infrastructure ownership, sovereign deployment options, workload placement, runtime governance and operational control.
The emergence of AI Factories signals a broader realization. Access to advanced AI increasingly depends on access to infrastructure. The ability to deploy, govern and operate AI systems is becoming a strategic capability in its own right.
This reinforces a trend visible throughout the Stockholm MLOps ecosystem:
Sovereign AI is no longer simply about where data resides.
It is increasingly about whether organizations have the operational flexibility, infrastructure control and deployment capabilities required to run AI systems under real-world conditions.
Sweden's AI Factory: From Strategy To Reality
The discussion around sovereign AI often remains abstract.
Mimer AI Factory makes it tangible.
As part of the European AI Factories initiative, Mimer represents Sweden's contribution to building the infrastructure required for the next generation of AI systems. Led by RISE, Linköping University and NAISS, the initiative provides access to compute resources, expertise and support designed to help researchers, startups, enterprises and public-sector organizations develop and deploy AI solutions.
What makes Mimer particularly relevant to this theme is that it addresses many of the challenges repeatedly discussed throughout Stockholm MLOps events.
Organizations increasingly understand the potential of AI. What they often lack is access to infrastructure, operational expertise and the practical support required to move from experimentation into production.
Mimer was created to help close that gap.
Rather than focusing solely on models, the initiative focuses on the broader operational environment needed to build and deploy AI systems. Access to compute, guidance, infrastructure and expertise are increasingly becoming strategic enablers for AI adoption.
This reinforces a recurring signal across Stockholm MLOps.
The future of AI will not be determined solely by access to models.
It will increasingly be determined by access to infrastructure.
Mimer represents one of the clearest examples of sovereign AI moving from policy discussions into operational reality.
Operational Signals From Stockholm MLOps
Across dozens of talks, panels and community discussions, several recurring observations continue to emerge.
Organizations increasingly want flexibility rather than dependency. Trust is becoming an operational requirement rather than a marketing claim. Infrastructure decisions increasingly determine what AI systems can and cannot do. Governance is moving closer to runtime systems. Public sector, healthcare and defense environments are often forcing infrastructure maturity earlier than the broader market.
Perhaps most importantly, the risk of failing to adopt AI is increasingly becoming as significant as the risks associated with adopting it.
The conversation is changing.
The focus is shifting away from models and toward operations.
Away from experimentation and toward production.
Away from AI as a feature and toward AI as infrastructure.
Emerging Patterns
Infrastructure Ownership Is Becoming Strategic
One of the strongest recurring signals across Stockholm MLOps is that infrastructure ownership is becoming a strategic question.
This does not mean every organization should own its own hardware. Rather, organizations increasingly want the ability to choose. They want flexibility in how workloads are deployed, optionality in how models are served, and confidence that they can adapt as requirements change.
As AI systems become more deeply embedded in business operations, infrastructure choices increasingly become business decisions.
Orchestration Is Becoming The Core AI Layer
Production AI increasingly consists of multiple models, routing layers, validation systems, retrieval architectures, fallback mechanisms and agent workflows.
The challenge is no longer simply generating intelligence.
The challenge is coordinating intelligence.
This is why many AI infrastructure discussions increasingly resemble distributed systems discussions. Routing, observability, failover, workload placement and runtime coordination are becoming core AI disciplines.
Governance Is Moving Into Infrastructure
Governance is no longer something performed after deployment.
The organizations making the most progress are embedding governance directly into infrastructure through deployment controls, runtime validation, observability systems, access controls and policy enforcement mechanisms.
The realization is simple:
Policies do not govern AI systems.
Architecture does.
AI Infrastructure Is Becoming A Competitive Advantage
The industry is slowly realizing that the hardest AI problems are increasingly operational rather than algorithmic.
The ability to deploy, govern, monitor and scale AI systems is becoming a competitive differentiator.
As model capability becomes more accessible, infrastructure capability becomes more important.
Timeline Of Evolution
February 2026 — AI Infrastructure, Orchestration & On-Prem AI (#28)
The conversation focused on infrastructure ownership, open-source AI stacks, orchestration, AI factories and secure deployment. It marked one of the first strong signals that governance was moving closer to infrastructure.
👉 Related Insight: AI Infrastructure, Orchestration & On-Prem AI
March 2026 — Optimizing Inference In Production (#29)
The discussion shifted toward inference economics, runtime scaling, KV-cache optimization and model orchestration. Production AI increasingly looked like distributed systems engineering rather than model engineering.
👉 Related Insight: Optimizing Inference In Production
April 2026 — Sovereign AI In Production (#31)
The focus moved toward trust, sovereignty, control and regulated deployment realities. Organizations discussed external API limitations, healthcare constraints and governance challenges.
👉 Related Insight: Sovereign AI In Production
June 2026 — European Technological Sovereignty Becomes Infrastructure Policy
On June 3, 2026, the European Commission significantly expanded its focus on technological sovereignty through a broader package of initiatives covering AI infrastructure, cloud infrastructure, compute capacity and long-term digital resilience.
The announcement reflects a major shift in how Europe approaches AI.
The discussion is no longer limited to regulation or AI safety. Increasingly, the focus is on building the infrastructure required to develop, deploy and operate AI systems within Europe. AI Factories, future AI Gigafactories, cloud infrastructure, supercomputing resources and operational AI capabilities are all becoming part of a broader strategy aimed at strengthening Europe's technological competitiveness and reducing critical dependencies.
For Stockholm MLOps, this announcement validates many of the themes that have repeatedly emerged across discussions on sovereign AI, AI infrastructure, inference optimization, governance and workload placement.
The conversation is increasingly moving beyond model capability.
The focus is shifting toward operational capability.
Who owns the infrastructure?
Where do workloads run?
How are AI systems governed?
Can organizations maintain control while still innovating?
These are the same questions that continue to surface across the Stockholm MLOps community.
2025–2026 — Mimer AI Factory Emerges As Sweden's AI Infrastructure Platform
While conversations around sovereign AI, infrastructure ownership and AI operations accelerated across Europe, Sweden simultaneously became part of the European AI Factories initiative through Mimer AI Factory.
Led by RISE, Linköping University and NAISS, Mimer provides access to compute resources, expertise and operational support for researchers, startups, enterprises and public-sector organizations building AI systems.
For Stockholm MLOps, Mimer represents one of the clearest examples of sovereign AI moving from strategy into implementation. It is not a concept. It is infrastructure that organizations can use today.
👉 Related Organization: Mimer AI Factory
June 2026 — Token Efficiency & Air-Gapped MLOps (#36)
The strongest signal yet that infrastructure is becoming the bottleneck. Retrieval, routing, workload placement, synthetic data and air-gapped environments dominated the conversation.
👉 Related Insight: Token Efficiency & Air-Gapped MLOps
Quotes Defining The Theme
"Running open models in production means owning the entire MLOps stack."
— Lucas Ferreira, Inceptron
"The optimization layer, not the model, is becoming the moat."
— Lucas Ferreira, Inceptron
"Companies prototype with closed models. Then the inference bill arrives."
— Lucas Ferreira, Inceptron
"If your product depends on one API, you don't have a system."
— Robert Luciani, Mediqtech
"We can't use external APIs. We need full control of the pipeline."
— Markus Boman, Careifai
"Calling something sovereign doesn't make it sovereign."
— Rui Gomes, 6G AI Sweden
"Even if AWS says data stays in Sweden, US law still applies."
— Ying Cheng, Arkus AI
"Scaling serverless inference is less about GPUs and more about what breaks when you add them."
— Matthijs Kok, evroc
"Inference infrastructure is becoming a distributed systems problem."
— Matthijs Kok, evroc
"You don't have to use just one model."
— Göran Sandahl, Opper
"Retrieval quality is what determines whether agents succeed."
— Ewa Szyszka, Qdrant
"Hybrid search should be your default."
— Ewa Szyszka, Qdrant
"Token waste happens in context accumulation."
— Ewa Szyszka, Qdrant
"You can either bring data to the model or bring the model to the data."
— Markus Kjellner, Cloudera
"You must innovate with GenAI to survive, but you cannot move your proprietary data to the public cloud."
— Markus Kjellner, Cloudera
"The software layer is becoming increasingly solved."
— Mikael Vesavuori, evroc
"We are moving toward workload orchestration across locations."
— Mikael Vesavuori, evroc
"The problem is not infrastructure. The problem is developer experience."
— Mikael Vesavuori, evroc
"Everything in the middle chain we must do ourselves."
— Peter Sundström, SAAB
"The data format itself reveals the capabilities of our sensors."
— Peter Sundström, SAAB
"Synthetic data is very important."
— Peter Sundström, SAAB
"We want to build a foundation model for the electromagnetic spectrum."
— Peter Sundström, SAAB
Definitions We Are Tracking
Stockholm MLOps follows several related concepts that frequently appear across community discussions, industry initiatives and public policy.
Sovereign AI
The ability to develop, deploy and operate AI systems while maintaining control over infrastructure, governance, deployment environments and critical operational dependencies.
Technological Sovereignty
The ability of organizations, nations or regions to make strategic technology decisions without excessive dependence on external providers or infrastructure. According to the European Commission, this includes capabilities across AI, cloud infrastructure, semiconductors, open-source technologies and digital resilience.
AI Factories
European AI Factories combine compute resources, expertise, data access and operational support to help startups, researchers, enterprises and public-sector organizations develop and deploy AI systems. AI Factories are a core component of Europe's strategy to strengthen AI capability and technological sovereignty.
Sovereign Infrastructure
Infrastructure that provides organizations with greater control over deployment, governance, security, workload placement and operational resilience while reducing critical dependencies.
Ecosystem Participants
Sovereign AI is not being built by a single company or initiative.
Across Stockholm MLOps, a growing ecosystem of infrastructure providers, AI platforms, governance specialists, AI factories, researchers and operators are contributing to the development of sovereign AI capabilities in Sweden and Europe.
While they approach the challenge from different perspectives, they are increasingly converging around common themes: infrastructure ownership, governance, operational control, deployment flexibility, trust and long-term technological independence.
evroc
One of the strongest voices in Sweden around sovereign cloud infrastructure, workload placement and European AI operations. Across multiple Stockholm MLOps events, evroc has consistently pushed the discussion beyond cloud adoption and toward operational control, sovereignty and infrastructure ownership.
Related Insights:
Mimer AI Factory
Sweden's AI Factory and part of the broader European AI Factories initiative. Led by NAISS, Linköping University and RISE, Mimer provides free access to compute resources, AI expertise and operational support for startups, researchers, enterprises and public-sector organizations. It represents one of the clearest examples of sovereign AI moving from policy into practical implementation.
Related Insights:
6G AI Sweden
A recurring contributor to the sovereign AI discussion, particularly around transparency, auditability, trust and what sovereignty actually means in practice. The organization has consistently challenged simplistic definitions of sovereign AI and helped move the discussion toward operational reality.
Related Insights:
Hopsworks
One of Europe's strongest examples of production AI infrastructure. Hopsworks has long advocated for open architectures, MLOps maturity, feature stores and operational AI systems that can be deployed across different environments. The company's work aligns closely with themes around infrastructure portability and operational independence.
Related Insights:
(Add when Insight is published)
Berget AI
Berget AI represents a growing movement toward European AI capability and infrastructure independence. Their contribution to Stockholm MLOps reflects increasing interest in building AI systems closer to where data, users and operational requirements exist.
Related Insights:
(Add when Insight is published)
Intric
Intric highlights another dimension of sovereign AI. As organizations operationalize generative AI internally, governance, knowledge management and deployment control become critical. Enterprise AI systems increasingly sit inside the broader sovereign AI conversation.
Related Insights:
8wave
8wave contributes an important perspective around AI governance and operational controls. The recurring message aligns closely with a major Stockholm MLOps theme: governance increasingly succeeds or fails inside workflows and systems rather than inside documents.
Related Insights:
AI Sweden
As Sweden's national AI center, AI Sweden plays a critical role in convening industry, academia and public-sector organizations around production AI. Many of the discussions around sovereign AI, infrastructure and operational AI systems intersect with initiatives taking place across the AI Sweden ecosystem.
SAAB
One of the strongest signals for sovereign and air-gapped AI in defense and mission-critical environments. SAAB's discussions around synthetic data, foundation models for the electromagnetic spectrum and operational AI systems illustrate where some of the most demanding infrastructure requirements are emerging.
Related Insights:
Related Insights
What We're Watching
Several developments are likely to shape this theme over the next 12–24 months.
The rise of sovereign inference layers. Air-gapped AI moving beyond defense into healthcare and critical infrastructure. Governance becoming embedded directly into runtime systems. AI Factories expanding across Europe. Increasing demand for infrastructure portability across clouds, sovereign providers, on-prem environments and edge deployments.
The common thread is simple.
Organizations increasingly want control without sacrificing capability.
What This Theme Represents
Sovereign AI & AI Infrastructure represents the shift from AI as a software feature to AI as controlled operational infrastructure.
It connects discussions around infrastructure ownership, governance, inference economics, healthcare, public sector adoption, defense requirements, orchestration, resilience and trust.
The core message is straightforward:
AI in production is no longer mainly about model capability. It is about who controls the system, how it operates, how it fails, how it is governed and whether organizations can trust it under real-world conditions.
This is the lens through which Stockholm MLOps continues to track the evolution of AI in production.