Most businesses now use AI for something, from automating routine tasks to analysing data. The more they rely on it, the more it matters who controls the technology underneath.
One answer is sovereign AI.
Germany's Kolibri model, built by Heidelberg-based Aleph Alpha, takes that approach: organisations can run it under their own control. It came out in October 2026 and is aimed at sensitive work in government and industry.
Below, we look at what sovereign AI is, why Europe wants it, and what it could mean for the tech sector.
What is sovereign AI?
Sovereign AI means AI that a country or organisation can build, run and govern itself, with real control over the technology, infrastructure and data involved.
Most organisations use AI services from large technology companies. They work well and they're easy to start with, but you end up tied to someone else's infrastructure and terms of service. If you handle sensitive information, that raises real questions about data protection, security, regulation and long-term control.
Sovereign AI gives organisations more say over how their AI is built and run. In practice, that usually comes down to four things:
- Data control: where information is processed and who can access it.
- Infrastructure independence: running AI on infrastructure you choose or manage.
- Technology ownership: models and components you can deploy and adapt yourself.
- Regulatory oversight: making it easier to meet legal, security and governance requirements.
Sovereignty comes in degrees. Using a European-built model doesn't automatically mean you control your data or infrastructure: how it's deployed, who owns it, who else has access and how it's run day to day all matter. The aim is to cut avoidable dependencies on critical technology.
What is Germany's Kolibri AI model?
Kolibri is a large language model (LLM) from Aleph Alpha, released on 3 October 2026. It's built for high-stakes work, mainly in public administration and industry, and it supports German and English, with a particular focus on German.
The key details:
- Open weights: the model weights are available under the Apache 2.0 licence, so developers and organisations can download, deploy and adapt it within the licence terms. The weights are hosted on Hugging Face.
- Mixture-of-Experts architecture: instead of using the whole model for every token, a Mixture-of-Experts design activates only selected parts, which keeps computing demands down.
- Size: about 78 billion parameters in total, with around 3.46 billion active per token.
- Long context: context windows of up to one million tokens (Aleph Alpha recommends shorter contexts for demanding tasks), handy for large documents and big collections of information.
All of these specs are vendor-reported from Aleph Alpha's model card. Independent testing is still to come.
Why was Kolibri developed?
Plenty of organisations want AI but can't send confidential documents, internal records or sensitive operational data to an externally hosted service.
Kolibri is designed to run on infrastructure the organisation controls, so businesses and public bodies decide how their data is handled. Likely uses include document analysis, information retrieval, internal knowledge systems and workflows built on specialist industry data. The open licence also makes it easy to test the model against your existing systems.
Open weights don't make a model secure, accurate or compliant on their own. You still need the right infrastructure, access controls, testing and governance.
Why is Europe investing in sovereign AI?
The US and China have produced the most influential AI companies, models and infrastructure. European governments and businesses want their own capabilities, less reliance on outside providers, and the freedom to make their own calls on critical technology. Three things are driving that.
1. Data protection and regulation
Financial services, healthcare and public administration handle confidential data every day. Sovereign AI gives them more options for controlling how that data is processed and where systems run. That can help when designing deployments around the GDPR and the relevant parts of the EU AI Act.
The technology doesn't guarantee compliance, though. Each organisation still has to assess its own use cases and legal obligations.
2. Technological independence
Depending on a handful of external AI providers carries risk. Price changes, service outages, access restrictions and new commercial terms can all hit the companies that rely on them. Having alternative models makes it easier to spread that risk.
3. European innovation and competitiveness
Competitive AI takes serious investment in research, computing power, engineering talent and commercial development. Projects like Kolibri help build Europe's AI ecosystem and produce technology suited to regional languages, industries and regulation. The opportunity goes beyond the models themselves to the infrastructure, software, security and specialist skills needed to use them.
What could sovereign AI mean for businesses?
For businesses, it's one more option when deciding how to adopt AI. A company might use a hosted AI service for everyday productivity, and run a separate model in-house for confidential documents or sensitive workflows. The right mix depends on your requirements, technical capability and budget.
Sovereign AI could suit organisations that need to:
- process sensitive internal information
- build AI tools around proprietary data
- keep tighter control of their technology infrastructure
- adapt models to specialist industry terminology
- reduce dependence on a single provider
There are trade-offs. Running AI in-house can demand significant computing resources, technical expertise and ongoing maintenance, and you'll need to weigh up performance, reliability, security and running costs. So sovereign AI is unlikely to replace every existing AI service. It's more likely to become a go-to option for the cases where control and independence matter most.
What does sovereign AI mean for AI jobs?
As AI adoption grows, so does the range of skills needed to build, deploy and manage these systems. Sovereign AI could open up opportunities in several areas, though how much extra hiring demand it creates is still unclear.
AI and machine learning engineers. Adapting language models takes skills in machine learning, model evaluation, optimisation and software engineering. People who can assess and tailor AI systems will be in demand as organisations try different models and deployment approaches.
AI infrastructure and cloud engineers. Models need computing power, storage, networking and reliable environments to run in. Engineers who know GPU infrastructure, distributed computing, cloud platforms and on-premises systems make AI practical for businesses.
Cybersecurity specialists. Securing AI means more than protecting the model. Teams have to think about access permissions, data exposure, vulnerabilities, integrations and the risks in AI-generated outputs. Cyber security professionals who understand AI infrastructure are likely to be increasingly relevant here.
AI governance and compliance specialists. Once AI is part of business-critical processes, someone has to assess risk, set policy, document systems and keep implementation responsible. People who understand both the technology and the regulation help bridge the gap between building and adopting.
None of these roles are new, but sovereign AI gives employers another reason to think about who will build, run and oversee these systems. For tech professionals, it's worth watching how AI infrastructure, governance and deployment skills develop alongside the models themselves.
What happens next for sovereign AI?
Kolibri shows that European companies are still building alternatives in a market dominated by large international providers. Its impact will depend on how it performs in real-world use, how easy it is to deploy, and whether it balances capability, cost and control well enough.
Competition won't ease off either. Open-weight models from developers around the world keep improving, so European projects will have to prove themselves on performance, usability and practical business value.
One thing to watch is ownership. Aleph Alpha has signed an agreement to combine with Canadian company Cohere, which still needs regulatory approval and is expected to close later this year. The combined business will operate as Cohere, with offices in Berlin and Toronto and Heidelberg kept as a research centre. It's a reminder that sovereignty depends on who owns the vendor as well as where the model was built. Kolibri's open weights stay usable whoever owns Aleph Alpha.
Sovereign AI comes down to giving organisations more choice over the technology they depend on. As AI becomes part of everyday operations, questions about ownership, infrastructure, data control and specialist skills will matter more. For employers and tech professionals, understanding them now will help with investment, workforce planning and career decisions.
At Priority Tech, we work across the Digital, Tech and Cyber markets as a specialist tech recruitment agency, connecting businesses with the talent they need to grow. As AI evolves, keeping an eye on developments like sovereign AI helps you work out which skills you'll need next.
Building a tech team? Get in touch with Priority Tech to talk through your hiring needs and find the right people for the job.
