AI arrives: From hype to controllable business infrastructure

On 25 August 2026, the talk at IHK Gießen-Friedberg was about more than the possibilities of artificial intelligence. The focus was instead on the question of how companies can concretely deploy AI today – and what it takes for individual applications to become a solid enterprise solution.

Carmine Squillace was on the ground for CCNet. His keynote and the discussions in the following barcamp were mainly about one point: enterprise AI has to be able to do more than produce good answers. It has to become controllable.

Don’t ask the internet. Ask your own company.

Many companies have meanwhile gained first experience with generative AI. The next step is, however, considerably more demanding. In day-to-day business, general knowledge from a language model is frequently not enough. What matters is your own information: internal documentation, process knowledge, guidelines, project experience, or the knowledge of long-serving employees.

An enterprise AI can connect these knowledge bases with one another and make them accessible through natural-language queries. This is not about simply replacing existing storage with a chatbot.

The real task is to build a controlled knowledge base:

• Preserve knowledge so that it remains available even when staff change.
• Make information easier to access and easier to find.
• Scale knowledge so that it is not only available to individual employees.
• Feed new insights back into the knowledge base in a structured way.

This is how AI is increasingly becoming a component of business infrastructure.

Sovereignty means more than data protection

A central point of the discussion was the concept of digital sovereignty. It quickly became clear: there is a difference between a general idea of “we keep control of our data” and a technically clearly defined implementation oriented towards security and authorisation concepts.

For companies, this raises concrete questions:

Where is our most valuable knowledge? Who is allowed to access it? And what would be possible if authorised employees could query this knowledge directly through an AI? This is exactly where enterprise solutions differ from freely available AI tools.

A controllable AI infrastructure has to take existing roles, access rights and information boundaries into account. Not every employee should automatically receive every piece of information – simply because it is technically present in a shared knowledge base.

Local AI also changes the cost logic

Alongside data protection and control, economic efficiency played an important role in the exchange. Many externally operated AI services incur costs that depend on usage, tokens or API calls. The more deeply AI is integrated into daily processes, the more relevant this variable cost structure can become. With a local or privately operated setup, this logic changes. An external per-use token or API bill can be eliminated. Instead, infrastructure, operation and energy become the decisive cost factors.

This does not automatically mean that local AI is always cheaper. It is an investment decision that has to fit the respective use case. In the scenario presented by CCNet, the calculated return on investment came to around 18 months.

Especially for enterprise applications that are used on a lasting basis, it is therefore worth comparing the ongoing, usage-dependent costs with your own infrastructure.

From AI experiment to concrete use case

The discussions of the evening also showed this: companies do not need as many AI projects as possible. They need the right ones.

A sensible starting point therefore does not begin with the question of which language model is currently achieving the highest benchmark scores. It begins with the company itself. Where does unnecessary searching effort arise today? Where does knowledge depend on individual people? Which information is needed again and again? Where could processes be accelerated by secure access to existing knowledge?

Only once the concrete application case is defined can it be sensibly decided which architecture, which models and which operating model are needed.

AI is arriving – this is where the real work begins

The evening at IHK Gießen-Friedberg showed that the discussion around artificial intelligence is changing. The question is increasingly no longer whether companies can deploy AI.

It is about how they deploy it so that it creates measurable benefit – without giving up control over data, authorisations and sensitive business knowledge. It is precisely at this point that the AI hype becomes an infrastructure decision.

CCNet supports companies with this, from the identification of suitable use cases through the architecture to the operation of controlled IT and AI infrastructures. The focus is on data control, authorisations, security and economically comprehensible operating models.

You would like to know which AI use case is actually suitable for your company? Get in touch with us. Together we will look at your existing knowledge bases and check where a local or private enterprise AI can create concrete added value.

Image source: AI Netzwerk DACH (https://www.ainetzwerk.de)