What GTC Berlin Means for AI You Control

Who controls the computer that answers your questions? Who can access the data you give it? What keeps working when a service changes?

Those questions connect the sovereign AI discussion at NVIDIA GTC Berlin to a much smaller setting: your desk.

The conference took place on October 21–22, 2026. NVIDIA’s published program included a sovereign and enterprise AI discussion with leaders from Deutsche Telekom, Mistral AI, and Dassault Systèmes. HPE’s Berlin sessions explored control over AI data, models, infrastructure, and operations.

The practical question for an individual or small business is how to put that control to work.

What sovereignty adds to the AI conversation

Performance tells you how fast a model can answer. Control tells you where that answer is produced, who can see the input, and who decides how the system operates.

At national scale, NVIDIA describes AI strategies that bring together domestic infrastructure, data, skills, models, and institutions. A local application addresses a smaller part of that picture. It can put inference—the process that produces an answer—on a computer you control.

Our takeaway from Berlin is that these questions deserve a place in everyday purchasing decisions, too. You should be able to understand an AI product’s data path before trusting it with personal notes or business information.

Three questions to bring back to your desk

Where does the model run?

An app on your laptop can still send its work to a hosted model. Check the location of the model itself. For local inference, your device runs the model and produces the answer.

That choice has practical limits. The model must fit the available memory, support your operating system, and handle your task well enough. A small local model can be useful without matching every capability of a large hosted system.

What else happens to the conversation?

Inference and storage are separate questions. A model can run locally while the application synchronizes conversation history between devices. Downloads, updates, backups, and external integrations also have their own network requirements.

Look for an explanation of each path. A statement about local processing should not have to carry the entire privacy story.

What do you keep control of?

Ask who holds the keys, how you recover your data, and which functions require an ongoing service. Include model licenses and software updates in that review. Owning the computer is a useful starting point; the surrounding choices still matter.

Where Merka fits

Merka Vault™ brings local chat and a personal cloud into one product. On a supported device, a built-in provider or downloaded model can produce answers locally. The app shows the selected model and where it runs. Chat history is saved on the device and can sync across your devices when chat sync is active.

Your files and photos have a home on computers you control. You keep your keys and recovery material. Local use is free, and dedicated hardware and managed services are optional choices.

That is the part of the sovereignty discussion we can make concrete: start with a computer you already own and understand what it does with your data. Merka’s local chat is available on supported devices. The calendar and email assistant illustrated on our marketing website shows a broader direction; it is not evidence of automatic access to those accounts.

Try it with a small task

Start with a short, invented note. Ask the model to turn it into a checklist. Check where the answer runs, review the result, and learn the app’s storage and sync behavior before moving to sensitive material.

Our guide to sovereign AI for individuals and small businesses gives you a practical set of questions. The local AI on Mac walkthrough takes you through the controls and an offline check.

Download Merka for Mac or explore local AI and your own cloud.

This article is Cosmic Rocks’ interpretation of the published conference material. It does not imply an NVIDIA partnership or certification.