How can local AI models be made more independent – and when are they really worth it, INGOLF TEETZ?
In this ASK INGOLF podcast episode, I speak to Ingolf Teetz, Chief Innovation Officer at EMBRACE, about a topic that is becoming increasingly exciting for businesses and private users: local AI models. Instead of sending data to OpenAI, Anthropic or other cloud providers, local models run directly on the user’s own hardware. This can mean greater data privacy, more control and lower costs in the long term – provided the use case and the required computing power are suitable.
One statement by Ingolf really stuck with me:
“If my token usage across the whole company is under 10,000 to 15,000 euros, I don’t really need to give it a second thought – unless I have a compelling reason.”
Among other things, we discussed the following topics:
🔐 Your data stays with you
Whether it’s contracts, invoices, applicant data, photos or other sensitive information: with local models, you can process data without transferring it to external cloud providers – an exciting approach, particularly with regard to data protection and the GDPR.
💸 When does on-premises AI really pay off?
No ongoing token costs sounds attractive at first – but powerful hardware is also expensive. Ingolf explains why scaling is the decisive factor and why companies should consider cloud costs, hardware, operations and security as a whole.
🤖 Your very own AI ‘house spirit’ is getting closer
Models such as Qwen are now becoming surprisingly powerful even on consumer hardware. Combined with local AI agents, they can analyse documents, help with programming, organise files or even control the smart home – without the cloud. However, when it comes to large-scale software projects and demanding office applications, frontier models currently still have the edge.
🔗 Further links
👉 Ingolf Teetz LinkedIn
👉 ASK INGOLF Spotify playlist