Summary#
Handelsblatt examines how the shift by major model providers to usage-based billing turns tokens into a resource that companies must actively manage, much like labor, energy, and capital. The trigger, among others, is GitHub Copilot moving to token-based billing. The article shows how companies are responding to the cost pressure through budgets, model selection, and new platforms.
My Quotes#
On how the cost logic changes for users:
“As a user, I then have to ask myself: is this request worth 100 dollars to me?”
My expectation is that AI costs will rise and that the large models from Anthropic or OpenAI will likely only be used for complex requests in the future. For specialized tasks, smaller models can be used instead, alongside open-source alternatives that are significantly cheaper to run.
On the importance of local models, for both cost and data-protection reasons: they run directly on a company’s own hardware, on laptops, smartphones, or in the company’s own data center. A local model first classifies a request by confidentiality and, depending on the result, forwards it to a large model in the cloud, transmits it anonymized, or processes it entirely on local hardware.
On the vision for Zühlke:
“For Zühlke, I envision offering our employees a platform that makes this choice automatically.”
And a forecast: if the model providers raise prices sharply, it could one day pay off again to hire junior talent instead of relying on AI.