The release of another AI model has already generated widespread media coverage. What compels me to add my two cents now? I think we are experiencing a truly disruptive moment this time, not the kind that clever “techbros” desire. This disruption is bringing down business models, especially those of Western tech companies, before they’ve properly taken off. In my opinion, this hasn’t been sufficiently appreciated in the media yet.
Whose Business Model?
European corporations are already far behind when it comes to developments in IT and particularly AI. So it can only refer to US corporations. In the US, with the change in government, a development strategy has been announced to strongly promote AI development and to simplify and reduce the cost of generating the energy needed for it. Stock prices, especially Nvidia’s, soared to new heights as sustained strong sales were anticipated. Furthermore, the technology gained is kept protected and under lock and key for as long as possible, to milk consumers over an extended period and protect the profit interests of the inventors.
What’s Changing Now
With the release of Deepseek, stock prices experienced significant losses. The business model for the coming decade has been destroyed before value creation could even begin. Deepseek, as every reader could gather from the media, achieves comparable complexity to the leading models (including ChatGPT) with only about 5% of the energy and computing power. The models can be trained cheaply, they can be operated cheaply, they require significantly fewer GPU chips than previously thought, and above all, it becomes drastically cheaper for end users. Additionally, the model is now not locked away but open source and accessible to everyone. Yes, we average consumers can now use it, won’t be milked to death, and for every small business owner, this is an enormous gain.
Deepseek will still generate enough profit, for example, by providers offering access to computing power with the colossal large model, which according to experts now overshadows ChatGPT. But it’s already so affordable – I would say with five euros you can get by for a few months.
This is no secret knowledge; you can view the model here. You can’t run such a huge model on a home computer, but a smaller version, which is still very powerful for most purposes, can be easily used by employing a tool like Ollama.
I think the implications for the future are still being underestimated. There is currently an attempt to fight back with reporting about an alleged data leak from the developer, somehow trying to undermine trust in the developers. For a project that is already open source and has already positioned itself at the top of the most used AI models, this will have little impact.