Deepseek’s silent revolution

The launch of the model Deepseek-R1 It has marked an inflection point in the AI ​​industry, demonstrating that it is possible to develop high -performance language models at a fraction of the cost of current leaders … and in an open source.

Why is it important. This development questions the business model of large technological ones, which have invested tens of billions of dollars in proprietary infrastructure.

They have done it under the premise that LLMS They are in themselves a sustainable competitive advantage. Deepseek (and not only her, although yesterday all the holders will take Calls) has burst that premise.

The alarm signal. Marc Benioff, CEO of Salesforce, He said it clear:

“The models and interface have become commodities. True value lies in data and metadata.”

This vision suggests that the future of AI is not in the models itself, but in its practical application and in the data that feed them.

This process reminds a lot of what happened with the search engines at the beginning of the century. Which was once considered high -value -owned technology – the search algorithm – became Commodity with the arrival of open source solutions such as Lucene / elasticsearch.

The true value was not in the search engine per sebut in what Google built on him: contextual advertising, user data, integrated services, user experience.

Today any company can implement the search, but only those that have managed to build a valuable ecosystem around them have prospered.

Deepen. Deepseek’s efficiency is no accident. Forced by the restrictions of access to advanced chips, the researchers had to innovate in the model’s own architecture, with techniques such as latent care Multihead (MLA), which allow to obtain better results with less resources.

Deepseek marks the beginning of a new phase of the AI ​​in which language models are a basic utility on which to build added value applications.

The big question. Where will the value in a world where the LLMS are Commodities? Everything indicates that differentiation will be in the ability to apply these models to specific problems. And in the quality of the owner data, in addition to the user experience.

The earthquake that Depseek has unleashed is the beginning of an era in which the competitive advantage will not reside in having the most powerful model, but in knowing how to apply it in the most effective way to solve real problems.

In Xataka | Deepseek is the fashion model. The problem is that nobody knows very well what you are doing with our data

Outstanding image | Xataka with Mockuuuups Studio