Desde que el Ómnibus Digital de IA entró en vigor a finales de julio, la pregunta que más me han hecho relacionada con el AI Act desde las PYMEs con las que tratamos en @Kairos_Tek (gestorías, despachos o agencias de menos de veinte personas) llega con un tono de alivio que se nota claramente: "He leído que el AI Act se ha aplazado a diciembre de 2027, entonces ya no me toca hacer nada todavía, ¿verdad?". Y la respuesta suele ser siempre la misma: lo que se aplazó casi con total seguridad no era tuyo, y lo que sí entró en vigor el 2 de agosto no te pide lo que crees que te pide. El Ómnibus no suavizó el reglamento para las empresas pequeñas. Movió las obligaciones caras, que afectan a muy pocas empresas, y dejó intactas las baratas, que afectan a casi todas. Y la parte que se mantiene implica dos puntos de vista diferentes: el de quien fabrica el sistema y el de quien lo usa, que casi todo lo que he leído estas semanas mezcla en uno solo. Ahí es donde se está tomando mal la decisión, y si tienes una PYME, eso es lo que tienes que revisar (no te preocupes, pero ocúpate). Te lo cuento en esta nueva publicación de la newsletter "Entre gigafactorIA y gestorIA". Qué cambió con el AI Act, y cómo afecta a una PYME.

NVIDIA's $12.9 billion acquisition of Hugging Face may be far more significant than it first appears. It's not simply a matter of NVIDIA wanting to add another AI company to its portfolio. It's a strategic move about where the true center of gravity of the next phase of AI might lie. For the past few years, the market seemed relatively clear: the large commercial models of OpenAI, Anthropic, and Google defined the technological frontier and concentrated a significant portion of the value. But something has been changing in recent months. Open source models are rapidly closing the gap with commercial ones. They are becoming increasingly powerful, efficient, and, above all, allow companies and developers to adapt, deploy, and control them in ways that closed models don't always permit. And NVIDIA seems to have recognized something fundamental here. Hugging Face is not simply a repository of models; it has become one of the major layers of distribution and collaboration within the AI ​​ecosystem: models, datasets, tools, communities, and developers. Controlling that layer can be as strategic as controlling the hardware on which the AI ​​runs. The operation would also have a particularly interesting logic for NVIDIA: the more diverse and competitive the open model ecosystem, the less dependent the market will be on a small group of closed model vendors who, moreover, have incentives to develop their own chips and reduce their dependence on NVIDIA. And the data shows that the open ecosystem is no longer a marginal alternative. Hugging Face registered nearly 3 million public model repositories this summer, while NVIDIA and AMD are among the organizations that have published the most new models in 2026. The connection between open models and hardware strategy is becoming increasingly clear: a good model optimized for your infrastructure is also an extraordinarily effective way to sell hardware. An example of a recurring successful strategy, which has an entire chapter dedicated to it in the book "Good Strategy, Bad Strategy," discusses how NVIDIA went from being a GPU company for video games to building CUDA when the accelerated computing market was still much smaller. Then it turned that investment into a massive competitive advantage for AI. And now, having clearly dominated the infrastructure layer, it seems determined to strengthen its position in the software, modeling, and community layers as well. It's not content with just selling shovels during the gold rush. It wants to be on the roads, in the warehouses, and in the markets where that gold is distributed. Watch the market for the consequences of this move. Link to the article in the comments 👇

Europe just launched the AI Gigafactories call tender. Up to 7 facilities, €10B public, €20B private expected, 18 Member States on board. I lead software for the Spanish AI Gigafactory, so I am not a neutral observer here. But €30B is less than 5% the investment of combined 2026 capex for Amazon, Microsoft, Google and Meta. And you have to take into account also China... This is not about how much money we spend. This is about sovereign AI, who gets to use the compute we build: startups, SMEs, universities, public bodies. Under EU rules, on EU infrastructure. Different race. Worth reading it that way 👇