When I am talking with projects that are developing in AI, they are taking either the approach to use the full LLM, booking and renting an H100 but look into getting the LLM beast under control. The interesting part is, an LLM is not only capable to give information as an example about auto mechanics or health care, but also about Goethe, the latest trends on X, Bitcoin movement and history and so on. It means that you will need an highend GPU because the model is "too smart". But if you are going to create a specialized model, as an example only to check onchain data, LP transactions in combination with CT sentiment on X, you don't need all the stuff with Goethe or how to create a rocket, right? This is where @OpenledgerHQ can help with their infrastructure. It can help to create an SLM based from a general model, also recreated the process how LLM's works on GPUs so they reduce the needed resources and reducing the cost. I've still not seen the whole technical background and how the architecture is working in real life, but if Openledger is delivering what they are currently prepping, than it would be a core infrastructure for a lot of AI startups. The best? AI startups can even create a second revenue stream with Openledgers Data reward system.