It's quite interesting. Ever since I discovered that aethir was very profitable last time, I've looked into some other GPU computing power rental projects and found that those in the "treasure trove" of GPU computing power rentals are also very profitable. For example, @SpheronFDN's ARR this year exceeds 10 million USD (estimated). In fact, with the explosion of AI agents and applications, selling computing power is like being a water seller on the gold rush road of the new era. The scarcity of GPU computing power also allows computing power rental/sale/trading matching platforms to charge a higher premium. However, in any profitable industry, more and more players will gradually enter, leading to increased competition. There are more and more decentralized GPU computing power platforms, such as io, Akash, aethir, hyperbolic, salad, render network, vast, and so on. So where is the key point for longevity in this track? One is the ability to obtain cheap and stable GPUs, which essentially means strong supply capability. On the other hand, there needs to be a focus on vertical aspects, such as providing customized tools and environments for AI model training and data processing. Alternatively, efficient rendering queue management for image and video rendering could be provided. I see that Spheron has launched KlippyAI (focusing on AI video) and Skynet AI (for building and deploying autonomous AI agents and workflows), which may be their attempt to differentiate in the vertical space. You can check out this demo.