If the value of AI ultimately comes from the combination of data and computing power, why are these resources still controlled by a few platforms?

This is another question that arose as I continued to explore @dgrid_ai. The core contradiction in the current AI ecosystem is not just about model capabilities, but about resource allocation rights. Whoever controls the computing power and the entry points determines the boundaries of applications. What DGrid attempts to break is precisely this structure.
Its idea is to split inference requests and distribute them to different nodes for execution. Nodes provide computing power, users initiate requests, and the system schedules and prices based on task complexity. If this process can operate stably, it essentially builds an open marketplace for computing resources, allowing resources to flow rather than be locked in.
On a deeper level, this structure will change developers’ choice pathways. In the past, you had to bind to a specific cloud platform or AI service provider, but now, in theory, you can directly connect to the network and choose the optimal node based on price and performance. This freedom will make the entire ecosystem more resilient.
But obvious practical issues remain. The stability and consistency of distributed inference are major challenges faced by all decentralized AI projects. Especially in high-concurrency scenarios, latency and result quality will directly impact user experience. If these issues aren’t resolved, users will still revert to centralized services.
In the long run, DGrid’s direction is correct; it aims to turn AI from a service into a market. But this process won’t happen overnight. It will take time to verify whether the network truly has the capacity to replace existing solutions. If successful, it will change not just the technical architecture but also the distribution of power across the entire AI industry.