The AI industry talks a lot about openness.
Open-source models.
Open research.
Open innovation.
But the more time I spend looking at the ecosystem, the more I feel like AI is becoming concentrated rather than distributed.
Most users interact with intelligence through a small number of platforms. Most developers rely on a small number of providers. Even when models are open, the infrastructure needed to run them at scale often isn't.
That's not necessarily a problem today.
The question is whether it becomes one later.
Technology has a habit of looking decentralized at the application layer while becoming increasingly centralized underneath. We've seen versions of that story before.
That's one reason OpenGradient caught my attention.
Not because it's another AI project.
But because it's trying to approach intelligence as infrastructure instead of just software.
The idea sounds simple: create a network where models can be hosted, inference can be executed, and results can be verified across a decentralized system.
Whether that model can attract enough builders and demand is a completely different question.
Infrastructure doesn't become important because it's technically elegant.
It becomes important when people actually use it.
For now, that's the part I'm watching most closely.
Will the future of AI be defined by better models, or by the networks that make those models accessible?