Building the playroom for AI toddlers

Agent runtimes as the missing revenue engine for trillion-dollar data center capex

I recently wrote that the hyperscalers are spending trillions on AI data centers, and that they have no clear path to generating revenues sufficient to justify the enormous expense. Their assumptions are, I averred, not even wrong. But what if there is a path forward? What if we can build a revenue model which could justify trillions in capex spend?

Well, a new paper¹, written by David Silver and Richard Sutton², Welcome to the Era of Experience, calls for a new approach to advanced AI. Large language models, they claim, have hit their limit:

To progress significantly further, a new source of data is required. This data must be generated in a way that continually improves as the agent becomes stronger; any static procedure for synthetically generating data will quickly become outstripped. This can be achieved by allowing agents to learn continually from their own experience, i.e., data that is generated by the agent interacting with the environment. AI is at the cusp of a new period in which experience will become the dominant medium of improvement and ultimately dwarf the scale of human data used in today’s systems.

If we proceed as if the authors’ claims are correct, then we can construct a revenue model for the hyperscalers which justifies their capex spend, and that’s what the rest of this post is about.