Why the order matters, and what each layer actually does.
AI doesn't run on one industry. It runs on a stack, and each layer depends on the one beneath it. Energy powers the data centres. Data centres house and cool the machines. Semiconductors do the computing. Networking lashes those chips into servers and wires the servers into one machine. Cloud rents the result out by the hour instead of selling it. Data and analytics stores and moves what the models are trained on. The models turn that data into answers. Applications package those answers into software people actually buy.
That dependency is the whole point. It means capital doesn't move between these groups at random — it moves along a chain, and the chain has a direction.
A conventional classification puts NVIDIA and Arista in the same bucket: technology, semiconductors and equipment. Accurate, and useless for the question most people are actually asking. It won't tell you that one sells the compute a data centre commits to first and the other sells the interconnect it buys once the compute is racked.
Sector membership describes what a company is. Layer position describes what it depends on and what depends on it. For a supply chain being built out at speed, the second is the more useful fact.
L1 is not worse than L8. The numbering is structural, not a quality judgement — L1 is the foundation, L8 is what the end user touches. A power company and an AI SaaS business are in the same stack, doing different jobs, and both can be the right thing to own at different points in a build cycle.