ChainCatcher reported that an analysis of AI semiconductor and infrastructure cycles says demand for AI infrastructure may persist longer than it did during the internet era. The argument centers on two exponential dimensions — user penetration and tokens consumed per person — whose combined effect, converted into token demand, materially raises the ceiling for infrastructure needs.
The analysis says user penetration has already exceeded 50%, and that the larger part of future growth is more likely to come from per-capita token consumption, which it describes as still being at an early stage. It puts median AI spending by U.S. enterprises at about $12 per person per month, and says that over the long run this figure could approach roughly 10% of white-collar wages, or about $1,000 per month, leaving nearly two orders of magnitude of room in between.
Higher inference costs point to heavier infrastructure intensity
The note contrasts AI with the internet model of flatter subscription revenue and very low marginal hardware consumption. In AI, it says, inference remains expensive, which lifts the marginal cost of each visit and requires a higher level of infrastructure intensity.
Broader programming seen as the next ARR driver
After developer coding, the next wave of annual recurring revenue growth is still expected to come mainly from what the analysis calls broader programming. In that model, non-programmers use programming infrastructure to complete non-programming tasks across industries, while programming becomes the default execution core for agents. Tasks tied to broader programming account for about 60% to 70% of ARR, according to the analysis.
OpenAI and Anthropic data cited in the note
As of June 2026, Codex accounted for 64% of the combined output tokens of Codex and ChatGPT among OpenAI enterprise customers, the analysis says. Since February, weekly active growth for Codex in legal, sales recruitment, marketing and other verticals has been far higher than in engineering.
The note also says that at Anthropic, narrow development and software contribute about 40% of revenue, while financial services and insurance make up more than 20%. Legal, life sciences and retail each represent meaningful shares as well.
The analysis concludes that the user-side logic for token demand growth remains intact, while capital providers and beneficiaries are highly overlapping.