One way to make aggregate ARR estimates for AI is to use this simple fact:

All OpenAI+Anthropic revenues must *eventually* come from actual human population who is directly or indirectly consuming all of the tokens.

It could be through subs or through AI powered apps but, regardless, costs eventually must be transferred to the end users.

Generously assuming 30% of the US population directly or indirectly pays $100/mo in steady state, US ARR is probably bounded to ~$120B. Typical global revenues are 50% of US revenues so global ARR is possibly bounded to ~180B.

Given incomes are not increasing and unemployment is currently rising, consumers have to sacrifice something else to make space for this AI cost from their incomes.

In short term, businesses may end up spending 2X-3X and subsidize for discovering the consumer demand curve. But they will eventually have to match ROI with consumer spend.

One way ARR continues to rise is if consumer income expands. However, there has been a leg between tech revolution and consumer income expansion.

One wild card for AI which didn’t exist in other tech revolutions is exponential decline of inference cost. This can allow to accelerate consumer income expansion faster than usual 10-20 years of leg.

A possible scenario is this:

Initial rise in unemployment causes huge uptick in self-employment with solo founder companies powered by AI agents. This trend can unchain a lot of people from employment where they get fixed wages instead of capital equal to the value they generated for the business (for ex, Starbucks barista adds huge value to the company but they get ~minimum wage while the value they generate by making over 100 cups of coffee every day simply flows back to the capital owners).

In long term AI could be a huge force in reducing income disparity and bringing sort of communist flare where workers accurs capital same as the value they generate as opposed to receive only the “market” wages from the capital owners. Notice that this would be only possible because reward vs risk proportion is significantly reduced by AI (i.e. almost no capital needed to start solo AI agent powered business).

An ultimate observation here is that the reward per risk determines the disparity in wealth. The ratio of value that goes to shareholders vs the value generated by the worker is the risk premium. AI makes this risk premium tantalizingly small.

While AI will cause massive unemployment in short term, those employees will tend to turn to using their knowledge and skills to start similar, complimentary or even competing businesses. Instead of creating value for the shareholders, they will create value for themselves.

Entrepreneurship was hard because getting funding was hard, building team was hard, building right product was hard, and, perhaps most importantly, marketing is hard. All this friction will start melting away as AI agent competes with the best engineers, best designers, best marketers while eliminating the need for external funding. These will come at the expense of large inefficient businesses that are unable to change, infested with burocracy and internal politics. They will get divied up into many small powerful businesses run by solo founders.

This is the ultimate Universal High Income that self-emerges in the economy due to AI in long term. In that future, potential ARR for tokens is approximately proportional to rest of the information economy.