Key insights
- The author expresses skepticism about AI's current business model, arguing that despite productivity gains, the cost of AI tools is high and may not be sustainable for widespread adoption. The article highlights that AI currently costs as much as or more than human interns for comparable tasks, with limited adaptability and the need for human oversight. Rising costs in semiconductors, infrastructure, utilities, and R&D are cited as key drivers, suggesting that AI valuations may be overstretched in the medium term.
First of all, forget the valuation, if AI is a bubble or not. I am also biased against AI being a sustainable cashflow machine in the medium term, 3-5years, to justify the valuations. The post is mainly my personal opinion based on my personal experience and experience at workplace in programming and research tasks.
Now, let's talk about how will they bring in cash.
It improves the productivity -- yes, agreed. I can do more 3x/4x more in the same time on specific tasks. But I have invested a lot of time in bringing the AI workflow to this point - from where it was 20%-30% productivity boost for me to a 3x/4x now. But, it still requires me to solve hard bugs to make the code usable. But, at what cost? I dont know how many of you are aware of the recent price changes in GitHub Copilot, which now costs more - on par with other western providers. Most people are using open source models or cheaper Chinese alternatives. Every new model is getting better, but also in the meantime more expensive.
If it improves productivity, either the total human output should skyrocket or should replace humans (which leads to other problems). Then it should cost as much or less than humans. At the moment, it costs me at least as much as an intern@hourly rate - but an intern can do more things - an intern can physically move - conduct experiments, have insights or own ideas. In other words, agentic workflows are nice for even complex and repeatable tasks, but cannot adapt to a new task or bring a new product idea.
In the tasks, I use the agents - as it stands now it costs as much as or more expensive than an intern or a junior. But the variety of tasks, to which a human can adapt themselves fast is much larger. So I assume that the cost has to come significantly down to justify the use case.
Let's think about what is composed in the cost.
-
semiconductor prices
-
infrastructure capitalisation / maintenance
-
Utility costs
-
R&D costs
Now, we justify that R&D is necessary and spend whatever we would need, how will the business model work forwards?
-
semiconductor - chips have to become cheaper for the same power or chipmakers will have to cut down their costs or margins. - both of them go against exponential growth trajectory, negatively affecting margins and limiting the earnings. 2. Infrastructure - money is already spent, must be factored in. The debt service comes due whether you subsidize or make money out of the infrastructure. The capex spend must also increase for a wider adoption to keep up with demand.
-
Utility costs - chips &/or models will have to become extremely efficient. However, with soaring energy prices, which is expected at least for the near to medium term - brings the question - how much efficient the chips and models will have to become to maintain the current costs?
So, what is the business model? those of you who are invested in the AI-drive, how do you see the business model working from here onwards.