Are we underestimating AI revenue by looking at software spending instead of labor costs?

Lately, there has been a lot of discussion about the AI bubble, particularly whether OpenAI, Anthropic, and other companies can generate enough revenue to justify the enormous investments being made. But let's look at it this way. Apparently, there are roughly 1 billion white-collar workers worldwide. If companies eventually spend an average of $200/month per employee on AI, that's $2.4 trillion in potential annual B2B spending. This sounds like a ridiculous amount of revenue, but consider this. I think an average US white-collar employee costs a company around $10,000/month, including benefits. If AI allows a company to reduce its workforce by just 2%, the savings alone could theoretically fund $200/month in AI spending for everyone else. And that's without considering other productivity gains. And if you are a company, would you rather have 100% of employees with $0 spent on LLMs or 98% of employees with each employee having a $200/month subscription to Claude/GPT/etc.? Now, obviously, wages are much lower in many other countries. There are also other factors such as cheaper open-weight models, self-hosting, and competition that could dramatically reduce how much revenue frontier