General Discussion

He is shredding AI on his blog and YouTube.

1) it is not intelligent. Current AI is a really good search engine, but it is not intelligent. It is not thinking. This limits its productive use.

2) it is insanely expensive to run. The amount of compute power needed to run a simple “dialogue” with the AI engine can eat up a monthly subscription in about an hour, depending on the complexity.

3) The entire supply chain is based on absurdly huge demand projections that would make AI 2-3x the current software industry. Why would people buy that much given its technological limitations?

4) Nvidia and others are stuffing their supply chains. They have already shipped more GPU’s then could be installed this year.

5) The data centers (think Google Cloud, Microsoft Azure, Amazon Web Services, Oracle) are buying the chips/GPUs, and many data centers have been announced but haven’t been started. A data center’s profitability is based on massive government subsidies, and there has been a bipartisan revolt against them by people. There needs to be a lot more data centers for AI to achieve profitability, and building has slowed down.

6) Circular deals - GPU buying has slowed, so Nvidia and others are paying OpenAI and Anthropic to buy Nvidia chips by buying stock in OpenAI and Anthropic. Amazon and Microsoft are doing similar with their data centers. This is a really bad sign, because it points out that the AI software companies are nowhere close to cash flow positive, but VCs are not willing to fill the breach. The vast majority of investors funding AI are able to treat that investment as revenue, thereby inflating their own stock price.

This is about a third of the problems with AI economics.

If any part of this slows down, the entire ecosystem, which has been carrying the stock market since 2022, starts to have serious problems.