Busy morning, but I was struck by this comment in OilPrice.com that repeated an observation from a Financial Times piece:

“You may not know it, but AI work is measured in tokens. According to the Financial Times, as of recently, Chinese AI groups have overtaken American ones in sales of tokens. Why? For one, the Chinese models appear to be more efficient, and they get good electricity rates, according to the authoritative peach colored paper. As a result, Chinese models charge $2-3 per million output tokens vs a $15 price charged by US firms.”

I remember back in the days of the Internet bubble, people began to make fun of the prospectuses that companies were putting forward for IPOs. In many cases, these were being written by 20-year-olds who had no clue what they were doing. Often, they would still manage to raise hundreds of millions from their IPO.

I recall the Wall Street Journal writing up one of these perspectives early in 2000, not long before the bubble began to deflate. The prospectus described the product the company looked to produce, and then was honest enough to say that they didn’t know how they would make a profit. They still raised over $100 million in their IPO.

I can hardly claim any expertise on AI, but if people can buy the same AI power from Chinese companies for $2-3 that OpenAI and the rest are charging $15 for, it is hard for me to see how the U.S. behemoths can expect to make the big bucks that would justify their share price. 

I’m sure there is some home-country advantage. US firms will look to buy from the leading US companies, but there has to be a limit to this. Will they pay 400-500 percent more just to “Buy American?” Maybe they hope Trump will use his ugly stare to intimidate US companies into buying their overpriced product, but that doesn’t seem like a good long-term strategy. 

Seems to me, we might be better off with the company that didn’t know how it would make a profit. At least they were honest.