China's AI Costs 7.5x Less Than US: Should You Worry?

Here is a number that should make every $NVDA bull pause for a second: $2.33 vs. $0.31. That is what a developer paid running the same coding task on Claude Opus (US) versus GLM, a Chinese open-source model. Same output quality, same five-and-a-half minutes of runtime, and a 7.5x price difference. Andrei Jikh laid this out in a recent video and it is the kind of data point that quietly rearranges your portfolio thesis.

The reflex response is to say the US still leads on raw capability, and that is true. But Jikh makes a sharper point: most businesses do not need the smartest AI. They need affordable AI for the unglamorous 90% of use cases, think customer service emails, document summarization, internal chatbots. For those tasks, China's models are not just competitive, they are a structurally cheaper alternative that is already available right now.

Meanwhile, a separate Andrei Jikh video takes a wrecking ball to the broader AI business model. Traditional software has near-zero marginal cost per new user, which is why it prints money. AI breaks that entirely because every single query burns electricity and compute. As Jikh puts it, it is like a restaurant that loses money on every meal served. The "just scale it" answer does not fix a model where costs scale with revenue. That is a real problem and not enough people are pricing it in.

So what does this mean for the actual trades? $PLTR just staged a beast-mode earnings run and is up nearly 40% after boosting its full-year forecast, with management describing commercial AI demand as essentially uncapped. Palantir is not selling raw AI compute, it is selling the software layer on top of it, which is a very different and more defensible business model. That distinction matters a lot in a world where underlying AI gets cheaper every quarter.

The bear case for US AI infrastructure names is not that China builds a better model. It is that China builds a good enough model at a fraction of the cost, and the enterprise market rationally defects to cheaper. The bull case is that the US still controls the top of the capability stack, and regulation, trust, and data sovereignty keep American models dominant for high-stakes use cases. Both things can be true simultaneously, which is exactly what makes this trade complicated.

The scariest part of the 7.5x price gap is not where it is today. It is the direction of travel.