In January 2025, a Chinese startup called DeepSeek released an AI model that reportedly cost a fraction of what American firms were spending.
Markets didn’t take to the news well… and for about 2 weeks, it wiped hundreds of billions of dollars off the market caps of Nvidia and other AI-linked stocks.
Last week, something similar happened again…
Two new Chinese AI models, Kimi K3 from Moonshot AI and Qwen 3.8 Max from Alibaba, were released within days of each other.
Both of these are "open weight," meaning anyone can download them or run them on their own servers.
More importantly, you can customize these models for a fraction of what it costs to use the big American alternatives.
And once again, some tech and AI stocks fell hard on the news.
Needless to say, this has OpenAI and Anthropic exces worried.
OpenAI has gone the red scare route… warning of "full AI communism”
Anthropic's top brass are hitting the cybersecurity angle hard.
But I don’t see it that way…
This looks more to me like regulatory capture than anything else. This is where incumbents attempt to use government regulation to box out alternatives…
But here’s the part almost nobody is writing about.
None of this actually threatens companies like Nvidia.
You see, there are two very different kinds of businesses in the AI world right now.
The first kind is what I'd call a winner-takes-all…
Companies like OpenAI and Anthropic spend enormous sums of money training AI models, then try to make that money back by charging you to use them.
Their entire business depends on being better than the alternatives (both free and paid)
So if a Chinese lab releases a model that's almost as good and costs nothing to run, them OpenAI’s entire reason for charging you money starts to look shaky.
The second kind of AI business are my favorites…
These are the toll booths…
Companies that get paid regardless of who wins the AI race, because every single AI model… American or Chinese… open or closed, has to pass through them to exist at all.
Nvidia makes the chips that train and run virtually every serious AI model on Earth, including the Chinese ones the article describes.
Taiwan Semiconductor is the factory that physically manufactures those chips.
And then ASML makes the one machine in the world capable of etching the tiny circuits onto them, a machine so specialized that no other company can currently build a competitor.
Meaning that if the Chinese models win… they were likely trained using Nvidia chips, manufactured by TSMC, using ASML's machines. Same story if OpenAI or Anthropic wins.
And every lab mentioned in this story, American or Chinese, is racing to build bigger, not smaller. Cheaper AI models tend to mean more usage, not less, because more people and companies can afford to build with them. More usage means more demand for chips.
This is the same reason that during the California Gold Rush, most of the people who struck it rich weren't the miners panning for gold. They were the merchants selling picks, shovels, and blue jeans to every single miner, regardless of who found gold and who didn't.
I don't have a dog in the OpenAI-versus-DeepSeek fight, and it isn't obvious yet who wins it.
What is becoming obvious is that the fight itself has to run through a very small number of chip and equipment makers, no matter who's throwing the punches.
Oliver