Hey, guess what? China managed to do a cheaper version of an American product.

All joking aside, by now you have surely learned that DeepSeek is an AI model produced by a Chinese company that appears at least competitive with the best models produced by American companies. They claim it was trained on just $5 million, which, according to experts we trust, is probably untrue and is probably a lie meant to disguise black market chip acquisition. What is not untrue, because anyone with mild computing skills can validate it, is that their model is extremely cheap to run.

And we mean it: you can, right now, order the right computer equipment online, which will run you about $6,000, assemble it, then download their open source model and its so-called weights (all 800 GB of them) and, after installing Linux on your computer, install the model, and you will have your own personal AI that will be nearly as good as ChatGPT—which, for its part, runs on massive data centers that cost billions of dollars to build and operate. With prudence and good online tutorials, a non-technical person can do this. Heck, your correspondent used to do it as a teenager for pocket money (minus the artificial intelligence bit).

The implications are profound.

First, for the industry itself, and what it was planning to do. If it turns out that you need much fewer computer chips than we thought to run AI, that involves significant changes to the industry, but also: what happens to the recent tech-backed movement to open up more nuclear and other sources of energy.

Second, for AI censorship. If advanced AI models can be run out of a machine that can be built for four figures and run at home, then there is simply no hope for a far-left (or far-right?) NGO complex to control AI. Our future is a distributed cyberpunk future where highly advanced cyberintelligences, either under the control of individuals or groups, or perhaps their own control, will be active in every sphere of our lives.

Third, for national security. Marc Andreessen called this "AI's Sputnik moment." And it does seem to be that way. There are direct implications. Reports suggest that OpenAI have determined that DeepSeek may have been created through illegally copying their work. Others have pointed to the fact that a growing, and implausibly large, portion of nVidia's revenues, come from "Singapore," suggesting that most of these chips are really headed for Taiwan. If we are in an AI race with China, plugging those leaks becomes all the more urgent.

But there are defensive and offensive implications. If this is AI's Sputnik moment, then it means that America must accelerate. It must remove brakes on the development of AI, particularly by startups and open source projects. It must make sure AI entrepreneurs (or anyone else, but particularly AI entrepreneurs) are not put under the yoke of blue-haired NGO commissars. And we need to put regulatory rules in place so that the data centers get built and then get the new energy they need. It's more urgent now than ever.