It was just last week that the kings of the U.S. Artificial Intelligence (AI) industry, the leaders in the world, gathered in a press conference with President Trump to talk about astonishing budgetary numbers (half a trillion!) and massive industrial transformation.
They were: Larry Ellison, the executive chairman of Oracle, Masayoshi Son, the CEO of SoftBank, and Sam Altman, the CEO of OpenAI, all rallying around the expensive miracle chip from Nvidia, the stock price of which reflects the investment mania now fueling the industry.
The project they were pitching was called Stargate, and Ellison made a quick for-the-internet speech about it. In the course of a two-minute soundbite, he somehow combined AI, mRNA, and a cancer cure in one big stew of far-flung promises. That speech rattled many close observers who wondered whether the new president was really going to buy what these people were selling.
Elon Musk, who has a personal and public feud with Sam Altman, quickly took to X and said that these companies are nowhere near creditworthy enough to raise this kind of money. One presumes that this helped undermine Stargate with the new Trump administration but there was not enough information forthcoming to know either way.
Then a strange thing happened. A Chinese company called DeepSeek, seemingly out of nowhere, had released an app (the day before the meeting) directly into the mainline stores—within a week it became the number one download for large language models. The company produced all its documentation and made the whole of its code open source.
The efficiencies relative to the status quo are mind-blowing. They did for less than a tenth of the resources that OpenAI has spent and plans to spend, and did so in a fraction of the time, in a way that requires not acres of data centers but simple laptops.
The code is open for all to see and includes a number of simplifications. It uses database calls far more efficiently for the same results. It deploys chips that are older and evade U.S. bans on chip exports, which seemingly makes the point that an industrial policy by the United States to bolster its leadership in this field is easily circumvented.
To be sure, as a program, DeepSeek can run more efficiently so that even small companies can compete in the AI world which has so far been dominated by corporate behemoths. The point is not to celebrate this new company over all existing ones but merely to highlight that foundational innovation in this sector is far from over; it’s just begun.
To be sure, DeepSeek is massively censored by the Chinese Communist Party (CCP). Ask any sensitive question about politics and the answer comes back that this question is beyond the scope. That is perfectly predictable and wholly expected. This is how the regime works. That said, the code can be copied and run on an uncensored basis, and that will surely happen in the coming weeks and months.
Where does that leave the tech bros and Nvidia in particular? The stock price as I type is showing the answer. It’s too soon to say that the AI bubble is popping but we’ve been here before and know the trajectory.
The best analogy here is the dotcom boom-bust of 1998-2001. There was a wild frenzy in the sector that exaggerated the valuation of some companies that went bust. Pets.com is a good example of that. The company’s stock valuation soared until it crashed and everyone was laid off. The idea merely needed refinement, however, despite appearances, and today that domain redirects to a company that has made a go of selling pet food.
The point is a more general one. Stock market manias can cause huge industrial disasters when they are exposed to competitive market environments. But such shaking out serves a valuable market function too.
The AI industry in the United States had already become pretty heady pretty quickly, and the stock prices of the companies involved reflected this. Nvidia’s stock fell $400 billion in valuation over the course of half a day of trading following the release of DeepSeek, which is close to being the same amount of resources that Stargate was crowing about spending on AI-based mRNA cancer cures. Markets do have a way of bringing reality to bear.
The question for the U.S. industry now and in the future is what to do about international competition for its functionality and coding. There is the option of creating a walled garden of a handful of big providers and insulating them from innovations of all sorts. That does not seem like a viable long-term solution in a digital world.
This might have worked in the late 19th century with the U.S. steel industry. It’s hard to see how the same approach can work in today’s world, however. Technology moves too quickly to make that model operational.
The only real solution from the U.S. point of view is dramatic deregulation, impressive innovation, competition at home and abroad, and fast adaptability to change. Tariffs can work as a band-aid approach for issues of physical manufacturing but have extremely limited use in a digital world of international competition.
As to AI in general, it is glorious for some functions and useless for others. It is great at assembling and deploying existing information—which is enough of an achievement to earn a place in the history of industrial innovation—but useless for genuinely creative thought. Yes, it threatens and redirects existing labor resources but in no way replaces the need for rational judgment by human beings. As with all tools, it makes existing work more efficient but changes nothing about the core human predicament or the value of the human mind in solving it.
The real problem with AI is that it has given rise to bleary-eyed speculation about some coming merging of the human and machine mind, as imagined in the fantasies of the transhumanist project. This vision constitutes the real threat insofar as governments and their industrial partners imagine a world in which, for example, the human immune system is replaced by fancy injections of DNA-altering drugs.
The results of such an experiment will be vast waste and vast injury as consequences of humanistic hubris. I have no doubt that some moguls in the tech space believe this is possible but it simply is not.
What we are seeing take place within the AI industry is consistent with a very long history of mania and industrial concentration, followed by competition and disruption, settling back over time into normalcy. This same trajectory happened with websites, databases, electricity, flight, steel and metallurgy, internal combustion, railroads, and onward back into time. At each stage, there is population panic about the meaning of it all, accompanied by stock frenzies in each case.
What was most delightful about the events of the past several days is that they serve to remind us of the absurd presumptions of the masters of the universe and how they so easily set themselves to be smited by new and unexpected forces from the outside.
All of which should remind us of the great principle that no one industry or government-created oligopoly will ever finally rule the world, a fact for which we should all be deeply grateful.
