The AI industry was rocked in recent days with the release of an open source AI model from Chinese start-up DeepSeek that can compete with the best AI models from U.S. companies despite purportedly costing just $6 million to train. While it's possible the claims about costs are exaggerated or flat-out untrue, the DeepSeek model appears to be the real deal.
While cheap, powerful AI models are a great thing for companies looking to deploy AI, it's potentially terrible news for Nvidia (NVDA -4.10%). The bull thesis for Nvidia, which dominates the market for powerful AI accelerators that are necessary to train the most advanced AI models, relies on the assumption that each successive generation of AI models will require more and more computational horsepower to train and run.
DeepSeek's breakthrough raises some serious questions. This uncertainty led to an epic plunge for Nvidia stock on Monday that wiped out hundreds of billions of dollars in market value.
Another shoe still must drop
A cheap AI model capable of trading blows with the best models from Open AI and Anthropic isn't enough on its own to scuttle Nvidia's growth story. Cheap and efficient AI models could spur more demand and ultimately drive sales of AI accelerators higher in the years ahead. Former Intel CEO Pat Gelsinger stated in a post on X that "Computing obeys the gas law. Making it dramatically cheaper will expand the market for it." Cheap AI models that don't require massive clusters of high-powered graphics processing units (GPUs) to train and run may not be a negative for Nvidia if it greatly expands the use of AI.
However, there's another problem lurking. Nvidia's multitrillion-dollar valuation rests on one more important assumption: AI models will continue to become more capable as more computing resources are thrown at them. This was certainly true in the early days of AI, but it may not be true for much longer.
AI companies have largely exhausted the data used to train AI models, and new AI models aren't leaping ahead in terms of performance anymore. The founders of Andreessen Horowitz noted late last year that improvements are slowing down, and AI models may be hitting a ceiling.
It's important to remember how large language models (LLMs) work. Ultimately, these models predict the next token in a stream of tokens. That's it. There's no reasoning going on, just the illusion of reasoning. While new techniques could unlock improved performance, it makes sense that there would be a limit to the capabilities of this class of AI model.
It's the combination of cheap AI models and this potential ceiling in capabilities that would be the death knell for Nvidia stock. If AI models stop improving meaningfully no matter how much computing power is thrown at them, and a top-tier model can be trained cheaply on second-rate hardware, that's the ballgame. Demand for Nvidia's expensive AI accelerators would likely fall off a cliff as the AI bubble bursts.
Even under this scenario, AI will still be a useful and potentially game-changing technology for companies that find good use cases. However, the GPU gravy train for Nvidia would come to an end, and tech giants pouring tens of billions of dollars into AI data centers would likely never recoup those investments.
A high-risk investment
Nvidia is still valued at more than $3 trillion. The company is selling a lot of pricey data center GPUs and churning out incredible profits, but this valuation depends on two assumptions that are starting to look like they may not be true. First, that AI models will require ever-growing amounts of computing power to train and run, and second, that there is no ceiling in their capabilities.
DeepSeek's cheap AI model represents a big crack in Nvidia's growth story. It's not enough on its own to derail the AI king. However, when cheap AI models are combined with the very real prospect that AI models just aren't going to get much more capable from here, the basis for Nvidia's stratospheric market value falls apart.
I can't predict the future, and neither can you, but the uncertainty surrounding the future of the AI industry makes Nvidia an incredible risky stock.
