Nvidia and other AI Infrastructure stocks are falling on news that China’s DeepSeek has a Large Language Model (LLM) that can be trained at 2% of the cost of other models. (Illustration from ChatGPT.)
I expect the bust to spread quickly. Microsoft, Google and Meta are all vulnerable given their huge investments in AI infrastructure. But Meta’s LLaMa was the basis for Deepseek’s work, and it’s inevitable that work will be disseminated.
In the end this is a victory for open source, for open discourse, and for start-up culture. The irony is, thanks first to the Inflation Reduction Act, then to Trump doubling-down on top-down change, taking power with an AI-infused crew of billionaires, it’s China that’s getting the gold medal.
Marcus sums up my thoughts on the matter. “The race to AGI will be won not by the country with the most chips but by the one that best fosters true innovation.”
That means you wait for Meta to fall, then double down on it. LLaMa isn’t true open source, since Meta demands the code added by users, but it’s the closest thing we have. I expect a lot of activity at the Linux Foundation and Apache Foundation over the next few weeks, organizing true open source projects around important AI concepts. A lot of laid off software engineers, at least those in their 20s and 30s, will find new opportunities in the next month.
Then we get to the real race, which is using this software and these AI concepts to deliver the productivity gains LLMs only promised us. Enterprise software companies like ServiceNow and Salesforce have an advantage here, being on the ground with customers, reflective of their needs. But this is going to be a huge year for start-ups.
Losers and Winners
The losers here are easy to name.
The Stargate Grifters go down first. A $100 billion data center isn’t necessary at this point. We may have adequate capacity already. That means we won’t need the fossil fuels Trump has promised to deliver the grift. Elmo can take off for Mars with our blessings. We don’t need him, either.
Cloud Czars will take a hit, because their dominance is based on owning huge data centers where demand won’t be rising as was once thought. Hardware stocks will fall, because we don’t need as much hardware as we thought.
Tech stocks generally are going to fall, because investors throw babies out with bathwater all the time. But there will be opportunities. Nvidia will still be heard from. It has enormous piles of foundational AI software. Make sure you have dry powder ready for the next boom.
Where will those opportunities lie? In software. Let 1,000 startups bloom, as Chairman Marc (Andreessen) might say. I think a lot of the action will come from India, where there is now a hyper-active startup culture and millions of programmers who are young, scrappy, and hungry. It’s possible that neither America nor China will be the big winners here because we’re just too old, a median age pushing 40, and start-ups are a young man’s game.
Beyond that, watch this space and every space you can find where smart people are discussing the latest AI software developments. Spread your bets. Prepare to be amazed. But only after the coming bust passes.
