Earlier this month, in this very newsletter I flagged a potential risk facing chip titan Nvidia that could temper its extraordinary share price run — commoditisation in AI:
“As AI models mature, the value could shift away from premium hardware to software optimisation and scale, leaving Nvidia holding an increasingly expensive shovel in a gold rush that’s moving elsewhere.”
That was precisely the risk on the mind of investors when they sent Nvidia’s stock plummeting as much as 17% overnight, shedding roughly USD600 billion ($960 billion) in the process and taking the rest of the tech indices down with it. The reason behind the correction was the market coming to grips with the latest open source AI models from Chinese AI lab DeepSeek, which go toe-to-toe with bleeding edge systems from OpenAI, Anthropic and Google at a fraction of the compute cost.
What’s more, DeepSeek claims to have trained these models on last-generation Nvidia hardware. While there are open questions about what the company might be leaving out of its story, this still represents a narrative violation for the AI investment thesis. Investors have largely tolerated extraordinary capex from big tech players like Microsoft, assuming that the world-historic infrastructure buildouts were necessary to drag tech into the next phase of hypergrowth.
