On Monday January 27, 2025, NVIDIA’s stock fell about 17 percent in a single session, erasing roughly 589 billion dollars of market value. It was the largest one-day market-capitalization loss for any company in US stock-market history, surpassing NVIDIA’s own prior record. The Nasdaq Composite dropped more than 3 percent, the chip sector fell hard, and the shock spread to Europe, where the semiconductor-equipment maker ASML fell about 7 percent.
The trigger was DeepSeek, a Chinese AI lab whose R1 reasoning model had been released in late December. DeepSeek claimed the model reached performance comparable to leading Western systems while having been trained for a strikingly low cost - figures of under 6 million dollars and roughly two months were widely cited. If a competitive frontier model could be built that cheaply, investors reasoned, then the enormous spending on GPUs underpinning NVIDIA’s valuation might be far larger than necessary. The fear was, in effect, peak demand for compute.
The reaction was fast and not fully reasoned. One analyst described “a shoot first and ask questions later quality to the sell-off,” and NVIDIA recovered part of the loss within days as skeptics questioned DeepSeek’s cost claims and pointed out that cheaper inference can increase total compute demand rather than reduce it. But the episode crystallized the market’s central anxiety about the AI buildout: that the relationship between compute spending and the value it produces was far less certain than the share prices implied.
Why a business reader should care: this single day showed how tightly the entire AI-investment thesis was wired to one company and one assumption - that more compute is always needed. A credible counterexample, even an unverified one, was enough to wipe out more value in a day than most companies are worth.