Nvidia loses USD 600 billion in a day. What the market was actually repricing
The largest single-day loss in US market history, triggered by a model released the previous week. The selloff was never really about Nvidia.
Nvidia shares fell about 17 per cent today, 27 January 2025, erasing close to USD 600 billion in market value — the largest one-day loss for a US company on record. The trigger was DeepSeek R1, released a week ago, and the accompanying claim that a frontier-class reasoning model had been trained for a fraction of the assumed cost.
What the market was repricing
Not Nvidia’s current business, which was unchanged that morning. The assumption underneath it: that frontier AI requires vast, sustained, capital-intensive training runs that only a few organisations can fund, and that every one of those runs needs an enormous quantity of specialised chips.
If a capable model can be produced far more cheaply than assumed, the projected demand curve for that hardware changes. That is what moved, and the same week had produced a USD 500 billion infrastructure announcement pointing the other way. The market was, reasonably, confused.
Why an SME should be calm about this
Because none of it changes what a thirty-person company can do on Tuesday.
Your AI costs are already a small line. You are not buying chips, not training models, and your provider’s share price is not your concern. If anything, an industry competing harder on efficiency is good for a buyer — cheaper answers, more providers, less lock-in.
The one real signal in the noise
The cost of capability is falling faster than almost anyone forecast, including people paid to forecast it.
That has a concrete consequence: do not sign anything long. A three-year commitment at today’s prices, in a market where a week can reprice the assumptions behind an entire industry, is an unnecessary bet. Twelve months with an exit is the sensible shape for almost every AI contract a Swiss SME will be offered this year.
And the thing worth repeating
Nothing in this week changes the bottleneck. The constraint in your company is not model capability or its price. It is knowing which process to point it at, having data clean enough to use, and having decided who checks the output.
Those three things cost nothing to work on and were unaffected by the largest single-day loss in American market history. Spend January on them.