A year since the model that moved markets
DeepSeek R1 appeared on 20 January 2025 and wiped hundreds of billions off technology stocks. Twelve months later, here is what it actually changed.
One year ago today, a Chinese laboratory released an open-weight reasoning model built for a fraction of the expected cost. Stock markets fell sharply, and for two weeks it was the only subject in technology.
This blog covered it at the time and advised Swiss SMEs to do nothing. An anniversary is a fair moment to check whether that held.
What the shock got right
That frontier capability was not defensible for long. Everything R1 demonstrated — reasoning at low cost, published weights, a credible challenger from outside the American labs — became normal within the year. By November an open-weight reasoning model was a Thursday.
That prices would fall. They did, substantially and repeatedly, throughout the year.
What the shock got wrong
That it was an event. It was the visible moment of a trend already underway, and the market reaction treated a direction as a surprise.
Companies that changed supplier in the weeks after, on the strength of the news, mostly changed back or wished they had not moved. Companies that did nothing lost nothing, which was this blog’s advice at the time and is the part we would repeat.
The lesson that generalises
When a technology story is large enough to move stock markets, it is a story about capital, not about your Tuesday. The two are related over years and almost never within a quarter.
That test has now held through R1, through GPT-5, through the November cluster and through a corporate restructuring. A company of thirty people has not yet needed to act on a headline.
What we are watching instead
The same three things as last January: where the processing happens, whether you can pin a version, and whether your people actually use what you bought.
None of those made a front page this year. All three determined whether a Swiss SME got anything out of the technology.