3 min read
DeepSeek, the Chinese lab that spent early 2026 giving Silicon Valley a collective aneurysm, is reportedly in talks to raise about $1.5 billion at a $71 billion valuation, then go public. Lovely stuff, bar one detail: roughly six weeks ago the same company was valued at around $50 billion. That is a $21 billion mark-up in the time it takes the rest of us to finish a tube of toothpaste, and DeepSeek has put out precisely nothing new to earn it.
The valuation speed-run
The round is reportedly backed by Tencent and a Beijing state-linked AI fund, per Bloomberg, and it lands barely a month after the company took its first-ever outside money. DeepSeek is also lining up a mainland IPO, with a filing possibly as soon as late 2026 and a listing targeted for 2027. For a firm founded in 2023 by Liang Wenfeng, a former quant-fund manager, that is a genuinely unhinged trajectory, and I mean that with a certain grudging respect.
Where the “China does AI on a shoestring” myth goes to die
Here is the bit that punctures the fairy tale everyone fell for in January. DeepSeek’s entire legend was built on training frontier-grade models for pocket change. But look at what the money is actually earmarked for: gigawatt-scale data centres, in-house inference chips, and a hard shove into AI agents. Turns out that once you have proven you can build the model, you still have to pay the same eye-watering electricity and silicon bills as every bloated Western incumbent you were supposedly humiliating. The laws of physics, disappointingly, do not offer a China discount.
Which is absolutely fine, if you happen to be a Beijing-anointed national champion with a state fund on speed-dial. It is rather less fine for the plucky-underdog story the West told itself to feel better about getting caught napping. DeepSeek is a strategic national asset being fattened for a domestic listing, and the valuation is climbing on narrative, nationalism and raw FOMO far more than on revenue.
The market-watcher’s note (not investment advice)
A private valuation is a number two parties agreed on in a room with the blinds down, not a fact about the universe. $71 billion for a three-year-old company that has jacked its own price up 40% in six weeks tells you an enormous amount about how much desperate capital is hunting for an AI story, and almost nothing about what the thing is actually worth. When the IPO finally lands, the public gets handed the bill and finds out which of the two it was. Eyes open, wallets closed.
The AI tool stack actually worth paying for
One email a week. The tools, models and moves that matter, minus the hype and the horseshit filter set to maximum. Free.
Did you know: DeepSeek’s January 2026 model release briefly vaporised hundreds of billions of dollars off US tech stocks in a single trading day, once investors realised a competitive model could be trained for a fraction of the assumed cost. The company then spent the next six months raising exactly the kind of money that assumption swore it would never need. Funny, that.