China's AI ambitions stopped being abstract this morning and showed up on a stock ticker. Shares in Zhipu — the Beijing lab behind the open-weight GLM models — surged as much as 42% in Hong Kong, briefly pushing its market value above one trillion Hong Kong dollars (around $128bn). It's the clearest market signal yet that the "open model" strategy China has been pursuing all month is being taken very seriously indeed.

Zhipu's trillion-dollar morning

The numbers are striking. Zhipu (which trades as Knowledge Atlas Technology, ticker 2513.HK) only listed in Hong Kong in January, at HK$116.20 a share. This morning it touched HK$2,980 — a rise of more than 2,400% since its January listing. The trigger was last week's release of GLM-5.2, its latest open-weight model, which ranked second in the world on the Code Arena leaderboard for front-end web development, behind only Anthropic's Claude Fable 5.

A few caveats worth keeping straight: this is a secondary-market spike, not an IPO debut (Zhipu has been public since January), and the trillion-HKD figure was an intraday peak rather than a settled valuation. The stock has also had more than one big move recently — it jumped sharply a week ago after JPMorgan named it an AI winner. But the direction of travel is unmistakable. A Chinese lab giving its best model away for free is now, on paper, one of the most valuable AI companies in the world. As we noted when GLM-5.2 first landed, Zhipu's founder Tang Jie has been pointedly dismissive of Western predictions that China is years behind — and the market just agreed with him.

Europe's answer: an open model of its own

If China's play is open models at national scale, Europe has been experimenting with the same idea. Apertus, the fully-open large language model Switzerland's ETH Zurich, EPFL and national supercomputing centre released last September, resurfaced on developer forums over the weekend. Its distinguishing feature isn't raw benchmark-topping performance — it's openness all the way down: open weights, open training data, and an open recipe, built on public infrastructure and lawfully-sourced data.

It's the same instinct behind the Netherlands' GPT-NL and the broader "sovereign AI" push we've tracked since the Anthropic export-ban fallout: the worry that depending on a handful of American labs — models a foreign government can switch off — is a strategic risk a continent shouldn't accept. Apertus is Europe's open-by-design answer to that question. Whether open-but-modest beats closed-but-frontier is the open debate; that two of the three big AI blocs are now experimenting publicly with openness is the story.

Still rolling: Anthropic's identity checks

The piece of news dominating discussion forums over the weekend was Anthropic's plan, which we covered yesterday, to begin age and identity verification for some consumer Claude users from 8 July. It stayed at the top of Hacker News for most of the weekend — a sign of how raw the nerve is around a safety-focused lab asking users for biometric ID. Nothing has changed on the facts, but the volume of reaction is itself worth noting: the trade between safety and privacy is one this industry is going to keep relitigating.

The throughline

Three stories, one thread. A Chinese open lab worth a trillion HK dollars, a European open model built on public money, and a US lab tightening who gets to use its most powerful features. The open-versus-closed question that has run through this whole month isn't settling — it's hardening into the defining strategic divide of the AI race, and this morning the markets put a very large number on one side of it.