Bloomberg reported on June 18, 2026 that shares of Zhipu — registered as Knowledge Atlas Technology JSC Ltd — are up roughly 170% since the end of March on the Hong Kong exchange, while MiniMax Group Inc. is down about 50% over the same period. Zhipu became the first foundation-model AI startup to list on a public exchange anywhere and is leaning into enterprise deployments of its GLM family. The divergence is large enough that Bloomberg flagged it as an emerging long-Zhipu, short-MiniMax pair trade. The spread is expected to widen in early July when a larger tranche of MiniMax stock comes off lockup.
China's frontier-model market is doing in public what the US private market is doing in private: concentrating. Investors are already paying for the eventual winner premium and discounting the also-rans, even before either company has demonstrated durable revenue at the Anthropic or OpenAI scale. The fact that this is happening on a public exchange matters because it gives global allocators a transparent price signal on a category that, in the US, only shows up in quarterly secondary-market data leaks. Hong Kong has effectively become the listing venue where the Chinese AI race gets repriced daily.
The bigger story is how fast Hong Kong has filled the China AI IPO vacuum. Mainland exchanges have been slow to absorb AI listings, and the Trump-era US restrictions made New York harder; Hong Kong has been the beneficiary. Zhipu's debut and now its 170% run-up is the kind of outcome that pulls more Chinese AI capital toward Hong Kong listings — which compounds Beijing's existing interest in domestic capital markets carrying the AI buildout. Watch the next two or three Chinese AI offerings: pricing and aftermarket performance there will tell you whether Zhipu is a one-off or a regime change.
Takeaway for learners: "the AI race" is not a single race — it's at least four, split by region and by capital-market access. Public-market signals are usually crude, but they are the only ones outside the labs you can actually see. A pair trade like Zhipu vs MiniMax is the kind of market-implied prediction worth tracking even if you never buy a share, because it tells you which model lab has convinced enterprise customers and which one has convinced only VCs.