Inside Zhipu’s 48% Surge: JPMorgan’s Pricing Power Thesis, GLM-5’s Open-Source Edge, and Washington’s Unintended Boost

Zhipu’s stock surged as much as 48% on Monday, June 15, in what looked at first like a straightforward analyst upgrade. JPMorgan had lifted its price target to HK$1,400 and declared Zhipu the stronger horse in China’s two-stock AI race . But beneath that headline, three forces collided to produce one of the most revealing single-day moves in the young history of publicly traded AI companies.

The JPMorgan call was the proximate trigger, but the deeper story is about how pricing power, open-source model performance, and an unprecedented US export control order combined to reframe Zhipu not just as a Chinese AI play — but as the only globally accessible frontier alternative at a moment when Washington was locking the doors on American models.

The JPMorgan call that split the ‘AI twin stars’

On June 12, JPMorgan issued a research report that created starkly contrasting ratings for the two Chinese AI model companies that had IPO’d in Hong Kong just months earlier. The bank maintained an Overweight rating on Zhipu (Knowledge Atlas Technology, ticker 02513.HK) and raised its target price sharply from HK$950 to HK$1,400. At the same time, it downgraded MiniMax (00100.HK) from Overweight to Neutral and slashed its target from HK$1,100 to HK$400 .

The entire divergence hinged on one variable: pricing power.

According to the report’s logic, whoever is cutting prices in China’s increasingly competitive AI market is the weaker player . JPMorgan concluded that Zhipu demonstrated sufficient model visibility and market position to sustain pricing, while MiniMax appeared to be conceding ground. The numbers that flowed from this judgment were stark: JPMorgan assigned Zhipu a valuation of 57 times its projected 2027 price-to-sales ratio, while MiniMax’s target price implied just 29 times — roughly half the multiple .

When trading opened on June 15, the market absorbed the call immediately. Zhipu jumped 33% on the day, touching an intraday gain of 48% at HK$1,620 before settling at HK$1,457, up 32.8% . MiniMax rose too — 7.4% on sympathetic momentum — but the performance gap told the story JPMorgan had framed: one of these companies was now seen as structurally advantaged, and the other was not .

It is worth noting that Bank of America initiated coverage on both companies the same week with buy ratings — HK$1,250 for Zhipu and HK$500 for MiniMax — meaning the market was not uniformly bearish on MiniMax . But JPMorgan’s call carried more weight because it was a downgrade from a previous bullish stance, and because its pricing-power thesis offered investors a clean relative-value story in a sector still opaque on fundamentals.

GLM-5: the open-source model that changed the narrative

JPMorgan’s call landed on already primed ground. Since February 2026, Zhipu had been building technical credibility with its GLM family of models, and by June the narrative around Chinese open-source AI had intensified.

GLM-5, Zhipu’s 744-billion-parameter flagship model released in February, was the first major open-source Chinese model to credibly challenge Western frontier systems. Trained entirely on Huawei Ascend chips without any Nvidia GPUs, the model scored 77.8% on SWE-bench Verified — within striking distance of Anthropic’s Claude Opus 4.5 at 80.9% — and outperformed Google’s Gemini 3 Pro on several evaluations . On Humanity’s Last Exam, GLM-5 scored 50.4% with tools, beating Claude Opus 4.5 and GPT-5.2 .

The model’s significance extended beyond benchmarks. It demonstrated that a Chinese lab could produce frontier-competitive AI without access to US chips, and it did so under an open-source license that made the model globally accessible . By June, Zhipu had released GLM-5.2, its most powerful model yet, and announced it would be made open-source later that week .

The timing was critical. Zhipu’s stock surged on June 15 — the same day the South China Morning Post reported that the GLM-5.2 open-source rollout “coincided with Washington’s abrupt order to suspend top US models overseas” . The model release gave investors a concrete product to value, and its open-source positioning made it a direct contrast to the restrictions unfolding simultaneously in the US.

The Anthropic order that changed the game

On June 12 — the same day JPMorgan issued its report — the US Commerce Department sent a letter to Anthropic CEO Dario Amodei ordering the company to suspend all access to its newest, most powerful models, Fable 5 and Mythos 5, by any foreign national “whether inside or outside the United States, including foreign national Anthropic employees” .

Anthropic received the instruction at 5:21 p.m. ET on June 12 and abruptly disabled both models for all customers to ensure compliance . The directive was unprecedented: never before had the US government forced a frontier AI lab to pull a deployed model offline . The order cited national security authorities and marked the most aggressive use of export controls against an AI system to date .

The scope was total. Foreign governments, businesses, individuals — and crucially, foreign-national Anthropic employees inside the US — were all blocked. Anthropic posted a statement saying it believed the order was a misunderstanding and was working to restore access, but the immediate effect was a global blackout of the company’s most capable systems .

For the investor narrative around Chinese AI stocks, this was an accelerant. As US frontier models became suddenly restricted, open-source Chinese models like GLM-5 looked like a hedge against a future where advanced AI access would be fragmented along geopolitical lines. CNBC reported that the JPMorgan call and the Anthropic restrictions combined to drive the surge in Zhipu shares, with analysts framing the company as a beneficiary of Washington’s harder line .

The 250% rally in context

The 48% intraday move on June 15 was dramatic, but it fit a broader pattern. Zhipu had been rallying since its January 8 IPO, when it debuted at HK$116.20 per share and closed its first day up 13.2% at HK$131.50 . The $558 million listing made it the first major Chinese generative-AI startup to go public .

After the IPO, the stock entered a sustained climb. By February, following the GLM-5 release, shares had roughly quadrupled from the IPO price and the company was planning a second listing on Shanghai’s Star Market . By late May, Zhipu briefly touched an intraday high of HK$1,993, pushing its market capitalization above HK$880 billion ($112 billion) — a gain of nearly 1,600% from its listing price .

The rally that included the June 15 surge — described in places as a 250% gain since the January IPO — reflected several compounding factors: index-inclusion speculation in May that pushed Zhipu up nearly 30% in a single session , the broader Hong Kong AI rally that had lifted both Zhipu and MiniMax, and the structural narrative that Chinese AI labs were closing the gap with Western competitors at a moment when US export controls were tightening rather than loosening.

JPMorgan’s call on June 12 crystallized this into a tradeable thesis: in a future where model access is restricted by nationality and chip sanctions force domestic innovation, the company that can sustain pricing power while keeping its best models open-source wins. The bank’s own numbers — a 57x multiple for Zhipu, 29x for MiniMax — put a price on that conviction .

What the divergence means going forward

The Zhipu-MiniMax split is not just a Wall Street ratings story. It reflects a real structural question about the Chinese AI market: can more than one independent model-maker sustain premium pricing, or will the market consolidate around whoever proves strongest?

JPMorgan’s answer, as of June 2026, is that Zhipu has the edge. The bank raised its revenue forecasts for the company between 2026 and 2030 by 46% to 78%, citing strong API demand and model iteration velocity . MiniMax, downgraded but still rising on the day JPMorgan cut it, retains defenders — Bank of America’s buy rating among them — but faces the harder case: prove it can price without losing share .

The Anthropic order added an unpredictable variable. If the US continues tightening export controls on frontier models — including the “deemed export” rule that treats showing technology to a foreign national inside the US as an export — then open-access models trained outside the US chip ecosystem become more than an alternative. They become a prerequisite for any non-US entity that wants to stay at the frontier.

That is the strategic logic behind the June 15 rally, and why Zhipu’s surge was about more than one analyst upgrade.