NextFin News -- At the start of this year, Zhipu and MiniMax’s race in Hong Kong stocks to become “the world’s first listed AI foundation-model company” briefly became a market focal point.

In the end, Zhipu got there first in Hong Kong on January 8, pricing its shares at HK$116.2. The stock rose about 13% on its debut, then climbed more than 20% for two straight trading days.

MiniMax rang the bell the next day, on January 9, but its surge was even stronger: it priced at HK$165 per share and skyrocketed 109% on its first day, then gained another 15% the following session before a modest pullback.

Nearly half a year has passed, and the two “titans” are still the only two mainstream independent foundation-model companies currently listed.

What’s more, around mid-year, the two companies almost simultaneously announced plans to kick off IPOs on Shanghai’s STAR Market. This time, MiniMax took the lead on May 31 by issuing an announcement confirming it was launching the process of “returning to A-shares.” Then, on June 1, Zhipu announced it intended to list on the STAR Market and raise RMB 15 billion.

Compared with the scene when they listed in Hong Kong, what has changed for the companies, the industry, and the market? And what new story can they tell in A-shares?

Stock Prices Soar Over Past 6 Months

Over the past six months, both Zhipu and MiniMax posted strong share-price performances at one point.

As of 12:00 on June 5, Zhipu’s share price was HK$1,300, up about 10x from its IPO price, with a market capitalization of HK$579.6 billion. MiniMax was trading at HK$575, up about 2.5x from its IPO price, with a market cap of HK$180.3 billion.

Although both stocks plunged on June 5, market enthusiasm for the two companies remained strong right up until they announced their A-share plans.

Zhipu in particular had just hit an intraday record high of HK$1,993 per share on May 29, with its market cap at one point approaching the HK$900 billion mark.

MiniMax hit its peak in March, when its share price reached as high as HK$1,333 per share and its market cap topped HK$380 billion. In May, the company’s market cap also stayed above HK$200 billion.

Previously, the market’s valuation of the two companies was largely driven by their “scarcity.” However, the push for A+H listings, together with the accelerated pace of U.S. IPOs by global large-model giants such as Anthropic and OpenAI, will all affect that valuation dimension. The recent pullbacks in both stocks may be related to this.

More worth watching, though, is how the two companies are performing financially and how the industry’s underlying logic is shifting. This also largely explains why, in secondary-market pricing, MiniMax started from a higher base while Zhipu has shown stronger staying power.

In March, the two companies released their first annual reports since going public, one after the other.

Zhipu posted 2025 revenue of RMB 724 million, up 132% year on year, with an adjusted net loss of RMB 3.182 billion; MiniMax reported revenue of US$79.038 million (about RMB 570 million), up 158.9% year on year, with an adjusted net loss of US$250 million (about RMB 1.75 billion).

On the surface, MiniMax grew slightly faster and its losses looked relatively more manageable, but Zhipu’s 41% gross margin was markedly higher than MiniMax’s 25.4%.

The market has long been familiar with the contrasting narratives: Zhipu focuses on B2B while MiniMax targets consumers; Zhipu digs deep locally while MiniMax primarily goes after overseas markets.

When they first listed on the Hong Kong Stock Exchange at the beginning of the year, consumer-facing and overseas expansion carried bigger growth expectations and more room for imagination. MiniMax’s business model was also seen as more mature, and its rally was at one point more aggressive.

But over the past six months, the core narrative has been AI landing in real use cases—entering office and production scenarios—and that narrative directly determines whether AI can break the “bubble curse” and create longer-term value. In particular, the blistering rise of B2B-focused Anthropic, along with Agent “breakouts” represented by “Lobster OpenClaw,” has significantly lifted expectations for the development outlook of large models in the B2B market.

Zhipu, while labor-intensive and burdened by heavy R&D costs, has already secured a relatively solid position in China’s government and enterprise market, with high average contract values and strong stickiness. Moreover, it moved early on coding and Agents, aligning with the industry’s mainstream trends over the past six months. At one point, in “U.S. stock mapping,” it was benchmarked against Anthropic, and its valuation performance has also been more resilient recently.

As for MiniMax, the good news is that its May operations update showed a global user base of about 300 million, and its ARR doubled over the past two months. But as competition for traffic among consumer AI apps intensifies, the company’s future growth rate will come under pressure, and its reliance on paid traffic acquisition will also weigh more heavily on profit expectations in the market.

More crucially, the AI pricing landscape has undergone a dramatic shift—and a clear split.

Up until the AI “price war” ahead of the Spring Festival, it looked as though the broader trend of large models going free and getting cheaper would continue. But then the Agent boom completely “lifted” the token economy: individual token consumption surged, and demand for tokens in enterprise and industry use cases kept exploding, pushing token prices back upward.

Zhipu CEO Zhang Peng said at an earnings briefing in March this year that the company had raised its API pricing by a cumulative 83% in Q1, while usage not only failed to fall but instead rose—up 400%. Zhipu raised model prices in both March and April, and its benchmarking approach was to compare its “cache-hit token price in coding scenarios” against Anthropic’s Claude Sonnet 4.6. Clearly, the company is now pursuing a “technology premium” path.

MiniMax is also raising prices. Its newly released flagship model, MiniMax M3 has again increased API pricing. However, on the consumer subscription side, the company is still emphasizing value for money; its Token Plan subscription is notably price-attractive within the industry. In fact, DeepSeek—positioned primarily toward consumers—has continued to cut prices sharply, but with more and more usage restrictions. Shortly after V4 was released, the upload-file and online-search functions in Expert Mode were removed one after another, citing “tight resources.” Meanwhile, Doubao, which has the largest user base and the highest activity levels, also rolled out a paid subscription model, sparking widespread industry debate.

The impact of shifting token-price trends on large models is still unfolding. For now, companies focusing on enterprise customers appear relatively resolute in repricing—and find it easier to gain market acceptance—but that also means greater pressure ahead to deliver value and prove the business model. On the consumer side, pricing is comparatively more fraught: everyone still needs to test what the market will accept and where the tipping point really lies.

Beyond industry metrics, both of the “two front-runners” are also facing lock-up expiry pressure. Zhipu will see restricted shares equal to 5.76% of its total share capital become tradable in July this year, and in January next year that figure will jump to as high as 40%. MiniMax, too, will face its first large-scale post-IPO lock-up expiry in July. Whether lofty valuations can withstand the shock of a surge in free-float is another question both firms must answer before returning to the A-share market.

What New Stories Will the Two AI Tigers Tell? 

Although there have already been plenty of “A+H” dual-listing cases, over the past two years the AI-driven tech wave has pushed market enthusiasm higher—while also making swings more violent—so the differences between the A-share and Hong Kong stock markets have drawn even more attention.

For the “two AI tigers” listing on the A-share market, they also need some new narratives.

At present, Hong Kong investors tend to value a globalization narrative and a liquidity premium; meanwhile, on today’s STAR Market in the A-share market, “hard tech” and “domestic substitution” are the most frequently mentioned keywords.

On this front, it may have been a pain point for Zhipu in the past, but it also has the potential to become a highlight in a new story.

After Zhipu released GLM-5 in February this year, it publicly apologized for service instability caused by insufficient computing power. At the time, the market noted that its computing capacity depended on third-party suppliers and that it lacked self-built computing centers—meaning its business could be constrained and its ability to control long-term costs was also in question.

In the company’s latest disclosed plan to raise RMB 15 billion on the STAR Market, RMB 12 billion is earmarked for investment in a general-purpose AI foundation-model project—primarily for R&D iteration of the next-generation base model (GLM-6 and subsequent series) and continued investment in large-scale compute clusters.

The core narrative of Zhipu’s A-share return will likely place greater emphasis on an independently developed, platform-oriented direction built around “foundation models + domestically produced computing power + agents,” increasingly underscoring its role as a technological infrastructure platform.

Notably, in late May, a relevant official at the National Development and Reform Commission (NDRC) also stated clearly that China would guide domestic large models to step up adaptation to domestically produced compute chips—maintaining rapid development while ensuring autonomy and controllability, responsible development, and steady long-term progress. For Zhipu, which has long focused on the government and enterprise market, synergy with domestic computing power will likely be emphasized even more in the A-share context.

MiniMax is also facing tight computing capacity. The “domestic computing power” story works here as well, though it may play out differently from Zhipu.

MiniMax follows a typical asset-light model, with low fixed-asset spending, and its computing power comes almost entirely from third-party cloud services. However, it has consistently responded quickly when it comes to integrating with domestic computing platforms. Going forward, its integration with domestic compute is more likely to focus on software-layer ecosystem adaptation and efficiency optimization. And playing the “ecosystem card” is also one of the development directions most closely watched by domestic compute providers that urgently need to expand their ecosystems, break out, and challenge NVIDIA’s CUDA monopoly.

In addition, MiniMax still has plenty of “hard tech” narratives to develop—such as multimodal capabilities and intelligence density—while concepts like cost-effectiveness and overseas expansion have also long been in the market spotlight.

As with their initial listing in Hong Kong, large-model companies’ return to the A-share market carries strong benchmark significance for both the capital market and the industry. Moreover, amid a denser cadence of earnings disclosures and heightened market scrutiny, the profitability outlook for the two companies—both still loss-making—will be examined more frequently and more rigorously. (Reporting by Hu Jiameng; Editing by Yang Lin)