Market capitalization exceeds 400 billion! The "Number One Stock in Large Models" Zhipu's revenue surged 132% last year. Management: API price increase of 83% still in high demand.

Image source: VCG

“China’s first listed company in large models,” Zhipu (02513.HK) has released its first set of results after listing.

On the evening of March 31, Zhipu disclosed its 2025 performance. For the full year, revenue exceeded RMB 724 million, up 132% year over year—making it the largest large-model company in China by current revenue scale. “(For the full year) revenue exceeded the company’s targets set at the start of the year,” management said plainly on the earnings call.

With the halo of “the world’s first listed company in large models,” Zhipu listed on the Hong Kong Stock Exchange on January 8 this year. The issue price was HK$116.2, and on the first day of trading its market cap immediately reached HK$57.89 billion. After the financial report was released, on the first trading day (April 1), Zhipu’s share price surged strongly intraday. By that day’s close, the stock jumped 31.94% to HK$915 per share, and its total market cap surpassed HK$400 billion. Within nearly 3 months since listing, the cumulative share-price gain exceeded 6 times.

However, behind the high revenue growth and the strong rally in the share price, the profitability model problem for large-model companies has shifted from a long-term issue to a real prompt that urgently needs answers.

In this money-burning track of large models, high R&D spending, among other factors, has left Zhipu still in losses. In 2025, Zhipu recorded a net loss of RMB 4.718 billion, widening its losses by RMB 1.76 billion compared with 2024.

According to Time Finance, on that evening’s earnings call, management responded to key questions including whether this year’s Q1 API price-hike wave is sustainable, the progress of Agent commercialization, and the competitive pressure under the “ecosystem enclosure” by big firms.

Zhipu’s 2025 financial report

API revenue surges, not yet profitable

Based on the financial report data, the “high-touch delivery” model, which leans toward localized deployments, still remains Zhipu’s core foundation.

In 2025, Zhipu’s revenue from localized deployment services was RMB 534 million, up 102.3% year over year. More surprisingly to the market was that its cloud deployment business (MaaS platform—i.e., the open platform and API business) generated revenue of RMB 190 million in the period, with a year-over-year growth rate as high as 292.6%.

Zhipu CEO Zhang Peng revealed on the earnings call that the comprehensive breakout of the MaaS platform is Zhipu’s core growth driver.

As introduced, the MaaS platform’s ARR (annual recurring revenue) is about RMB 1.7 billion, reaching 60x growth over the past 12 months. Meanwhile, through engineering optimization on the inference side, Zhipu significantly reduced the unit cost of Tokens it calls, boosting the MaaS platform’s gross margin by nearly 5x to 18.9%.

Zhipu’s management clearly stated that over the next two to three years, the open platform and API business will become the core vehicle for the company’s revenue scale and profit release, while the Agent solution will serve as a supplement for customer acquisition or scenario validation.

This means that Zhipu is changing its growth narrative—from leaning toward localized deployment to shifting toward cloud-based APIs, i.e., MaaS.

However, since the large-model market is crowded, external parties are also paying even more attention to Zhipu’s core moat. On the earnings call, Zhang Peng responded that, from a technical perspective, at this stage Zhipu is mainly focusing on Coding (programming) and the Agent layout.

In the programming track, Zhipu began positioning ahead of time as early as late 2020. In September 2025, it rolled out an AI-driven coding tool subscription plan called GLM Coding Plan. As of March 2026, the number of paying developer users had exceeded 242,000. Since its launch 6 months ago, the Token call volume of the Coding Plan has grown 15 times.

On the Agent layout, last year Zhipu open-sourced the world’s first AI Agent model AutoGLM that can “operate your phone.” In March this year, it launched AutoClaw, providing users with a local version of OpenClaw that can be installed and deployed with one click, and it also launched “Lobster set menus” (Claw Plan) aimed at both individual and enterprise-level users. Two days after the package launch, the number of subscribing users exceeded 100,000; after 20 days, it surpassed 400,000.

According to management, as of March 2026, Zhipu’s registered enterprises and users exceeded 4 million, serving more than 218 countries and regions worldwide.

However, even flattering growth data has not yet stopped Zhipu’s losses.

Due to factors such as high R&D spending, Zhipu’s net loss in 2025 reached RMB 4.718 billion. Full-year R&D expenses were RMB 3.180 billion, 4.39 times the revenue for the period. This implies that over the past year, Zhipu had to invest RMB 4.39 in R&D expenses to obtain RMB 1 of revenue.

Looking back, Zhipu has always maintained high-intensity R&D investment. The prospectus shows that from 2022 to 2024, Zhipu’s R&D expenditure surged from RMB 84.40 million to RMB 2.195 billion. Over the same period, net losses were RMB 143 million, RMB 788 million, and RMB 2.958 billion, respectively.

Behind this loss ledger lies a predicament the entire pre-training model (base model) industry has yet to resolve.

Under the “Scaling Law” followed by large models, the iteration of base models cannot be separated from support from massive compute clusters and the consumption of enormous volumes of data—this inevitably becomes a capital-intensive marathon.

As Zhang Peng admitted in a media interview last April, “The company’s view is that the path to achieving AGI is still very long and requires large amounts of technical exploration and research investment, and the cost of this process is extremely high. The trial-and-error cost for innovation exploration will be relatively higher, which is unavoidable. No matter how much Zhipu has raised or how much revenue it has obtained, actually it is all the money for the road on the path to AGI.”

Ambitions to match Silicon Valley’s giants

Profitability—the “Sword of Damocles” hanging over all large-model companies—always remains.

Against the backdrop of the whole industry, with Zhipu’s MaaS business as its core growth engine, it seems to be trying to replicate the commercial route of Silicon Valley’s current AI giant Anthropic: by providing API and Token-based billing services through closed-source large models.

On the earnings call, Zhipu’s management did not hide its ambitions to benchmark Anthropic, and said: “Zhipu will continue along a commercial path similar to Anthropic”“By continuously raising the upper bound of model intelligence, we will establish pricing power in high-value scenarios, rather than taking the simple low-price competition route of ‘high volume to keep everyone fed.’”

Zhipu believes that in the future the industry will gradually diverge. Low-end, high-volume Tokens will move toward scale and cost competition, while high-end, high-quality Tokens will move toward capability and value competition. Zhipu is firmly betting on the latter. However, Zhang Peng also mentioned that Zhipu will, as always, keep walking on two legs: open-source and commercialization.

In reality, nearly all leading domestic large-model enterprises are trying to “cross the river by feeling for stones in Silicon Valley.” However, because of differences in paid market soil, enterprise-level SaaS usage habits, and the like, it will still take a long time to truly match overseas giants in commercial scale and monetization capability.

From the revenue structure perspective, about 80% of Anthropic’s revenue comes from enterprise and developer API calls, while Zhipu’s current performance “core base” is still localized deployment services (with a share of 73.7% in 2025). In terms of scale, Anthropic’s ARR reached $9 billion at the end of 2025, and as of March 2026 it was close to $19 billion, whereas Zhipu’s MaaS business ARR is about $250 million—there is still a clear gap between the two.

At present, domestic internet giants are important target customers for Zhipu’s MaaS business. According to Zhipu’s disclosure, among China’s top ten internet companies, 9 have already been using GLM models deeply.

But in the face of internet giants such as Alibaba and Tencent fully ramping up their own large-model research and development, investors remain concerned about the survival space for independent model companies.

In response, Zhang Peng said on the earnings call: “From a competitive perspective, big companies will definitely have self-developed models, but limited by resources, they may not be able to maintain leadership competitiveness across all scenarios. Big companies themselves are a complex ecosystem, so they won’t rely on themselves completely at every node. They will also tap into excellent external technology suppliers to keep the ecosystem’s blood fresh and ensure the business doesn’t lose its first-mover advantage. At the current time when technology iterates quickly and model capability leads, independent large-model companies have very strong inherent advantages. This is also the foundation on which Zhipu stands.”

Worth mentioning is that this year, as Agent applications represented by OpenClaw ignited market demand, Zhipu was among the first to taste the commercial dividend brought by “lobsters.”

Before this year’s Spring Festival, leading model companies represented by Zhipu sparked a wave of price hikes. According to management, in the first quarter, Zhipu’s API call pricing increased cumulatively by 83%. Even so, the market still showed a situation of supply failing to meet demand, and call volume grew by 400%.

But is this wave of “both volume and price rising” sustainable?

Zhipu’s management said directly that the market widely cares about whether the call-volume surge brought by “lobsters” at this stage is a short-term boom or a long-term sustainable process. “From our perspective, the growth of Agent products represented by the Claw Plan is not a stage phenomenon of growing through discounting to chase volume. Instead, it is natural expansion after high-quality models establish value in real-world scenarios. We are very confident in the continuation of this trend.”

In addition, when discussing the industry compute bottleneck behind the surge in demand, management said that there indeed is currently a compute supply constraint and bottleneck. They believe this is an industry-wide issue. “At present, the actual demand of major platforms and users is roughly 1 to 2 times the actual call volume we support today.”

On the short-term strategy, Zhipu will, on one hand, make up supply in key links through external compute procurement or internal resource reallocation. On the other hand, it will further focus on high-value scenarios and core customers, prioritizing supply efficiency of high-quality Tokens. In addition, in overseas markets it will explore cooperation with local inference platforms and advance the business via a revenue-sharing model tied to model deployment.

(责任编辑：Guo Jiandong)

     【Disclaimer】This article only represents the author’s personal views and is not related to Hexun. The Hexun website remains neutral regarding statements and judgments made in the text, and provides no express or implied guarantees regarding the accuracy, reliability, or completeness of any content contained herein. Readers are only for reference, and bear all responsibility on their own. Email: news_center@staff.hexun.com