Market capitalization exceeds 400 billion yuan! "The first stock in large models" Zhipu's revenue surged 132% last year; management: API price increase of 83% still in high demand

Why does demand increase despite an 83% price hike for AI · Zhipu API?

Source: Times Finance Author: Pang Yu Lin Xinlin

Image source: TuChong

“First Stock of Large Models” Zhipu (02513.HK) delivers its first performance report since listing.

On the evening of March 31, Zhipu disclosed its 2025 performance, achieving over 724 million yuan in revenue for the year, a year-on-year increase of 132%, making it currently the largest domestic large model company by revenue. “The (annual) revenue exceeded the target set at the beginning of the year,” management stated directly during the conference call.

With the halo of being the “world’s first large model stock,” Zhipu listed on the Hong Kong Stock Exchange on January 8 this year, with an issue price of HKD 116.2. On the first day of trading, its market value reached HKD 57.89 billion. On the first trading day after the earnings release (April 1), Zhipu’s stock price surged strongly during trading, closing up 31.94% at HKD 915 per share, with a total market capitalization surpassing HKD 400 billion. Nearly three months after listing, the stock price has increased more than sixfold.

However, behind the high revenue growth and rising stock prices, the profit model of large model stocks has shifted from a long-term topic to a pressing practical issue that needs answers.

In the money-burning race of large models, high R&D costs and other expenses have kept Zhipu deeply in the red. In 2025, Zhipu recorded a net loss of 4.72B yuan, an increase of 1.76 billion yuan compared to 2024.

According to Times Finance, during the evening conference call, management responded to core questions about whether the API price hikes in the first quarter of this year are sustainable, the progress of Agent (intelligent agent) commercialization, and the competitive pressure under the ecosystem encirclement by major tech giants.

Zhipu 2025 Financial Report

API revenue explodes, profitability still out of reach

From the financial data, the “heavy delivery” model focused on localized deployment remains Zhipu’s mainstay.

In 2025, Zhipu’s revenue from localized deployment services was 534 million yuan, a year-on-year increase of 102.3%. Even more surprising to the market was its cloud deployment business (MaaS platform, including open platform and API services), which achieved revenue of 190 million yuan, a rapid growth of 292.6% year-on-year.

Zhipu CEO Zhang Peng revealed during the conference call that the full-scale explosion of the MaaS platform is the core growth driver for Zhipu.

According to reports, the ARR (annual recurring revenue) of the MaaS platform is about 1.7 billion yuan, reaching a 60-fold rapid growth over the past 12 months. Meanwhile, through engineering optimizations on the inference side, Zhipu significantly reduced the token unit cost of calls, boosting the gross margin of the MaaS platform nearly five times to 18.9%.

Zhipu’s management explicitly stated that in the next two to three years, the open platform and API business will become the core carriers of 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 Zhipu is shifting its growth narrative from localized deployment to favor cloud API services, i.e., MaaS.

However, the current crowded large model track has led outsiders to pay more attention to Zhipu’s core barriers. Zhang Peng responded during the call that, from a technical perspective, Zhipu is currently mainly focusing on the layout of Coding (programming) and Agent.

In the coding track, Zhipu started early positioning at the end of 2020. In September 2025, it launched the AI-driven coding tool subscription plan GLM Coding Plan, and by March 2026, the paid developer users exceeded 242k. Since its launch six months ago, the Token call volume of Coding Plan has increased 15 times.

On the Agent front, Zhipu open-sourced the world’s first “phone-operating” AI Agent model AutoGLM last year; in March this year, it launched AutoClaw, providing users with a one-click installable local version called OpenClaw, and introduced the “Lobster Package” (Claw Plan) for individual and enterprise users. The package gained over 100k subscribers in two days and surpassed 400k in 20 days.

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

However, these impressive growth figures have not yet stopped Zhipu’s losses.

Due to high R&D expenses, Zhipu’s net loss in 2025 reached 242k yuan. R&D costs for the year totaled 3.18 billion yuan, 4.39 times the current year’s revenue. This means that over the past year, Zhipu “spent 4.39 yuan on R&D for every 1 yuan of revenue.”

Historically, Zhipu has maintained high-intensity R&D investment. The prospectus shows that from 2022 to 2024, Zhipu’s R&D expenditure soared from 84.4 million yuan to 100k yuan; during the same period, net losses were 143 million yuan, 788 million yuan, and 400k yuan respectively.

Behind this loss ledger lies the unresolved dilemma of the entire foundation model industry.

Under the “Scaling Law” followed by large models, the iteration of foundational models depends on massive computing clusters and vast data consumption, making it a capital-intensive marathon.

As Zhang Peng candidly said in a media interview last April: “The company believes that achieving AGI (Artificial General Intelligence) is still a long way off, requiring extensive technological exploration and research investment, and the costs of this process are very high. The trial-and-error costs of innovation are relatively increased, which is unavoidable. No matter how much Zhipu raises or earns, it is all capital on the road to AGI.”

Ambitions to rival Silicon Valley giants

Profitability has always been the Damocles sword hanging over all large model companies.

Looking across the industry, with MaaS as the core growth engine, Zhipu seems to be trying to replicate the business model of the popular Silicon Valley AI giant Anthropic, offering API and token billing services based on closed-source large models.

During the conference call, Zhipu’s management did not hide its ambition to benchmark against Anthropic, stating: “Zhipu will continue along a similar commercial path as Anthropic,” “by continuously improving the upper bounds of model intelligence, establishing pricing power in high-value scenarios, rather than engaging in simple low-price competition based on ‘large volume’.”

Zhipu believes that the industry will gradually differentiate in the future: low-end, volume-driven tokens will compete on scale and cost, while high-end, high-quality tokens will compete on capability and value. Zhipu is firmly betting on the latter. However, Zhang Peng also mentioned that Zhipu will continue to pursue a dual approach of open source and commercialization.

In fact, almost all leading domestic large model companies are trying to “cross the river by feeling the stones” from Silicon Valley. But due to differences in market payment environment, enterprise SaaS usage habits, and other factors, there is still a long way to go before they can truly match foreign giants in business scale and monetization.

From the revenue structure, about 80% of Anthropic’s income comes from enterprise and developer API calls, while Zhipu’s core remains localized deployment services (which accounted for 73.7% in 2025). In scale, Anthropic’s ARR reached $9 billion by the end of 2025 and nearly $19 billion by March 2026, while Zhipu’s MaaS ARR is about $250 million, with a significant gap.

Currently, domestic internet giants are the main service targets for Zhipu’s MaaS business. Zhipu disclosed that among China’s top ten internet companies, nine are deeply using the GLM model.

However, with Alibaba, Tencent, and other internet giants fully developing their own large models, the survival space for independent model vendors remains a concern in the capital market.

In response, Zhang Peng said during the call: “From a competitive perspective, large companies will definitely develop their own models, but due to resource constraints, they may not maintain a leading edge in all scenarios. Large companies are complex ecosystems and will not fully rely on themselves at every node; they will also incorporate external excellent technology providers to ensure fresh blood in the ecosystem and maintain a competitive advantage. In the current era of rapid technological iteration and model capability dominance, independent large model vendors have inherent advantages. This is also the foundation of Zhipu.”

It is worth noting that this year, with the market demand explosion driven by Agent applications represented by OpenClaw, Zhipu has also first tasted the commercial dividends brought by “Lobster.”

Before the Spring Festival this year, leading model companies including Zhipu launched a wave of price hikes. Management revealed that in the first quarter, Zhipu’s API call prices increased cumulatively by 83%. Even so, the market still shows a supply-demand imbalance, with call volume increasing by 400%.

But is this “price and volume increase” trend sustainable?

Zhipu’s management said frankly that the outside world is generally concerned whether the surge in “Lobster” call volume at this stage is a temporary prosperity or a long-term sustainable process. “From our perspective, the growth of Agent products represented by Claw Plan is not a phase of price-for-volume, but a natural increase after high-quality models establish value in real scenarios. We are very confident that this trend will continue.”

Additionally, regarding the industry’s computing power bottleneck behind the demand surge, management stated that there is indeed a supply constraint and bottleneck, which they see as a common industry issue. “Currently, the actual demand from major platforms and users is about 1 to 2 times the current support capacity.”

In the short term, Zhipu plans to supplement key supply links through external computing power procurement or internal resource allocation; simultaneously, it will focus more on high-value scenarios and core clients, prioritizing the efficiency of high-quality token supply. Moreover, it will explore cooperation with local inference platforms in overseas markets, promoting business through model deployment and revenue sharing.