"AI New Generation" market capitalization surpasses HKD 400 billion! Zhipu raises prices, ending API price war: Can the valuation logic of "China's version of Anthropic" hold up?

Ask AI · Zhipu’s valuation logic compared to Anthropic: how does it handle profit loss pressure?

This newspaper (chinatimes.net.cn) reporter Shi Feiyue Beijing reports

“Currently, Zhipu has become one of the domestic vendors with the highest paid Token consumption.” On the evening of March 31, CEO Zhang Peng of Zhipu stated at the earnings release call, including reporters from Huaxia Times, and frequently compared Zhipu with Anthropic. On April 1, Zhipu’s stock price soared 31.94% to HKD 915 per share, with a total market value surpassing HKD 400 billion for the first time, exceeding MiniMax by over HKD 70 billion.

On the evening of March 31, Zhipu disclosed its first annual report since listing, showing high growth in both revenue and gross profit for 2025. However, the other side of high growth is still increasing losses and the ultimate challenge to its business model. Comparing to Anthropic is easy, but truly becoming like Anthropic is difficult. Can the Token economy shift from a short-term craze to a sustainable profit pillar? The test is not only the increase in call volume but also Zhipu’s future model capabilities.

Targeting the “Chinese version of Anthropic”

At the earnings call, Zhang Peng repeatedly mentioned Anthropic and extensively discussed the similarities between Zhipu and Anthropic’s business models, indicating an intention to become the “Chinese version of Anthropic.”

Anthropic is recognized globally as the “second pole” in AI, and also OpenAI’s strongest challenger. Founded by former core members of OpenAI, media reports indicate its latest valuation has reached $380 billion, with annual revenue surpassing $19 billion, growing over 10 times annually over the past three years, dubbed by the media as “the fastest-growing enterprise software company in history.”

“Anthropic’s ARR (Annual Recurring Revenue) growth proves one thing: when the model is strong enough, the API itself is the best business model,” Zhang Peng said. The growth logic of Anthropic is very clear: focus on delivering the most powerful models via API to enterprises and developers, allowing intelligent participation in creating economic value.

Similarly, Zhang Peng pointed out, this business model and growth are happening at Zhipu: Anthropic’s early growth relied on the model being used frequently by global developers. In 2025, Zhipu launched China’s first programming package “GLM Coding Plan,” with the number of paying developers rapidly exceeding 242,000, and Token call volume increased 15-fold in six months; about 80% of Anthropic’s revenue comes from enterprise API services. Relying on BigModel.cn, Zhipu’s MaaS platform has become a hub connecting foundational models with 4 million enterprise applications and developers.

Zhang Peng also revealed that nine of the top ten internet companies in China deeply call Zhipu’s GLM daily. After each model release, major companies like ByteDance, Alibaba, Tencent, etc., immediately integrated officially within 24 hours.

The 2025 financial data supports Zhang Peng’s statement: revenue from the open platform and API was 190 million yuan, up 292.6%; revenue from enterprise-level intelligent agents was 166 million yuan, up 248.8%; revenue from enterprise general large models was 366 million yuan, up 70.5%; revenue from technical services and others was 2.544 million yuan, up 31.6%.

Of course, from the perspective of revenue scale, Zhipu still lags behind Anthropic. In 2025, Zhipu’s revenue reached 724 million yuan, up 131.9%; gross profit was 297 million yuan, up 68.7%.

However, like MiniMax listed at the same time, the sharp increase in revenue did not improve Zhipu’s losses. According to the financial report, the company lost 4.718 billion yuan in 2025, with an adjusted net loss of 3.182 billion yuan.

Zhipu explained in the financial report that this was mainly due to continued increases in R&D investment. The reporter noted that, due to rising employee costs and increased payments to third-party computing service providers, Zhipu’s R&D expenditure in 2025 increased by 44.9% to 3.18 billion yuan, far exceeding annual revenue; besides the significant growth in R&D spending, the company’s sales, marketing, and general administrative expenses also increased last year.

Additionally, due to the rising proportion of cloud deployment business and the phased decline in gross margin of localized deployment, Zhipu’s gross margin dropped from 56.3% in 2024 to 41% in 2025.

Token economy “turning point”

In fact, looking at the revenue structure last year, enterprise-level general large models remained Zhipu’s largest income source. But from recent trends, including the Token economy boom triggered by OpenClaw and the growth trend of various income streams last year, it can be inferred that Zhipu’s open platform and API revenue scale is expected to further increase, possibly becoming Zhipu’s pillar income in the near future.

Since 2024, to reduce costs and compete for market share, domestic large model vendors launched a price-cutting wave. But in the first quarter of this year, as the Token economy trend emerged, raising prices became a new choice for some large model vendors. Zhipu’s API call pricing directly increased by 83%. Even so, Zhang Peng revealed that the market still shows a supply-demand imbalance, with Zhipu’s API call volume growing by 400%.

It is reported that Zhipu has realized the monetization of Token value globally. Its GLM models have been fully deployed on cloud services such as Google Vertex AI, AWS Bedrock, Fireworks, Cerebras, and have entered international mainstream model aggregation platforms like OpenRouter and Vercel, ranking No.1 in OpenRouter paid models.

This means that by 2026, Zhipu has achieved a double leap in Token economy through “price increases” and “volume growth.” Does this indicate that profits will substantially improve this year? “Although price hikes and volume increases are positive signals, Zhipu’s losses in 2026 are likely to ‘improve’ but not ‘turn profitable.’ To truly break even, it may still require the MaaS flywheel to operate more fully and the entire Token economy system to mature,” said Zhang Xiaorong, director of the Deep Technology Research Institute, to Huaxia Times.

Another key point is that once a certain business becomes a core revenue source for a company, its sustainability directly affects future growth prospects. So, how long can the Token economy benefits last?

Zhang Peng believes that the growth of API business is a structural, long-term trend. The industry has been developing for years, seeking a simple, economically powerful business model.

He emphasized that API is “transforming AI infrastructure capabilities into resources for economic operation,” not a one-time dividend. Today, AI capabilities have shifted from being usable and fun to solving increasingly complex and important problems, turning Token API calls and consumption into real economic value. “Moreover, the emergence of new applications like OpenClaw and the expected device-level native intelligence will lead to exponential growth in API and Token consumption.”

Zhang Xiaorong also pointed out that Token economy is forming and may become a reality faster than expected. Although currently small-scale, Token could become a core business model for large model vendors. It may not be “the biggest business,” but it can be regarded as the “new oil” of the AI era.

In Zhipu’s 2025 financial report, the concept of TAC (Token Architecture Capability) was also introduced, defined as TAC = intelligent call volume × intelligent quality × economic conversion efficiency. The company stated that in the era of large models, once they possess long-term task execution closed-loop capabilities, the core competitiveness will be reshaped into TAC. Zhipu’s goal is to become the infrastructure that enhances society’s TAC, turning every drop of Token into deliverable economic growth.

Editor: Huang Xingli Chief Editor: Han Feng