It is not enough to have the "Six Little Dragons". Hangzhou takes action and invests in Zhipu.

Written by Zhou Xinyu

Edited by Su Jianxun

"Intelligent Emergence" has learned that on March 1, Zhipu announced that it has obtained a strategic financing of over 1 billion yuan. It is understood that the main investors in this round of financing are funds from Hangzhou, including Hangzhou Chengtou Industrial Fund and Shangcheng Capital.

It has only been three months since the official announcement of Zhipu's last round of financing - in December 2024, Zhipu received financing from several strategic investors and state-owned capital, raising a "winter survival fund" of 3 billion yuan for itself.

From the perspective of its establishment time, as the eldest brother of the "AI Six Tigers" in this round, Zhipu is not only aggressive in attracting funds, but also acts quickly in "gold panning".

According to the exclusive news of "Intelligent Emergence", Zhipu recently won an overseas order from a certain "Belt and Road" country, worth tens of millions of US dollars. At the same time, with the signing of large orders such as Samsung, Zhipu's income has exceeded 100 million yuan in less than a month after the festival.

Under the pressure of DeepSeek's open source, Zhipu's ability to achieve such results is equivalent to verifying the model's hematopoietic ability. According to "Intelligent Emergence", after the beginning of the year, the paid growth of Zhipu's large model API platform has also exceeded 30%.

Accelerating commercialization is the consensus of the AI industry in 2025. It is undeniable that the model price war and investment and marketing war since 2024 have disrupted the pace of "gold panning" of AI manufacturers to a certain extent.

In 2025, the open source of high-performance models such as DeepSeek V3 and R1 "turned the table", which has led many manufacturers to take this opportunity to re-evaluate the hematopoietic ability of the closed-source model itself. It can be seen that MiniMax has recently contracted its B-end business. And earlier, ZeroOneEverything directly gave up the pre-training of the super-large model and adjusted the algorithm team to Alibaba.

Zhipu's rapid progress in commercialization at the beginning of the year also reflects several aspects of the "gold panning" of AI model manufacturers at present: going overseas and using open source to establish the model's influence have become increasingly important.

The "National Team" Also Needs to Go Overseas for Gold Panning

It is worth noting that Zhipu, which has been active in the domestic government and enterprise market as the "national team", has set its commercial territory overseas for the first time.

The wind vane of the commercialization of new technologies is going overseas, which has become a consensus that has been talked about too much. In addition to a more mature payment awareness and a stronger payment ability, more importantly, in overseas markets, open-source models and closed-source models have a more mature coexistence ecosystem, and open-source models are difficult to monopolize the market of closed-source models.

For example, as of August 2024, the download volume of Meta's open-source model Llama has approached 350 million times, a 10-fold increase compared to the same period in 2023. However, this does not prevent the number of OpenAI's paying enterprise users (2 million) from doubling compared to September 2023.

However, Zhipu's going overseas does not target the mainstream markets in North America and Europe. According to informed sources, unlike previous product exports, the order that Zhipu won from a certain "Belt and Road" country is to jointly build a "national sovereign large model", which is the export of large models.

This also gives some inspiration for other manufacturers to go overseas: It is a strategy to avoid directly competing with European and American model manufacturers such as OpenAI and Anthropic by entering from the marginal and emerging markets with model needs.

The "Intelligent Elephant Going Overseas", an overseas service institution, released the "2024 China New Technology Going Overseas Report", which pointed out that with the growth of digital infrastructure in Southeast Asia, Latin America, and the Middle East region mainly composed of Gulf countries, Chinese enterprises will have more opportunities to go overseas.

More cases show that non-European and American regions are likely to be underestimated markets. Some AI applications focused on markets such as Japan and India have already obtained considerable profits or financing, such as the AI conference transcription tool Notta with an ARR (Annual Recurring Revenue) of tens of millions of US dollars, and the Japanese adult learning platform UMU with a total financing of nearly 50 million US dollars.

At the same time, meeting the geopolitical needs and doing a good job in localization have become the core factors that enterprises need to consider when going overseas.

According to "Intelligent Emergence", the main reason why Zhipu was able to win the overseas order for the "national sovereign large model" is that it meets a core demand: The increasing attention of third-world countries to technology and data sovereignty, and the fact that a safe, autonomous, stable and controllable large model has become a manifestation of the country's hard power.

Fighting the Open Source Battle and Making Money from Closed Source

2025 is not only the year of the commercialization test, but also the year of model open source. With the open source of DeepSeek R1, manufacturers such as Baichuan Intelligence, StepStar, and Baidu have all joined the open source team.

"Intelligent Emergence" has learned that Zhipu will soon release new models, including the base, reasoning, multimodal, and Agent. And these models will all be open source.

"Open source is not a simple follow." An industry insider told "Intelligent Emergence", "Open source, like ChatBot, is a display of the model's ability and brand."

This means that by launching the open source battle, model manufacturers can compete with DeepSeek for some customers and developers with technical capabilities and cultivate their own innovation ecosystem.

At the same time, today's manufacturers, in the face of DeepSeek's "table-turning", need to think clearly about a question: What kind of market is being eaten up by DeepSeek's open source?

The most obvious is that currently DeepSeek is having an impact on the B-end business of model manufacturers. Many enterprise customers with technical capabilities are turning to open-source models with lower costs.

But some industry insiders believe that This part of customers turning to open source is not the target customer group of model enterprises in the first place.

"Developers and enterprises that can download open source code to do development themselves already have relatively strong IT capabilities." The above-mentioned industry insider said, "But in the long run, there is an 80% possibility that these people or enterprises will train their own models in the future and will not become customers of closed-source models."

This also forces closed-source model manufacturers to look forward and think about how to focus on seizing the remaining B-end customers who are willing to pay for the model and launching AI products that the open source ecosystem is difficult to provide.

It can be seen that the "AI Six Tigers" have divided into two distinct factions in answering this question. One faction is represented by MiniMax and Dark Side of the Moon, contracting the B-end and concentrating resources on the already popular star C-end products.

The other faction is represented by Zhipu and StepStar, betting on multimodal and Agent - the core reason is that there are still only a few manufacturers that master multimodal technology, and at the same time, the landing space of Agent in vertical scenarios is still very large, and open source cannot eat up all the markets.

On the Ecological Open Day on February 21, 2025, StepStar brought in partners such as Luckin Coffee and OPPO to support the research and development of Agent. And Zhipu's recent cooperation with Samsung is based on the Agent product AutoGLM released in October 2024, which can realize functions such as AI real-time voice and video calls, as well as visual understanding and intelligent system function calls.

"With the open source of more high-performance models, the higher the level of market education, the higher the acceptance and willingness to pay for AI models and products." An industry insider of the "AI Six Tigers" told "Intelligent Emergence", "For manufacturers that independently develop models, the prosperity of the open source ecosystem is a positive factor."

But to seize the opportunity brought by open source, model manufacturers also need to reshape their core competitiveness. The muscle that model manufacturers should show has changed from technical ability to how to turn technology into product power.

Welcome to communicate!

Welcome to follow!