Zhipu Analyst Meeting: Year-end ARR guidance raised by 25% to $3 billion; cloud vendor revenue sharing to be recognized starting in October.

Zhipu is accelerating its computing power expansion and commercialization efforts. On September 16th, according to information from Zhipu's analyst conference, the company completed a $5 billion equity and debt financing round on September 11th and raised its year-end 2026 annual recurring revenue (ARR) guidance from $2.4 billion to $3 billion, an increase of approximately 25%. As of the first half of September, the company's total monthly ARR across all businesses had reached $1.8 billion.

This round of financing will primarily be used for expanding computing power. The company stated that after the financing, its total computing power reserves have entered the ranks of the top independent large-scale model vendors in China, and computing power supply will no longer be a major constraint on revenue growth in the short term . Previously, due to computing power limitations, the Coding Plan had been subject to sales restrictions. After being fully reopened from the end of July to the beginning of August, sales increased more than 15 times.

In terms of commercialization, the company has signed revenue-sharing agreements with several leading domestic and international cloud service providers, with related revenue expected to be recognized starting in October 2026. Meanwhile, businesses such as Coding, cloud services, and CoWork are all progressing. Z Code has surpassed 2 million users, and within a month of the release of GLM-5.3, industry orders for the to-work scenario exceeded 1 billion yuan.

At the meeting, the company also disclosed its computing power investment, cooperation with cloud vendors, and progress on the CoWork business, and introduced its preparations for an A-share listing. It should be noted that the $3 billion ARR is an annualized order revenue guidance from management's perspective and is not entirely equivalent to final accounting revenue. Actual revenue recognition still depends on business performance and relevant accounting standards.

With the expansion of computing power, coding business has resumed growth.

Over the past year, computing power supply has been a significant constraint on the company's business expansion. After the release of GLM-5.5 in February 2026, the demand for models increased nearly tenfold compared to the beginning of the year. The company's original computing power reserves were rapidly depleted, and its flagship product, Coding Plan, was once forced to limit sales, providing services only through a daily quota of 500 slots.

From February to March, the company improved computing power utilization efficiency through an "all in infrastructure" strategy, increasing overall efficiency by approximately 2.3 times. However, as model call demands continued to grow, computing power supply still could not fully meet the demand. After the release of GLM-5.2 in June, overseas call demands increased further. At that time, the company's original equipment manufacturer (OEM) could only handle about 10% of the call traffic, with the remaining demand mainly handled by third-party deployments of open-source models.

Following its $4 billion refinancing in July, the company began expanding its computing power footprint. From late July to early August, the Coding Plan, which had been suspended for six months, resumed full operation, resulting in a more than 15-fold increase in sales. The company stated that with this latest $5 billion financing, computing power supply will no longer be a major constraint on revenue growth in the short term.

According to calculations disclosed at the company's press conference, the $5 billion financing corresponds to approximately 30 billion RMB in computing power investment, which can purchase nearly 100,000 computing cards, with 40% allocated to training and R&D and 60% to inference operations. The company anticipates that with the increase in computing power reserves, businesses such as Coding, which were previously constrained by supply, can further expand their service scale.

Cloud vendor revenue sharing to be recognized starting in October

In terms of commercialization, the company has signed revenue sharing agreements with several leading domestic and international cloud service providers, deploying the GLM series of open-source models on cloud platforms in the form of managed APIs. The two parties will share revenue according to the agreed ratio, and the relevant revenue is expected to be recognized starting in October 2026.

The aforementioned collaboration aligns with the commercial arrangements following the open-sourcing of GLM-5.3. The company previously disclosed that cloud service platforms operating MaaS businesses with revenue exceeding $10 billion require security reviews and commercial licenses to use its cutting-edge models. Currently, the company has not disclosed specific partner names, individual platform revenue figures, or revenue sharing percentages.

The company stated that partnering with cloud vendors can expand model distribution channels, lower the barrier to entry for users, and reduce the direct impact of some calls on its own computing resources. As related services are gradually launched, revenue sharing with cloud vendors will become one of the new revenue sources in the fourth quarter.

CoWork orders exceed 1 billion yuan, with cybersecurity being implemented first.

CoWork was another key business disclosed by the company at this conference. The company stated that within one month of the release of GLM-5.3, industry orders for the to-work scenario exceeded 1 billion yuan, and the business has begun to enter the production-level deployment stage.

Cybersecurity is one of the earliest vertical applications to achieve large-scale deployment. Currently, over 100 cybersecurity companies have integrated the GLM model into their security products and business processes. At the product level, GLM-5.3 primarily handles tasks such as vulnerability discovery and complex code auditing, while GLM-5.3 Flash covers high-frequency scenarios such as batch scanning and alert analysis.

The company believes that cybersecurity scenarios are characterized by relatively easy verification of results and relatively easy quantification of value, making them more suitable for integrating AI models into actual business processes. At the meeting, the company disclosed that GLM-5.3 achieved a score of approximately 85% on the Cyber Games, and more than 1,500 high-risk vulnerabilities were discovered during the red team testing period.

The company stated that CoWork's business model will be further expanded to professional scenarios such as finance, law, industrial manufacturing, life sciences, and mathematics. Meanwhile, the company completed its A-share IPO pre-listing guidance and acceptance process in June and is currently in the pre-application acceptance stage. It has also clarified that after September 30th and before its listing on the Science and Technology Innovation Board, it will no longer conduct equity financing from new investors.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.