Chinese artificial intelligence firm Z.ai reported a significant increase in revenue for the first half of 2024, driven largely by rapid expansion in its cloud-based services. The company, also known domestically as Zhipu AI, posted a 400 percent growth in revenue for the six months ending June 30, reaching 953.89 million yuan (approximately HK$1.1 billion). This surge helped reduce overall losses despite heightened research and development expenditures.

Analysts polled by Bloomberg project full-year sales for Z.ai to climb 514 percent to 4.45 billion yuan. The company's annual recurring revenue (ARR), which estimates subscription income over a 12-month period, reached US$1.6 billion by the end of August. For comparison, MiniMax, a Shanghai-based competitor and China’s other publicly traded large language model developer, reported an ARR of US$800 million in the same month.

Despite the revenue growth, Z.ai’s total loss for the first half narrowed by 12.1 percent to 2.07 billion yuan. However, the company’s adjusted net loss rose by 12.1 percent to 1.96 billion yuan. Research and development costs increased by 33.6 percent to 2.16 billion yuan, reflecting substantial ongoing investments in computing infrastructure and enhancements to its base AI models.

The company’s shares on the Hong Kong Stock Exchange rose 9.63 percent to close at HK$1.195 ahead of the earnings announcement. Yet, they remain approximately 60 percent below the all-time high of HK$2.98 recorded in June, when Z.ai’s market capitalization briefly neared HK$1 trillion. Both Z.ai and MiniMax experienced sharp valuation increases following their initial public offerings in January but have since retreated from those peaks.

The bulk of Z.ai’s revenue growth stemmed from its cloud-based deployment services, which soared 2,736 percent year-on-year to 825 million yuan, up sharply from 29 million yuan in the same period last year. This segment now represents 86.5 percent of total revenue, compared with 15.2 percent a year earlier, marking a strategic shift from customized, on-premises solutions to cloud application programming interfaces (APIs) and subscription-based agent products. Meanwhile, revenue from on-premises deployment declined by 20.5 percent to 128.7 million yuan.

Gross profit grew 163.7 percent to 252 million yuan, with a gross margin of 26.4 percent—down from 50 percent the previous year—largely due to the higher computational costs associated with cloud inference services compared to on-premises offerings.

To accelerate its transition toward recurring revenue models, Z.ai released several new versions of its large language models this year. The company launched GLM-5-Turbo in March, followed by GLM-5.1 in April, GLM-5.2 in June, and its latest flagship version, GLM-5.3, in August. GLM-5.3 achieved a score of 60 on Artificial Analysis’ Intelligence Index, comparable to Moonshot AI’s Kimi K3 model, widely regarded as a benchmark in China’s open-weight models. This version improved coding and long-horizon agent capabilities through post-training refinements, without modifying the underlying base model.

Additionally, Z.ai introduced GLM-5.3-Flash last week, a lower-cost alternative designed to enhance accessibility. These efforts reflect the company’s broader ambition to evolve from a model-as-a-service provider into a comprehensive agent platform, encompassing its ZCode coding tool, AutoGLM technology, and consumer-facing products.

The Z.ai GLM platform’s registered user base surpassed 7.4 million in August, a 144 percent increase since the start of the year, underscoring strong market demand amid China’s increasingly competitive AI sector.