China’s large-model companies are beginning to show what customers will pay for, even if neither has solved the profitability question. A September 3 DIGITIMES report compared the first-half results of MiniMax and Z.ai, the overseas brand of Zhipu. The common thread is that APIs and enterprise-facing services are becoming central revenue engines, while research, computing, and expansion continue to produce substantial losses.
MiniMax’s own first-half disclosure reported total revenue of US$116.6 million, up 283.1% from a year earlier. Revenue from its open platform and other AI-based enterprise services reached US$73.9 million, up 703.1% year on year and accounting for 63.4% of total revenue. Zhipu, according to a 36Kr/Yiou comparison, reported 954 million yuan in first-half revenue, with 825 million yuan from its open platform and API business.
The revenue lines suggest a shift in how the two companies reach customers. Rather than relying mainly on one-off project deployments or consumer subscriptions, both are increasingly selling capacity that developers and businesses can call as they need it. That does not make the underlying models interchangeable. It does show that the commercial race is moving toward recurring infrastructure-style usage.
API Revenue Becomes the Largest Line at Both Companies
MiniMax’s first-half results give a clear picture of the change. The company said its open platform and other enterprise AI services brought in US$73.9 million, compared with US$9.2 million in the comparable 2025 period. The result represents 63.4% of MiniMax’s total revenue, up from 30.3% a year earlier. MiniMax attributed the increase to growth in paying individual and enterprise users, higher API-call volumes, and adoption of its Token Plan.
The company’s AI-native products also generated US$42.6 million in the first half, up 100.9% year on year. That means consumer-facing products still matter, but the faster growth came from platform and enterprise services. MiniMax’s statement linked its revenue growth to model inference demand and a larger global customer and user base. It did not break out revenue by every individual model or product.
Zhipu’s figures point to a similar endpoint from a different starting point. The 36Kr/Yiou report said its open-platform and API business reached 825 million yuan in the first half, up 2,735.7% from a low base. That segment made up 86.5% of Zhipu’s revenue, compared with 15.2% in the same period a year earlier. The report also said revenue from localized deployment declined to 129 million yuan from 161 million yuan, while the share of revenue tied to enterprise general large models fell.
The comparison is useful because it explains more than a headline growth rate. MiniMax built a larger platform-and-enterprise line alongside AI-native products. Zhipu, according to the report, moved away from a business weighted toward project-style deployment and toward API use. Both companies are now more exposed to the economics of inference, usage volumes, and developer demand.
Higher Revenue Has Not Yet Eliminated Large Losses
Revenue growth is only one part of the first-half picture. MiniMax reported an adjusted net loss of US$293.0 million, compared with US$138.7 million a year earlier. Its research and development expense reached US$296.9 million, up 138.8% year on year. The company said the higher R&D expense was driven largely by cloud-service costs connected to training and by model iteration and multimodal development.
Zhipu reported a first-half net loss of 2.07 billion yuan, according to 36Kr/Yiou. The same report said Zhipu’s R&D expenditure was 2.13 billion yuan. These amounts are not directly comparable with MiniMax’s figures without making currency conversions or accounting adjustments that the sources do not standardize. The broad conclusion is more reliable: both groups are spending heavily to improve models and serve a larger volume of inference requests.
MiniMax’s results also showed gross profit of US$20.8 million and a gross margin of 17.9%, compared with 12.1% a year earlier. It said improving infrastructure efficiency helped drive the margin change. The company’s reported loss shows that improved gross margin does not automatically offset research, administrative, and other costs during a period of aggressive development.
The tradeoff is visible elsewhere in the company’s strategy. MiniMax recently raised its Alibaba Cloud purchase cap to US$1.2 billion, illustrating the scale of computing commitments that can accompany a rapidly expanding model business. The company needs capacity to supply developers and enterprise users, but that same capacity is a significant input cost.
Usage-Based AI Becomes the Commercial Test
API revenue has an intuitive appeal for AI developers. Customers can use a model through their own applications, pay according to usage, and avoid building or operating a full model stack. For the provider, recurring calls can create a more scalable business than custom delivery work. But the model works only if revenue per request, infrastructure efficiency, and customer retention improve enough to cover training and inference costs.
The MiniMax and Zhipu results show why investors are watching API demand closely. The companies’ top-line revenue is growing quickly, and the share tied to open platforms or enterprise services is rising. That supports the idea that businesses are moving from experimentation toward paid access. At the same time, the size of the reported losses shows that demand has not yet turned either company into a mature, high-margin software business.
The trend also helps explain why Chinese AI companies are competing on more than benchmark scores. They need developer tools, service reliability, model choices, enterprise support, and pricing structures that keep users calling the API after an initial test. Zhipu’s cloud-business revenue story already showed how central this transition has become for the company.
The first-half results do not settle which business model will win. They do establish that the commercial center of gravity is changing. MiniMax and Z.ai are both reporting businesses increasingly tied to model calls and enterprise use, even as their financial statements show the cost of remaining competitive. The next earnings updates will reveal whether the increase in paid usage is enough to narrow losses, or whether the need for ever-more-capable models will keep pushing costs ahead of revenue.