China's artificial intelligence (AI) company Z.ai increased revenue 400 percent in the first half from a year earlier, extending rapid growth. Cloud-based deployment and its API business led results. Losses narrowed slightly, but spending pressure from research and development and AI inference costs remains a challenge to profitability.

Hong Kong's South China Morning Post (SCMP) reported on July 31 local time that Z.ai's first-half revenue, for the period ended June 30, totalled 953.89 million yuan. That was a 400 percent increase from a year earlier.

Bloomberg-compiled market forecasts projected Z.ai's annual revenue would increase 514 percent this year from 724.30 million yuan last year.

Losses improved slightly. Total first-half loss was 2.07 billion yuan, down 12.1 percent from a year earlier. Adjusted net loss, however, rose 12.1 percent to 1.96 billion yuan. R&D expenses also increased 33.6 percent to 2.13 billion yuan over the same period.

Z.ai is continuing aggressive investment to increase computing power and improve base model performance. The most striking part of the results was growth in its cloud-based deployment business. Revenue in that segment surged 2,736 percent to 825.00 million yuan from 29.00 million yuan in the first half last year. Its share of total revenue also expanded sharply to 86.5 percent from 15.2 percent.

By contrast, on-premises deployment revenue fell 20.5 percent to 128.70 million yuan. That is seen as meaning Z.ai's business structure is moving quickly from customised deployments for individual companies to cloud APIs and subscription-based agent products.

Z.ai is expanding its business beyond being a service provider that simply supplies AI models. It is broadening into an agent platform spanning the Zcode coding tool, AutoGLM and consumer products.

Rising revenue is not directly translating into improved profitability. First-half gross profit increased 163.7 percent to 252.00 million yuan, but gross margin fell to 26.4 percent from 50 percent. Cloud inference generates ongoing computing costs in operating services, giving it a relatively lower margin structure than on-premises deployment.

Z.ai is also accelerating the pace of AI model launches alongside performance growth. It released GLM-5-Turbo and GLM-5.1 in March, followed by GLM-5.2 in June. It unveiled its latest flagship model, GLM-5.3, in August.

GLM-5.3 scored 60 points on the Intelligence Index of AI model evaluator Artificial Analysis. That is on par with Moonshot AI's Kimi K3, which is regarded as a leading Chinese open-weight model.

Z.ai said GLM-5.3 improved its coding and long-term task execution abilities through post-training alone, without changing the base model itself.

Last week, it also unveiled the low-cost multimodal model GLM-5.3-Flash. The model was anonymously tested as "Oaks Alpha" on OpenRouter and OpenCode, and test traffic was processed on a large cluster of Chinese-made AI chips, the company said.

Market reaction is mixed. Shares of Z.ai, listed in Hong Kong, ended July 31 up 9.63 percent at 1,195 Hong Kong dollars ahead of the earnings announcement. The stock is still about 60 percent below the record high of 2,980 Hong Kong dollars set in June.

Z.ai's market capitalisation at the time briefly approached 1 trillion Hong Kong dollars. Market expectations for Chinese large language model (LLM) companies remain high. But as demand for AI services grows, the burden of computing resources and inference costs also rises, making it Z.ai's core task to link growth to actual profitability.

Z.ai has previously acknowledged that service quality was affected by a lack of computing resources during peak-demand periods. It said customers are accepting price increases due to model performance.

Ultimately, the task ahead for Z.ai is expected to be converting rapidly rising cloud and API demand into stable recurring revenue while effectively controlling inference costs and computing supply issues.