Revenue at Z.ai rose 400% year on year to $141.9 million in the six months to June 30, driven by a surge in cloud-based deployment.
The Shenzhen-based developer of the GLM family of AI models, known domestically as Zhipu AI, narrowed its total loss to $308.0 million.
Its adjusted net loss widened to $291.6 million.
Annual recurring revenue reached $1.6 billion by the end of August.
Gross profit climbed to $37.5 million, but gross margin fell sharply to 26.4% from 50% a year earlier.
Research and development spending rose 33.6% to $316.9 million, reflecting heavy investment in compute and base-model development.
Cloud services drive growth
Cloud-based deployment and application programming interface (API) services, which let outside developers integrate Z.ai's models into their own products, were the main source of growth.
Revenue from that segment jumped 2,736% to $122.8 million, accounting for 86.5% of first-half sales.
On-premise deployment revenue, where clients run the software on their own servers rather than through the cloud, fell to $19.2 million.
The shift marks a significant change in Z.ai's business mix toward the cloud-based model that has become standard among major AI providers.
Rapid product expansion
Z.ai has accelerated its pace of model releases this year.
The company launched GLM-5.3 in August, along with a lower-cost variant called GLM-5.3-Flash, which has been tested under the name Ox Alpha on domestic Chinese chips.
Registered users on the company's GLM platform topped 7.4 million in August.
Compute remains a constraint
An executive said the company was expanding compute capacity "at a very healthy pace."
Z.ai acknowledged that its shift toward cloud inference remains constrained by compute supply and high inference costs, a challenge facing AI developers across the industry as demand for processing power continues to outstrip supply.
Analysts polled expect full-year sales to reach $662.1 million.
