Morgan Stanley has issued a research report indicating that Z.AI (02513) achieved an annualized recurring revenue (ARR) of $1.8 billion by mid-September, while also raising its year-end ARR guidance from $2.4 billion to $3 billion. The firm reaffirmed its "Overweight" rating on the stock with a target price of HK$1,800.

The bank noted that the management framework shows approximately $5 billion in funding could support around 94,000 GPUs, with roughly 40% allocated for training and 60% for inference purposes. The approximately 56,000 inference GPUs imply a theoretical annual revenue ceiling of about RMB 41 billion, which does not yet account for utilization rates and actual pricing realization, with theoretical inference gross margins potentially reaching as high as 80%.

Easing compute bottlenecks is expected to unlock the next phase of growth, while cloud partnerships could improve capital efficiency and global distribution, with revenue sharing contributions anticipated to begin in October. On the research and development front, the training allocation could support average annual training expenditures of approximately RMB 3 billion over four years, sufficient to underpin multiple rounds of GLM-5.3-class or larger model training.

Furthermore, the bank pointed out that Cowork is expanding beyond coding, with cybersecurity emerging as the first scaled vertical application. More than 100 cybersecurity companies have already integrated GLM models, and orders for GLM-5.3 have surpassed RMB 1 billion within one month of its launch.

The upward revision in ARR is primarily driven by three key engines: coding, cloud partnerships, and Cowork. The critical discussion has shifted from whether demand exists to how quickly incremental compute capacity, partner distribution channels, and vertical workflow monetization can translate into recognized revenue and sustainable gross margins.