Latest Search
Quote
| Back Zoom + Zoom - | ||||
|
<Research> CMSI: KNOWLEDGE ATLAS (02513.HK) Breakthrough in Inference Efficiency Further Enhances Commercialization Prospects; Rated Overweight
Recommend 9 Positive 13 Negative 6 |
||||
|
||||
|
CMSI issued a report stating that KNOWLEDGE ATLAS (02513.HK) has achieved breakthroughs in inference efficiency, further enhancing its commercialization prospects. The broker assigned an Overweight rating and is reviewing its TP. Last Friday (22nd), the company launched the GLM-5.1 high-speed API version, with output speed reaching 400 tokens/s. Model output speed improved significantly, setting a new global upper limit for large-model vendors API speed and, for the first time among domestic large models, bringing both flagship capabilities and ultra-low latency into production environments simultaneously. The report noted that the new architecture reduces inference costs. Last Thursday (21st), the company released ZCube, its next-generation inference network architecture. A thousand-GPU production cluster has been deployed, with throughput increasing by more than 15%, TTFT P99 decreasing by 40.6%, and network hardware costs reduced by 33%. CMSI stated that under Agent scenarios involving dozens of invocation rounds, speed differences are exponentially magnified. The high-speed version enables product forms previously unachievable due to latency constraints, such as real-time collaboration, 3D interactive modeling, and instant tool generation. Inference efficiency is the core lever of the MaaS business model: the combination of 400 TPS and ZCube (throughput up 15%, network costs down 33%) optimizes both ends of the flywheelenhancing intelligent supply rates while lowering unit token production costs. If the high-speed version is fully rolled out, it is expected to further lift ARR expectations and margin improvement prospects. The broker remains positive on KNOWLEDGE ATLASs strategy of pursuing the upper bound of model intelligence as its core, continuously strengthening technological barriers and commercialization closed-loop capabilities in programming, and enhancing model pricing power. Key future catalysts include: (1) inclusion in HSI/HSTECH in early June and subsequent inclusion in Southbound Stock Connect; (2) continued iteration of the GLM model; (3) ARR growth of the open platform and margin improvement; (4) progress of its STAR Market listing in 2H26; and (5) increased supply and adaptation progress of domestic chips. Major risks include: (1) potential shareholding reductions by major shareholders after lock-up expiry; (2) model iteration falling short of expectations and intensifying competition; and (3) higher-than-expected investment. (da/a) Auto-translated by AI This article was automatically translated by AI, the original language version should be considered the authoritative version. AASTOCKS.com Limited does not guarantee its accuracy or completeness and accepts no liability for any damages or losses arising from the use of this translation. More Details
AASTOCKS Financial News |
||||
Copied to Clipboard
 |
||||
