Zhipu's GLM-5.3 Goes Live on Amazon Bedrock With a Usage-Based Revenue Share — and the Stock Surges 7%

Zhipu's overseas monetization machine just got its biggest lever yet. Amazon Bedrock, the model-hosting platform inside Amazon Web Services, has officially integrated the Chinese lab's GLM-5.3 — and rather than a plain listing fee, AWS will pay Zhipu a revenue share based on model call volume, according to an exclusive report from Chinese financial outlet Stardaily News (科创板日报) confirmed by multiple domestic outlets. Zhipu's Hong Kong-listed shares (02513.HK) surged as much as 7% in early Tuesday trading and were still up about 5.4% at around HK$701 by mid-morning.

The structure matters more than the headline. A usage-based revenue share turns Zhipu from a vendor selling licenses and custom deployments into something closer to infrastructure: every inference a customer runs on Bedrock — wherever that customer is, from a Saudi data center to a Southeast Asian e-commerce backend — generates dollar-denominated income for the lab. Zhipu has previously described overseas cloud revenue sharing as a commercial path with real scale potential, and said it has recently landed similar arrangements with multiple overseas cloud providers. Specific terms — the split ratio and settlement mechanics — have not been disclosed.

The deal also completes a two-sided distribution network. Domestically, Zhipu has signed comparable revenue-sharing agreements with Alibaba Cloud's Bailian platform, and Huawei Cloud has listed GLM-5.3 with a similar cooperation agreement reached in principle. Overseas, Bedrock now becomes the flagship channel. The result is what Zhipu positions as a domestically-and-internationally connected revenue-share system — a distribution strategy no other Chinese frontier lab has assembled at this breadth.

The financial backdrop explains the urgency. Zhipu's H1 2026 results show total revenue of RMB 954 million, up 399.7% year over year — but with an attributable net loss of RMB 2.071 billion, narrowing just 12.1%. The company raised roughly $5 billion in September through a share placement of about $2 billion and a convertible bond issuance of about $3 billion, earmarked for next-generation GLM base models, its "Fully Self Training" system, and large-scale training and inference infrastructure. Distribution deals that convert existing models into recurring overseas revenue are exactly what a cash-burning lab needs between fundraising cycles.

The market reaction is notable for its consistency. When GLM-5.3 landed in the Cursor coding editor on Sunday with a 42.6% CursorBench score — the top open-weight result, though Cursor's own benchmark — the stock jumped 6%. Tuesday's Bedrock news pushed it further. Investors are reading distribution wins, not benchmark claims, as the signal that Zhipu's commercial model is working.

For AWS, the integration fits a pattern: Bedrock has been aggressively adding third-party and open-weight models — from xAI's Grok to DeepSeek-family and now GLM — positioning itself as the neutral hosting layer for enterprises that want model choice rather than a single lab's stack. A Chinese frontier lab on Bedrock is also strategically awkward for some buyers in Western regulated markets, but for the cost-sensitive, high-volume workloads that dominate emerging-market AI adoption, price-performance is usually the deciding factor.

One caveat belongs on the record: the revenue-share arrangement comes from Stardaily News' exclusive reporting, and while the Bedrock integration itself is described as officially announced, AWS's own announcement text and the commercial terms have not been independently reviewed. If the usage-based model performs as advertised, though, the interesting question becomes which overseas cloud signs next — and whether Zhipu's "distribution over exclusivity" playbook becomes the template every Chinese AI lab exiting the domestic market ends up copying.