AWS Bedrock Integrates Zhipu GLM-5.3 With Revenue-Sharing Model as Chinese AI Lab Builds Global Distribution

Bedrock Integration and Revenue-Sharing Mechanics

Amazon Web Services added Zhipu AI's GLM-5.3 model to its Bedrock platform on October 5, enabling eligible enterprise customers to invoke the open-weight model directly through AWS infrastructure. Under the agreement, AWS will share revenue with Zhipu based on customer usage volume, rather than charging a flat hosting fee.

GLM-5.3 is accessible through both the OpenAI-compatible Responses API and AWS's native Invoke and Converse APIs, with prompt caching available to reduce latency and input costs on repeated calls. The model supports US and global cross-Region inference profiles, allowing enterprises to route requests while maintaining their existing IAM roles, billing relationships and compliance postures with AWS.

On Bedrock's Standard tier, GLM-5.3 is priced at $1.68 per million input tokens and $5.28 per million output tokens for global cross-region inference, with cache reads at $0.312 per million tokens. Priority tier pricing carries a 75% premium, while Flex and Batch tiers offer a 50% discount.

Market Reaction and Strategic Significance

Shares of Zhipu (2513.HK), the first publicly listed large model company globally, rose more than 7% intraday on October 6 following the announcement. By mid-morning trading, the stock was up approximately 5.6% at HK$702, valuing the company at roughly HK$342 billion.

For Zhipu, the Bedrock listing represents more than incremental API revenue. It provides a front door inside the cloud platform most American enterprises already trust with their data, sidestepping procurement friction and compliance concerns that typically slow adoption of new AI vendors. Enterprise teams already running Claude or Llama workloads on Bedrock can now call GLM-5.3 without establishing a separate vendor relationship with a Chinese company.

Multi-Cloud Distribution Strategy

Zhipu has been systematically building a domestic and international cloud distribution network. In China, the company has established similar revenue-sharing arrangements with Alibaba Cloud's Bailian platform and Huawei Cloud, which has also listed GLM-5.3. Zhipu is additionally pursuing comparable partnerships with multiple other overseas cloud providers beyond AWS.

The company has previously stated that it aims to build a commercially scalable path through overseas cloud platform revenue sharing. This strategy mirrors the model pioneered by Anthropic, whose Claude models generate substantial revenue through cloud marketplace distribution. According to market reports cited by Chinese financial media, approximately 47% of Anthropic's estimated $4.6 billion in 2025 revenue came through Amazon and Google cloud marketplaces, with the two platforms collecting roughly $351 million in combined distribution fees.

Challenges for Open-Weight Models on Cloud Platforms

The open-weight nature of GLM-5.3 creates an inherent tension between ecosystem growth and revenue capture. Enterprises can obtain the model weights and self-deploy rather than paying for managed API access through Bedrock, potentially diverting cloud-based revenue. The more openly available a model is, the faster its ecosystem expands, but the greater the risk of unpaid self-hosting.

Additionally, overseas enterprises maintain strict requirements around data security, regulatory compliance, model liability and supply chain stability. Whether Zhipu can convert developer interest into sustained enterprise usage and recurring revenue through cloud channels remains the central question for this distribution model's long-term viability.