Recently, the '2026 Outstanding Cases for High-Quality Development of Computing Infrastructure' selection results, organized by the Cloud Data Committee of the China Association of Communications Enterprises (CACE-CDPC), were officially announced. The project 'Local Private Deployment and Public Health Fine-Tuning of the DeepSeek Large Model at a CDC Based on the GenstorAIge AI Solution,' submitted by GenstorAIge Technology (Beijing) Co., Ltd., stood out from numerous entries and was successfully selected as a 'New Technology Application Case.'

GenstorAIge VP of Products, Xu Wei (5th from left), receives the award on stage

On May 27, GenstorAIge will attend DCIC 2026 (the 15th Data Center Industry Development Conference). GenstorAIge VP of Products, Xu Wei, accepted the certificate of honor on behalf of GenstorAIge. This award signifies high recognition from industry authorities for the technical roadmap and implementation capabilities of GenstorAIge's 'GenstorAIge AI Solution.'

Technological Breakthrough:

Compute-Storage Convergence Architecture Breaks the Memory Wall

In traditional AI training architectures, GPU memory capacity is the main bottleneck for local fine-tuning of large models. Taking a 70B parameter model as an example, conventional full-parameter fine-tuning requires 8 to 16 high-end GPUs, with hardware investment reaching millions of yuan, and imposes stringent requirements on data center space and power capacity.

Simultaneously, CDC data involves a large amount of citizen health information and national security-sensitive information, which legally cannot be uploaded to public clouds and must be deployed locally and privately.

GenstorAIge VP of Products, Xu Wei (4th from left), receives the award on stage

The 'GenstorAIge AI Training and Inference Integrated Machine' utilizes a self-developed GenstorAIge AI training and inference solution, deploying only 2 GPUs + 2 GenstorAIge AI accelerator cards. Through middleware, it virtualizes enterprise-grade NAND flash into GPU-addressable extended memory, enabling automatic offloading and transparent access to data such as model parameters, optimizer states, and activation values.

Key Results:

Consolidated Computing Power, Localized Security

The project was implemented at a provincial-level Center for Disease Control and Prevention, covering core operations such as infectious disease monitoring and early warning, epidemiological investigation, and emergency response decision-making.

Specific results are as follows:

• Hardware Cost: Under optimal conditions, the overall investment was reduced by approximately 80% compared to traditional solutions (from the million-yuan level to the hundred-thousand-yuan level).

• GPU Efficiency: The number of GPUs required was reduced by 75% to 87.5% (compressed from 8-16 cards to 2 cards).

• Privacy Protection: With local device deployment, both data security and user privacy are well protected.

In terms of data security, the full-stack equipment is deployed within the customer's intranet. Data governance, model training, and inference services are all completed in a physically isolated environment, eliminating public cloud transmission links.

The GenstorAIge AI middleware employs encrypted address mapping and access auditing, meeting the highest requirements of the Data Security Law, the Personal Information Protection Law, and the health industry'sdata classification and grading standards.

Industry Value:

From CDC to More Data-Sensitive Fields

This case provides a lightweight, low-cost, high-security computing infrastructure model for the public health system, which has been verified as replicable and scalable to prefecture-level CDCs, health inspection institutes, and primary healthcare institutions.

Simultaneously, its technical pathway is equally applicable to industries where data cannot leave their domain, such as finance, government affairs, and healthcare, effectively facilitating the in-depth vertical implementation of the 'East Data, West Computing' project.

Future Outlook

GenstorAIge will continue to optimize the architecture, promote the adaptation of the GenstorAIge AI Solution with more domestic GPUs and domestic NAND storage, and continuously iterate on multi-node collaboration and linear acceleration, building a more consolidated, secure, and efficient data foundation for the intelligent world.