First Release! ZStack AIOS Supports DeepSeek V3/R1/ Janus Pro, Various CPU/GPU for Private Deployme
On February 2, 2025, in response to the growing demand for AI inference and enterprise-level AI application private deployment scenarios (Private AI), ZStack announced that its AI Infra platform, ZStack AIOS, fully supports the private deployment of three models: DeepSeek V3/R1/ Janus Pro. It can adapt to various CPUs/GPUs from Hygon, Ascend, NVIDIA, Intel, etc., helping further implementation of enterprise-level AI applications.
It is reported that following DeepSeek’s launch and open-sourcing of DeepSeek V3/R1/Janus Pro in December 2024, global public cloud platforms such as AWS, Azure, and Huawei Cloud have successively announced support for DeepSeek R1 or R1/V3. As an enterprise-level private AI Infra platform supporting DeepSeek, ZStack AIOS will fully leverage the open-source models and the cost-effective, high-performance characteristics of DeepSeek to further advance the commercial process of enterprise-level AI:
Full Support for Three DeepSeek Models to Meet Diverse Enterprise AI Needs
ZStack AIOS, the AI Infra platform, provides various essential tools and components for model development and application at the model layer, supporting lifecycle management of both open-source and proprietary AI models. ZStack AIOS initially supports DeepSeek V3/R1/Janus Pro. V3 is suitable for general natural language processing tasks, R1 focuses on complex inference tasks, and Janus Pro excels in multimodal understanding and generation, meeting the different AI needs of enterprises.
Support for Multiple CPUs/GPUs, Adapting to the Diverse Computational Resources of Enterprise Data Centers
ZStack AIOS provides computing, storage, networking, security, and other fundamental resources and services at the computational layer. It can support the private deployment of DeepSeek on a variety of CPU/GPU resources such as Haiguang, Ascend, NVIDIA, and Intel, adapting to the diverse computing resources of enterprise data centers.
ZStack AIOS has an intelligent heterogeneous scheduling engine that automatically matches hardware features. It supports CUDA, ROCm, CANN, and other architectures, as well as GPU-less testing. The CPU deployment of the DeepSeek-R1-7B lightweight model achieves a usable performance of 9.26 tokens/s on a 16-core cloud host.
Providing Flexibility and Customization for Enterprises, Building the Next Generation of Digital Intelligence Platforms
ZStack AIOS can deploy DeepSeek models in a private or hybrid cloud environment based on enterprise needs, ensuring data security and privacy protection. It also offers elastic scaling of bare metal, virtual machine, and container computing resources, supporting integration with various hardware and software, allowing enterprises to easily incorporate AI capabilities into their existing systems. This flexibility and customization meet the diverse AI application needs of enterprise users.
ZStack AIOS has technologies for resource optimization, such as multi-GPU concurrent inference to enhance the availability of small memory GPUs and reduce idle time. GPU partitioning technology divides a single GPU’s computing power and memory, improving the utilization of large memory GPUs. The model quantization technology allows the platform to quantize models, significantly enhancing AI efficiency in combination with DeepSeek’s low-cost and high-performance features.
The AI Infra platform is a key engine platform for enterprises to accelerate the unleashing of AI productivity, focusing on enterprise-level AI application private deployment scenarios (Private AI). It supports the development, deployment, operation, and management of artificial intelligence applications with a series of foundational tools and software platforms, featuring computational management, model management, and application management capabilities. According to CCID Consulting, 2025 is the inaugural year for the application of China’s AI Infra platforms.
The surge in AI inference computational demand has spurred new needs for enterprise-level intelligent computational resource management; AI applications are accelerating penetration into enterprise-level scenarios, and model toolchains and operational management components help lower the threshold for AI applications; data privacy and security drive the private deployment of AI applications, and the new generation of enterprise digital transformation bases AI Infra platforms are showing a rapid development trend. In January 2025, CCID Consulting released the “2025 China AI Infra Platform Market Development Research Report,” predicting that the AI Infra platform will reach 1.94 billion yuan and 3.61 billion yuan in 2024 and 2025, respectively, with a year-over-year growth exceeding 86% in 2025.
