AI Computing Power Over 90% Utilized at Launch! Yawan Supercomputing, a Foxconn Subsidiary, Activates 5MW NVIDIA HGX B300 Cluster and Signs First Long-Term Contracts

Demand for AI computing power is shifting from 'securing GPUs' to competing for stable, commercially viable computing resources. Yawan Supercomputing, a subsidiary of Foxconn (2317), announced today (31) the official launch of its first 5MW AI cluster built on the NVIDIA HGX B300 system. The cluster has not only secured long-term leasing contracts with its first clients but has also entered commercial operation with over 90% utilization. Yawan Supercomputing stated that this is currently the largest cluster of its type in Taiwan, reflecting how demand for large-scale model training and inference is accelerating from the testing phase into formal production environments.

This cluster has also passed NVIDIA Exemplar Cloud validation for HGX B300 training workloads, making Yawan Supercomputing one of the world’s first ten and the second in Asia to complete this production-grade AI training validation as an NVIDIA Cloud Partner (NCP).

For Foxconn, the activation of the 5MW cluster represents more than just adding a batch of AI equipment. As Yawan begins to handle GPU cloud services, model training, inference, and AI agent deployment, Foxconn’s role in the AI market is expanding from server and infrastructure manufacturing further into computing power and platform operations.

Beyond just securing GPUs, enterprises now require AI computing power capable of formal commercial deployment. With the rapid development of generative AI, agentic AI, and multimodal models, enterprises need more than just GPUs—they require comprehensive AI infrastructure capable of long-term support for model training, fine-tuning, inference, and formal external services.

Yawan Supercomputing noted that enterprise demand for AI infrastructure has shifted from 'owning GPUs' to 'whether it can stably support production environments.' Unlike short-term testing or proof-of-concept projects, large-scale AI models, once deployed in production, must contend with challenges such as power supply, cooling, high-speed networking, cluster stability, and continuous operations.

Therefore, the newly launched 5MW cluster employs liquid cooling, a 2N high-availability architecture, and NVIDIA Quantum-X800 InfiniBand networking to support the high-speed data transfer and long-term collaborative computing of numerous GPUs required for large-scale model training and inference.

Yawan Supercomputing secured long-term leasing contracts with its first clients even before the cluster’s official launch and entered commercial operation with over 90% utilization, indicating that market demand has evolved beyond merely acquiring AI chips to seeking complete computing services that can be directly integrated into production environments.

Top 10 Globally Validated: Tested with Four Large Models. The NVIDIA Exemplar Cloud validation completed by Yawan Supercomputing is designed to verify whether AI infrastructure can deliver predictable, production-scale computing performance.

The validation process used NVIDIA Performance Benchmarking tools to run large AI workloads such as DeepSeek-V3, Llama 3.1, Qwen3, and Grok-1 to evaluate the platform’s computing performance, network efficiency, cluster stability, and consistency during prolonged operation.

For enterprise customers, this validation serves as a reference to assess whether an AI platform meets formal commercial deployment criteria, reducing the system integration and operational risks associated with building large GPU clusters in-house.

Yawan Supercomputing has become one of the world’s first ten and the second in Asia to pass NVIDIA’s production-grade AI training validation for the HGX B300. The term 'top ten globally' refers to being among the first ten companies worldwide to complete this validation, not to being among the top ten in terms of global AI computing power or performance.

From Model Training to AI Agent Deployment: Yawan Offers More Than GPU Rental. In addition to providing GPU computing resources, Yawan Supercomputing is attempting to expand its service scope from traditional computing power leasing to platform management and application deployment required for enterprises to adopt AI.

Yawan Supercomputing calls this service framework the 'Tokenized AI Factory,' integrating AI infrastructure, GPU Cloud, an AI Global Visibility Operations Platform, and the Token Factory platform to cover model training, fine-tuning, inference, and AI agent deployment.

This strategic move indicates that Yawan Supercomputing aims not only to offer GPU leasing based on computing resources but also to take on management services required for the formal deployment of models and AI applications. As AI computing supply continues to expand, pure GPU leasing may face price competition. The ability to provide model deployment, system operations, and application platforms will become a key factor in enhancing customer stickiness and service value.

Foxconn’s AI Expansion Extends to Services: Next Steps with Vera Rubin. Yawan Supercomputing is the entity within the Foxconn group dedicated to developing supercomputing centers and cloud computing operation platforms. Historically, Foxconn primarily played the role of a supplier of servers, racks, cooling systems, key components, and data center infrastructure in the AI market. Now, by entering supercomputing center and GPU cloud operations through Yawan Supercomputing, the group’s role is extending from hardware manufacturing to computing power and platform services.

In other words, Foxconn is not only manufacturing AI servers for clients but also directly operating computing resources and beginning to handle model training, inference, and AI agent deployment, thereby expanding its participation across the AI infrastructure and application service value chain.

Looking ahead, Yawan Supercomputing stated it will continue to expand AI infrastructure powered by NVIDIA Blackwell and Vera Rubin architectures across the Asia-Pacific region, leveraging Foxconn’s resources in key components, system manufacturing, and supply chain integration.

With the first 5MW cluster entering commercial operation at over 90% utilization, market attention will now focus on Yawan Supercomputing’s progress in adding new computing capacity and establishing new Asia-Pacific sites, as well as whether its operational model can expand beyond GPU leasing to include model, platform, and enterprise AI agent deployment services.