Sharp Corporation will enter the AI server business by accepting orders for AI servers equipped with NVIDIA technology in Japan starting today, September 15, from corporations, local governments, research institutions, data center operators, and other customers. Ahead of the planned commercial launch in fiscal 2027, Sharp aims to address the growing demand for AI infrastructure driven by the proliferation of generative AI. The first product available for order is the AI Server for Enterprise Workloads (NVIDIA RTX Pro Server). Sharp also plans to begin accepting orders for the AI Server for Large-Scale Training and Inference (NVIDIA HGX System) at a later date.
Left: AI Server for Enterprise Workloads (NVIDIA RTX Pro Server)Right: AI Server for Large-Scale Training and Inference (NVIDIA HGX System)
In developing its AI server business, Sharp is building out its sales structure as well as its operation and maintenance structure following deployment. Sharp has respectively reached a basic agreement on collaboration with two sales agents, Daiwabo Information System Co., Ltd. (head offices: Kita-ku, Osaka City and Minato-ku, Tokyo) and Ryoyo Ryosan, Inc. (head office: Chiyoda-ku, Tokyo). In addition, Sharp will provide maintenance support in collaboration with TOMORROW NET Co., Ltd. (head office: Shinagawa-ku, Tokyo), which will be responsible for operation and maintenance.
Through collaboration with these partners, Sharp will build a one-stop structure covering customer proposals, deployment, operation, and maintenance, and will support the development and expansion of AI infrastructure in Japan.
An AI server is a high-performance server that handles AI training and inference processing. As generative AI becomes more widespread and data utilization becomes more sophisticated, there is growing demand in Japan from corporations, local governments, research institutions, and others who wish to utilize AI on-premises (*1) while ensuring security. At the same time, the global supply-demand balance for key components such as GPUs (*2) remains constrained, and procuring AI servers can require considerable time and coordination.
In this environment, Sharp will respond to market needs through collaboration with Hon Hai Technology Group (Foxconn, head office: New Taipei City, Taiwan), which excels in procurement and manufacturing.
Product Name
Model Number
Price
Order Acceptance Date
AI Server for Enterprise Workloads (NVIDIA RTX Pro Server)
AI-B43100
AI Server for Large-Scale Training and Inference (NVIDIA HGX System)
FY2027 (plan)
*1 A form of operation in which servers are installed and operated within one's own company or organization rather than in the cloud. A key feature is that confidential information and business data can be managed in-house.
*2 Abbreviation for Graphics Processing Unit. A computing device that performs large volumes of calculations at high speed, such as image and video rendering, and is also used for AI processing.
■ Overview of the AI Servers
1. AI Server for Enterprise Workloads (NVIDIA RTX Pro Server)
This is a modular GPU server that can be configured with up to eight high-performance NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. It supports AI deployment in business environments, simulation, and visual computing workloads, allowing customers to select the GPU, memory and storage configuration according to their specific applications. It is designed for inference, AI application development, and the integration of AI into enterprise systems.
2. AI Server for Large-Scale Training and Inference (NVIDIA HGX System)
This server is equipped with the NVIDIA HGX Rubin NVL8, a server board that interconnects eight GPUs via NVIDIA NVLink, and adopts a Direct Liquid Cooling method (*3). It also supports NVIDIA ConnectX-9 SuperNIC adapters and NVIDIA BlueField-4 DPU configurations, making it optimal for the training and inference of generative AI models, including Large Language Models (LLMs). Delivering stable high-density, high-power operation, it is a model well suited for AI training infrastructure and core data center applications.
Common to Models 1 and 2: High-Performance Networking
Both models support cluster deployment based on the NVIDIA Enterprise Reference Architecture (Enterprise RA), and by combining NVIDIA Quantum InfiniBand and NVIDIA Spectrum-X Ethernet, they realize the implementation of scalable AI factories.
*3 A cooling method using liquid instead of conventional air cooling to dissipate heat generated inside equipment more efficiently.
■ Main Specifications of the AI Server for Enterprise Workloads (NVIDIA RTX Pro Server)
Model Number
AI-B43100
GPU
NVIDIA® RTX PROTM 6000 Blackwell Server Edition x 8
CPU
Intel® Xeon® 6 Series processors x 2
Memory
DDR5 RDIMM 96 GB x 32 (Total 3 TB)
Storage
M.2 NVMe 3.84 TB x 2 (Total 7.68 TB)
E1.S NVMe 1.92 TB x 16 (Total 30.72 TB)
Power Supply
CRPS PSU 3200 W x 4 (3+1 redundancy)
Dimensions
175H x 438W x 800D mm (4U rack mount)
Weight
Approx. 56 kg
Cooling Method
Air Cooling
● Specifications and appearance are subject to change without notice due to improvements and other reasons.
● Sharp is preparing to begin accepting orders for the AI Server for Large-Scale Training and Inference (NVIDIA HGX System) in FY2027, and detailed specifications will be announced once finalized.
● Even within Japan, related services may not be available in some cases due to geographic constraints, such as the distance from maintenance service bases.
● NVIDIA, NVIDIA RTX Pro, NVIDIA Blackwell, NVIDIA HGX, NVIDIA Rubin, NVIDIA NVLink, NVIDIA ConnectX, NVIDIA BlueField, NVIDIA Quantum InfiniBand and NVIDIA Spectrum-X Ethernet are trademarks and/or registered trademarks of NVIDIA Corporation in the United States and other countries.
● Intel and Xeon are trademarks and/or registered trademarks of Intel Corporation or its subsidiaries in the United States and/or other countries.
■ Customer Inquiries (Japanese only)
For consultation and inquiries, please visit the following website.
https://smj.jp.sharp/bs/ssl/sh/ai-server/form.php
