Sharp announced on September 15, 2026, that it had begun accepting orders for AI servers aimed at companies, government agencies, and data center operators in Japan, in a move that expands the company’s business from hardware solutions to infrastructure dedicated to AI workloads. Sharp is targeting sales of 250 billion yen from its AI server business during fiscal 2030.
The company plans to begin sales in fiscal 2027, offering customizable solutions based on customer needs. The initial offering includes a server designed for enterprise use that is based on NVIDIA RTX Pro Server technology, while Sharp is also preparing to begin accepting orders for servers aimed at large-scale training and inference using the NVIDIA HGX platform.
RTX Pro Server Designed for Enterprise Workloads
The enterprise server supports up to eight NVIDIA RTX 6000 Blackwell Server Edition graphics processing units. Sharp says its modular design allows customers to select GPUs, memory, and storage according to the intended use, such as AI applications, simulation, visual computing, and the integration of AI into enterprise systems.
The announced specifications include two Intel Xeon 6 processors and 96 gigabytes of DDR5 RDIMM memory distributed across 32 modules, for a total of up to 3 terabytes. Storage consists of two M.2 NVMe drives with a combined capacity of 7.68 terabytes and 16 E1.S NVMe drives with a combined capacity of 30.72 terabytes. The system uses four 3,200-watt CRPS power supplies, configured in a 3+1 redundant arrangement.
The server comes in a 4U rack-mountable chassis, with dimensions of 175 centimeters in width, 438 centimeters in height, and 800 millimeters in depth according to the data extracted from the source, and has a stated performance of up to 32 petaflops.
Another Platform for Large-Scale Training and Inference
For larger workloads, Sharp referred to a server based on the NVIDIA HGX Rubin NVL8, which connects eight GPUs through NVIDIA NVLink. The design also includes direct liquid cooling and support for the NVIDIA ConnectX-9 SuperNIC and NVIDIA BlueField-4 DPU networking components, making it intended for training and inference of large AI models within data centers.
Sharp intends to use NVIDIA Quantum InfiniBand and NVIDIA Spectrum-X Ethernet networking technologies, based on the NVIDIA Enterprise Reference Architecture, with the aim of supporting the construction of scalable AI clusters and production lines.
What Changes in Practice?
Sharp’s announcement is not limited to adding a server to its product portfolio; it reflects an attempt to build an integrated business that begins with hardware design and customization and extends to operation and maintenance. The company says it will work with partners in NVIDIA platforms and server components to provide solutions suited to different use cases, and it also links the project to expanding AI applications within companies, local government agencies, and data centers.
The importance of the move lies in the fact that Sharp is not merely announcing a sales date; it is also defining a three-stage path: beginning server sales, then developing services and solutions based on customer requirements, and finally expanding applications toward physical AI and edge AI. However, the source does not clarify server prices, expected capacity, or details of after-sales services, and the 2030 sales target remains an announced commercial goal rather than an achievement to date.