Sharp, a company most people associate with televisions and home appliances, is stepping onto new ground with enterprise AI servers. On September 15, 2026, the company began taking domestic orders for AI servers built on NVIDIA technology. Actual sales will not start until fiscal 2027, but Sharp is opening the order window early to capture demand. Backed by the manufacturing capacity of its parent company, Foxconn, the move is aimed squarely at Japan's growing appetite for AI infrastructure.
The First Model Is a 4U Server That Holds Eight GPUs
Orders have opened for the AI-B43100, positioned as a business-oriented AI server. It is a modular GPU server that can hold up to eight NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs, paired with two Intel Xeon 6 series processors. Memory consists of 32 DDR5 RDIMM modules of 96GB each for a total of 3TB, while storage combines 7.68TB of M.2 NVMe with 30.72TB of E1.S NVMe.
The chassis is a 4U rackmount unit measuring H175 x W438 x D800mm and weighing roughly 56kg. Power comes from four 3200W CRPS PSUs in a 3+1 redundant configuration, and cooling is air-based. That last detail matters: a company setting up its first in-house AI platform can install this in an ordinary server room without liquid cooling. Pricing is open, and GPU, memory, and storage configurations can be selected to match the workload.
Sharp describes the model as suited to inference, AI application development, and embedding AI into existing business systems. It also supports simulation and visual computing, which places it closer to a general-purpose GPU server than a machine built solely for generative AI.
The Liquid-Cooled Large-Scale Training Model Arrives in Fiscal 2027
A second line aimed at large-scale training and inference is also in preparation. It is built on the NVIDIA HGX Rubin NVL8 server baseboard, which links eight GPUs over NVLink, and uses Direct Liquid Cooling. On the networking side it supports NVIDIA ConnectX-9 SuperNIC adapters and NVIDIA BlueField-4 DPU configurations, a specification clearly aimed at training and serving large language models. Orders are scheduled to open in fiscal 2027, with detailed specifications to be published once finalized.
Both models support cluster deployment based on the NVIDIA Enterprise Reference Architecture. Combined with NVIDIA Quantum InfiniBand and NVIDIA Spectrum-X Ethernet, the lineup gives customers a path that starts with a single server and scales up to an AI factory later.
Building a Structure That Does Not End at the Sale
The hard part of the server business is not building the box, it is everything that happens after it ships. Sharp appears to have planned for that, locking in partners at the same time as its market entry. It has reached basic agreements with two distributors, Daiwabo Information System and Ryosan Ryoyo, and will provide maintenance support in cooperation with Tomorrow Net. The goal is a single point of contact covering proposal, deployment, operation, and maintenance.
The hardware itself is supplied through cooperation with Foxconn, which brings procurement and manufacturing scale. Supply of GPUs and other key components remains tight worldwide, and buyers who want an AI server often cannot get a reliable delivery date. Access to a group procurement network is a direct competitive advantage in that environment. Sharp is targeting 250 billion yen in revenue from this business in fiscal 2030, which suggests it intends to make AI servers a centerpiece among its new ventures.
The demand Sharp is chasing is on-premises rather than cloud. Companies, local governments, and research institutions increasingly want to use AI without sending confidential information or operational data outside their own walls, and the winner will be whoever can pair that hardware with hands-on deployment support. The fact that a consumer electronics maker is now presenting itself as an AI infrastructure provider says a good deal about how far Japan's AI investment has spread.
Summary
On September 15, 2026, Sharp announced its entry into the AI server business and opened domestic orders for the AI-B43100, its business-oriented model. It is an air-cooled 4U server holding up to eight NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs, with sales scheduled to begin in fiscal 2027. A liquid-cooled model for large-scale training is expected to follow. By combining Foxconn's manufacturing capacity with domestic sales and maintenance partners, and targeting 250 billion yen in revenue by fiscal 2030, Sharp is adding one more option for Japanese users who want to run AI on their own premises.
