Sakura.ne.jp Koukaryoku PHY H100 Model, JPY 3,046,120.00/mo.

This plan is the Interconnect Section B H100 GPU bare-metal server. It uses the Supermicro SuperServer SYS-821GE-TNHR chassis and includes NVIDIA H100 80GB x 8 GPUs, 2TB memory, redundant power, Broadcom SAS-3808 RAID controller, 25G x 2 port service NICs, and 400G x 4 port interconnect interfaces. The available operating systems are Ubuntu 24.04 and Ubuntu 22.04. The server is delivered powered off with no OS installed, and OS installation later applies the shared global network configuration automatically so the server can communicate immediately. The standard global network is 100Mbps, with 250Mbps and 500Mbps upgrades available for shared or dedicated global network use. Local network bandwidth is 4Gbps standard, with 10Gbps and 20Gbps options. One global network and up to ten local networks can be attached. The order listing shows monthly pricing from 3,046,120 JPY and setup from 0 JPY, and the listed stock status indicates no stock at the time of collection. NVMe storage consists of SSD 400GB x 2 on a RAID controller and SSD 7.68TB x 4 without the RAID controller. The service is delivered with RAID1 already configured, and the RAID configuration cannot be changed.

Koukaryoku PHY is a GPU bare-metal service designed for demanding AI development, large language model training, generative AI, machine learning, deep learning image analysis, natural-language processing, data analysis, risk prediction, and scientific simulation. It provides a complete dedicated physical server rather than a shared virtual instance, so the GPU, CPU, memory, local storage, and high-speed network interfaces are reserved for one customer environment. The service is positioned as a domestic Japanese GPU infrastructure option with data hosted in Sakura Internet infrastructure, and the Ishikari data center is presented as using renewable-energy-focused operations with zero annual CO2 emissions. The platform is intended for large enterprises, AI startups, research institutions, automotive AI training environments, and other users that need sustained high-performance compute with monthly fixed billing instead of per-second or per-hour usage. The service can be used for multi-node training because high-speed interconnect is available by section, and servers in different interconnect sections cannot communicate through that interconnect. Network bandwidth is best-effort. Interconnect setup is normally completed within three business days from the day after service start, with completion notification sent by email. The workflow is account creation, project creation in Sakura Cloud, payment after invoice confirmation, and then service start. The control panel is used for server operation, power-related work, OS installation, network connection changes, local networks, and other self-service administrative actions. Optional surrounding services include dedicated global networks, local networks, load balancer, firewall, name server, WAF, SSL certificate, external line connection, hybrid connection, console work, and equipment reboot support. A network SLA applies to the applicable network when monthly network uptime falls below 99.95 percent, excluding listed exceptions such as removing the recommended bonding configuration, interconnect network events, maintenance, user environment issues, attacks, disasters, time-billed services, and other excluded conditions.