NTTPC Communications, Getworks, and Fixstars began offering the Liquid-Cooled GPU Container DC-PKG on September 8. The package bundles a containerized data center, liquid-cooled GPU servers, integrated monitoring, and operational support into a single offering. It targets companies that want an AI platform on their own premises but lack the cooling facilities and the staff to run it.
An AI Platform That Arrives Inside a 40-Foot Container
At the center of the package is a 40-foot container. Inside sit the facility components: a CDU chiller that cools the liquid-cooled GPUs, ceiling-mounted and InRow air conditioning, and security cameras.
The compute layer is built on NVIDIA accelerated computing servers. The reference configuration includes the HPE ProLiant Compute XD685 with NVIDIA HGX B300. Firewall and UTM security appliances and network switches are part of the same package.
The other pillar is monitoring and operational support. Integrated monitoring puts everything from the container facilities to the GPU servers on a single view, combined with performance visualization and operational assistance. The point is to avoid the situation where an outage goes unnoticed and expensive GPUs sit idle.
Power Keeps Rising, Cooling Capacity Does Not
The three companies frame the offering around a shift already underway: corporate AI work is moving from proof of concept to production. Agentic AI and generative AI inference are being embedded into real business processes, and while servers keep getting more capable, their absolute power draw keeps climbing. Data centers equipped to cool that load are in short supply.
Layered on top are the desire to keep confidential data in-house, network constraints, the cost of moving large volumes of data, and the operational burden of managing equipment at a remote site. Together these push companies toward placing AI infrastructure on their own property.
Building it alone, however, runs into three walls: a shortage of design and construction know-how for liquid-cooled servers, a shortage of specialists who can extract full GPU performance, and the ongoing load of post-deployment operations and maintenance. According to the three companies, these walls are why many projects drag on from evaluation to production. In December 2025 the group announced a successful proof of concept on improving the operational efficiency of liquid-cooled GPU servers, and this package turns those findings into a product.
Aimed at Confidential Data and Physical AI in Manufacturing
Two use cases are outlined. The first is working with confidential data: even under strict data and network requirements, AI analysis and generative AI can be used without moving information outside the organization.
The second is manufacturing and physical AI. Examples include putting visual inspection and anomaly detection into production using design data and video from production lines. Factories typically already own the land and the electrical infrastructure, so the preconditions for placing a container are often already in place.
From 300 Million Yen Upfront, With an 8-Month Lead Time
Pricing is quoted individually, but reference figures have been published. A small configuration with a single GPU server starts at 300 million yen upfront and 750,000 yen per month, both excluding tax and based on estimates as of August 31, 2026. The lead time for a standard configuration is a minimum of 8 months and may extend depending on site conditions and equipment availability. Construction and electrical work required to install the container are not included in the package and are charged separately.
Getworks, which handles the container side, announced its first containerized data center in 2013 and reports 300 units built as of the end of January 2026, comprising 270 twenty-foot and 30 forty-foot units. The company has also installed and operated more than 3,000 servers and over 10,000 GPUs. Fixstars, responsible for software optimization, specializes in accelerating both training and inference for AI models.
The three companies plan to add support for next-generation servers such as the NVIDIA Vera Rubin platform, and to expand into distributed AI infrastructure and edge AI.
Summary
The defining feature of the Liquid-Cooled GPU Container DC-PKG is that a container, liquid-cooled GPU servers, integrated monitoring, and operational support come as one unit that can be placed on a company's own site. Starting at 300 million yen for a single-GPU-server configuration is not cheap, but if it removes the work of contracting design, build, and operations to separate vendors and the coordination time that follows, price is not the only variable. For companies holding data they cannot hand to a public cloud yet still want to reach production, one more realistic option now exists.
