Sources: NTT Data is in advanced talks to build a data center for Paradox in Osaka of up to 500 MW, among Japan's largest, potentially costing $6.3B (Nikkei Asia)

NTT Data Group is a Japanese technology and data-center operator. Paradox is a U.S. AI company. They are discussing a major facility in Osaka, in western Japan. The talks matter because the project would support Paradox’s expanding computing needs and deepen Japan’s role in advanced AI infrastructure.

The proposed arrangement would have NTT Data build the data center for Paradox. The site could provide up to 500 megawatts of electrical capacity. That power would run large computer systems used to train and operate AI models. The project could therefore become a major piece of Paradox’s technical infrastructure in Japan.

Multiple sources told Nikkei that the companies were in the final stages of talks. The facility could be among Japan’s largest data centers and may cost as much as $6.3 billion. The source does not provide a final agreement, construction schedule, or confirmed location within Osaka, so those details remain unresolved.

A data center is a secure facility that houses servers and related equipment. Servers are powerful computers connected through high-speed networks. The building also provides electricity, backup systems, monitoring, and cooling. Together, these systems keep computing services available and protect equipment from heat, outages, and physical disruption.

For an AI company, the center runs two main workloads. During training, many computers process huge datasets repeatedly to adjust a model’s internal settings. During operation, the same infrastructure answers users’ prompts and performs other tasks. Specialized chips accelerate the calculations, while networks let thousands of chips work together.

The proposed Osaka facility illustrates this role. NTT Data is discussing a center for Paradox with capacity of up to 500 megawatts. That figure describes potential electrical capacity, not a guaranteed constant draw. If completed, the site would give Paradox a large, dedicated base for AI computing in Japan. The source does not state whether Paradox would own or operate it.

A megawatt measures power: the rate at which electricity is being used. Five hundred megawatts is the same as 500,000 kilowatts. It describes the maximum electrical capacity proposed for the Osaka data center, not necessarily its average consumption. A facility may draw less when computers are idle, equipment is being maintained, or capacity has not yet been installed.

If the center drew 500 megawatts continuously for a full year, the calculation would be 500 megawatts multiplied by 8,760 hours. That equals 4.38 million megawatt-hours, or 4.38 terawatt-hours. The figure includes the site’s total power needs, such as computing equipment, cooling, networking, lighting, and other systems.

The Nikkei report describes the project as potentially among Japan’s largest data centers. Its “up to 500 megawatts” wording signals an upper capacity limit. Actual annual electricity use would depend on how quickly the site is built, how intensively Paradox uses it, and how efficiently its equipment operates. The source gives no confirmed usage forecast.

Advanced AI systems learn by processing enormous amounts of data through mathematical operations. Training repeats those operations across many examples, often using thousands of specialized chips at once. Larger models and heavier workloads require more chips, faster connections, and longer computing runs. After training, answering many users simultaneously creates another large demand.

A data center makes this possible by grouping servers into tightly connected clusters. Electrical systems feed the chips and networking hardware. Cooling systems remove the heat produced when the chips operate. High-speed links coordinate the machines so they can share parts of one training job. Backup equipment helps protect long-running calculations from interruptions.

The Osaka proposal shows the scale now associated with AI infrastructure. NTT Data is discussing a facility for Paradox with up to 500 megawatts of capacity. That does not prove every watt would serve one model or be used constantly. It does show the planned site could support a very large computing operation. The source does not identify Paradox’s models or exact hardware.

A data center of this size needs a large, hardened building and a powerful electrical system. Construction can include land preparation, foundations, server halls, transformers, substations, backup power, cooling equipment, fire protection, security, and network connections. The building must also support dense computer hardware and continuous operation.

The electrical and cooling systems are especially substantial. Hundreds of megawatts require high-capacity connections to the grid and equipment that distributes power safely throughout the site. AI chips turn much of their electricity into heat, so cooling machinery, pipes, pumps, and heat-rejection equipment must be designed at comparable scale. Servers and networking equipment add another major cost.

Nikkei reports that the Osaka project could cost as much as $6.3 billion. That is a potential figure, not a confirmed final budget. The article does not break down the estimate or say whether it includes all computing hardware. Costs could depend on the final design, capacity, location, equipment, and construction schedule, none of which the report specifies.

The proposed facility could place a very large, concentrated demand on Osaka’s electricity system. The grid would need to deliver power reliably, even when AI workloads are heavy. Connecting the site might require new substations, transformers, transmission equipment, or other upgrades. The exact requirements would depend on the chosen site and available local capacity.

The center would also affect infrastructure beyond electricity. Construction could require roads, fiber-optic links, water or other cooling arrangements, and substantial industrial equipment. During operation, cooling systems would remove heat from the servers. If the facility operated near its full 500-megawatt capacity, its electricity demand could be continuous and would require careful planning by utilities and local authorities.

The report identifies no confirmed grid upgrades, water demand, traffic plan, or environmental assessment. Therefore, these are potential consequences, not announced project details. NTT Data and Paradox are still in advanced talks, according to multiple sources. The final design, location, operating level, and connection plan would determine the actual effect on Osaka’s infrastructure.

Electricity is the energy source for servers, specialized AI chips, memory, networking equipment, and cooling machines. The chips perform the many mathematical operations that adjust a model during training or generate an answer during operation. More chips and faster workloads generally require more electrical power, although efficiency varies by hardware and task.

Inside a data center, networks divide and coordinate work across many computers. Chips process data and produce heat as electricity flows through them. Cooling systems move that heat away using methods such as air or liquid cooling, allowing equipment to remain within safe operating temperatures. Power distribution, cooling, and computing must be designed as one system because failure in any part can interrupt AI work.

The proposed Osaka facility connects these needs at an unusually large scale. Its capacity could reach 500 megawatts, and it would be built for Paradox’s AI operations. That figure represents available power for the whole facility, not just chips. The source does not specify the chip types, cooling technology, or models involved.