NTT is in the final stages of arranging to build a data center in Osaka Prefecture for the U.S. AI company Anthropic, Nikkei reported on October 13, 2026.
Construction costs could reach as much as ¥1 trillion, according to the report. NTT DATA Group and Anthropic are said to be in talks over a site in the bayside area of southern Osaka Prefecture. If the plan goes ahead, it could become the first dedicated data center in Japan for Anthropic, the developer of the generative AI model Claude.
Reuters reported the same day that total power receiving capacity could reach up to 500MW, with partial operations beginning as early as 2028. The two companies reportedly aim to sign a contract by the end of fiscal 2026.
However, the reports are based on people familiar with the matter. The construction plan and contract terms have not been formally confirmed.
NTT is already building out data centers in the Kansai region, and if this plan materializes, it would further strengthen Japan's domestic AI computing infrastructure. But having a data center in Japan is not necessarily the same as Japanese Claude users having their AI processing handled domestically, or being able to store their data in Japan.
To judge how the plan could affect Japanese companies, it will be necessary to look beyond the size of the facility and check what services will actually be offered and under what data-handling conditions.
The New Southern Osaka Site Is Separate From NTT's Existing Osaka North and Keihanna Projects
Apart from the data center reportedly planned for Anthropic, NTT DATA Group is developing several large facilities in Kansai.
One is the "Osaka North Data Center" under construction in Ibaraki, Osaka Prefecture.
NTT DATA Group outlined the construction plan in an announcement on May 27, 2025. Under the plan at the time, the first building would open in the second half of fiscal 2027, with a second building to be completed in 2032.
The first building is planned to provide 18MW of IT power capacity, rising to a total of 36MW once the second building is complete.
Another is the "Keihanna OSK11 Data Center" in the Keihanna area of Kyoto Prefecture.
This facility was officially opened, as announced on April 9, 2026, and has 30MW of IT power capacity. It is equipped with high-efficiency cooling systems and other features to meet demand for cloud and AI services.
The reported Anthropic facility is planned for the bayside area of southern Osaka Prefecture, so its location also differs from the two existing facilities.
The plans and facility scales break down as follows.
| 施設・計画 | 所在地 | 2026年10月13日時点の状況 | 電力容量 |
|---|---|---|---|
| Anthropic向け新データセンター | 大阪府南部の湾岸地帯 | 建設に向けた最終調整中と報道 | 最大500MWの総受電容量（報道） |
| 大阪北データセンター | 大阪府茨木市 | 2027年度下期に第一期棟の提供を開始する計画 | 第一期18MW、完成時36MWのIT電力容量 |
| 京阪奈OSK11データセンター | 京都府 | 2026年4月に正式開設を発表 | 30MWのIT電力容量 |
Note that these power figures cannot be compared directly.
The numbers for Osaka North and Keihanna OSK11 refer to "IT power capacity," the electricity that can be supplied to IT equipment such as servers.
The reported maximum of 500MW for the new Anthropic facility, by contrast, is "total power receiving capacity," the amount of power the entire facility can draw from utilities and other sources.
Total receiving capacity includes not only the power needed to run servers but also power for cooling systems and other facility equipment. This means not all 500MW could be used for AI computation.
The maximum construction cost of ¥1 trillion is likewise an estimate based on reports at this stage, not a confirmed contract value.
Power and Cooling Are Critical for AI Data Centers
NTT's design for the Osaka North Data Center offers a useful reference for understanding this plan.
NTT DATA Group says the facility can accommodate high-heat servers equipped with multiple GPUs, as well as racks that house computing equipment at high density.
It also plans to support "Direct Liquid Cooling," which uses coolant to cool heat-generating components directly.
Training and running generative AI requires large-scale computation using vast numbers of GPUs. But simply installing many high-performance GPUs is not enough to deliver stable AI services.
The more GPUs there are, the greater the power consumption and heat output, so facilities need both sufficient power and cooling systems that can remove that heat efficiently.
Conventional air cooling can struggle to handle the heat from densely packed GPU servers. As a result, liquid cooling is becoming increasingly important for large-scale AI computing infrastructure.
In the Osaka North plan, NTT likewise emphasizes securing land and power as well as supporting high-density racks and liquid cooling.
Still, even with such equipment, power receiving capacity alone does not determine how much AI computing power can actually be used.
The facility's overall computing capability varies greatly depending on the type and number of GPUs installed, the cooling method, and the efficiency of power delivery.
Networking that connects data centers is also important.
For Osaka North, NTT has indicated plans to adopt "IOWN APN (All-Photonics Network)," its next-generation optical communications platform.
The idea is to enable high-speed, high-capacity connections with other data centers owned by the NTT Group and its customers, so the facility can also serve as an expansion site for existing ones.
Being able to connect multiple facilities widens the options for distributing computing resources and data while operating them as a single integrated system.
That said, these are specifications published for the Osaka North Data Center. It has not been decided that the same equipment will be installed at the newly reported Anthropic facility.
NTT DATA Group positions Osaka not only as a backup site in case of a disaster in the Tokyo area, but also as an important region supporting large-scale cloud services.
At the opening of Keihanna OSK11, it likewise highlighted improved business continuity through geographic distribution from Tokyo, and its response to cloud and AI demand in western Japan.
If the new Anthropic site is realized, it would add large-scale computing facilities for a company developing cutting-edge AI models to Kansai's growing cluster of data centers.
A Japanese Data Center Doesn't Guarantee Claude Data Will Be Stored in Japan
Anthropic has also been expanding its business in Japan's enterprise market.
On April 24, 2026, the company announced a strategic collaboration with NEC.
Under the collaboration, about 30,000 NEC Group employees around the world will be able to use Claude, and the companies plan to co-develop AI products emphasizing safety and reliability for Japan's financial, manufacturing, and local government sectors.
Note that the figure of about 30,000 refers to eligible employees across the entire NEC Group, not the number of Claude users in Japan.
The collaboration shows that Anthropic is working to embed Claude in the operations of Japanese companies.
However, even if a data center for Anthropic is built in Japan, data sent by Claude users will not necessarily be processed or stored in Japan automatically.
The key distinction here is between where data is processed and where it is stored.
In an official explanation dated June 15, 2026, Anthropic set out its policy on data processing and storage locations for commercial services such as Claude Enterprise and the Anthropic API.
According to that explanation, unless otherwise agreed or specified by the customer, customer requests may be sent to facilities in the United States, Europe, Asia, and some countries in Australia.
Data storage, meanwhile, is stated explicitly as the United States.
In other words, even if there is a mechanism for choosing where data is processed, storage will not necessarily be in the same region.
For example, even if facilities for AI inference are set up in Japan, customer data could continue to be stored in the United States.
This difference matters when companies adopt Claude for their operations.
For companies handling confidential information, customer data, or regulated data in particular, where data is processed and where it is stored can be a deciding factor in choosing a service.
Anthropic also explains that for internal operations such as safety reviews, product support, and incident response, it may process data in countries where it or its affiliates operate.
That means companies need to check not only where ordinary requests are processed, but also how data is handled in operations such as maintenance and support.
The latest reports do not make clear exactly which services the Osaka data center would be used for.
The impact on Japanese Claude users will also differ depending on whether the facility is used for training AI models or for inference, which generates responses to users.
For example, if it is used mainly for model training, it would not necessarily directly reduce latency for users in Japan accessing Claude.
If it handles inference for Japan, on the other hand, it could reduce communication latency for access from within the country. But actual response speed also depends on model processing time and system configuration, so placing equipment in Japan does not by itself guarantee faster responses.
It is also not yet known whether services will be offered that let customers specify domestic data processing or storage.
Such conditions are separate from the construction plan, and will need to be confirmed through the service specifications and contract terms Anthropic announces in the future.
What to Watch: Contract Signing, Start of Operations, and Claude Service Terms
The reports show that NTT DATA Group and Anthropic are in talks to establish large-scale AI computing infrastructure in Japan.
Construction costs could reach up to ¥1 trillion and total power receiving capacity up to 500MW, which, if the plan proceeds, would make it a major investment in Japan's AI infrastructure.
At this stage, however, neither a formal contract nor final facility specifications have been confirmed.
The first thing to watch is whether the two companies sign a contract. After that, the site, facility scale, equipment delivery schedule, and actual start of operations will be key factors.
On NTT's side, important questions are when the necessary power can be secured, what cooling systems will be installed, and how many GPUs and other computing machines the facility can house.
On Anthropic's side, the focus is whether the new facility will be used for AI model training or inference, and which Claude services it will cover.
For Japanese companies in particular, whether a new domestic base widens their options for where data is processed and stored will also be significant.
In Kansai, large-scale data centers are being developed by operators including NTT DATA Group.
If this plan is realized, Anthropic's computing infrastructure would join that cluster of facilities, potentially expanding the options for AI infrastructure in Japan.
However, a large-scale AI data center being built in Japan is a separate matter from whether Japanese users can complete AI processing and data storage within the country.
What benefits this massive construction investment will bring to Japanese companies will become clear not just from the size of the facility, but from the Claude services actually offered and the data-handling conditions that are disclosed.