NTT Plans Osaka AI Facility for Anthropic in Record-Scale Build

A Six-Billion-Dollar Bet on AI Infrastructure

NTT Data Group is nearing an agreement to construct a data centre for Anthropic in western Japan, a facility expected to cost approximately $6.3 billion and rank among the country's most powerful computing installations. The project, centred in Osaka, would mark the first time the San Francisco-based AI company has commissioned a dedicated data centre in Japan, and the scale reflects both Anthropic's regional ambitions and the infrastructure demands of frontier AI models.

At Opentechwire, we've tracked the migration of hyperscale AI infrastructure beyond the traditional hubs of Northern Virginia and Oregon, and this deal fits a pattern: foundation-model companies now treat data-centre location as a strategic variable, balancing power availability, latency to end users, and regulatory environment. Japan offers stable grid capacity, a sophisticated technical workforce, and a government keen to host AI compute within its borders. Anthropic's choice of NTT Data Group as builder signals confidence in local execution at a moment when global supply chains for GPUs, cooling systems, and high-voltage switchgear remain stretched.

Partial Opening by 2028

The Osaka facility is expected to begin partial operations in 2028, though the full build-out timeline remains under discussion. For context, a $6 billion data centre of this type typically houses tens of thousands of GPUs, requires multiple substations, and consumes power in the hundreds of megawatts. Even a phased opening will demand coordination across NTT's real-estate, engineering, and power-provisioning units, as well as close collaboration with Anthropic's infrastructure team on rack design, networking topology, and thermal management.

The 2028 target is aggressive but not implausible. Comparable facilities in the United States have moved from site selection to first-rack power in roughly 24 to 30 months when permitting, grid interconnection, and component availability align. Japan's streamlined industrial permitting in designated zones may help, though the sheer scale of electrical infrastructure required for a facility of this capacity remains a bottleneck everywhere.

Why Osaka, Why NTT

Osaka has emerged as Japan's second-largest data-centre market after Tokyo, driven by lower land costs, proximity to Kansai's manufacturing and logistics base, and deliberate government investment in digital infrastructure. The region also benefits from diverse fibre routes and redundant power feeds, both critical for a facility that will run inference and training workloads around the clock.

NTT Data Group brings more than construction expertise. As part of the NTT conglomerate, it can tap the parent company's fibre backbone, peering relationships, and enterprise customer base across Japan and Southeast Asia. That vertical integration matters for Anthropic, which may eventually serve Japanese enterprises through this facility rather than routing all API calls through US-based clusters. Low-latency access to Tokyo, Nagoya, and Fukuoka from Osaka also positions the centre well for real-time applications in finance, healthcare, and autonomous systems.

The deal also reflects NTT's pivot toward hyperscale AI clients. The group has historically focused on enterprise co-location and managed services, but the economics of AI compute have shifted: foundation-model companies are willing to commit to long-term capacity in exchange for guaranteed power and custom configurations. For NTT, a single anchor tenant of Anthropic's calibre can underwrite a facility's capital cost and de-risk future expansion.

Processing Capacity and Regional Implications

The planned facility would rank among Japan's largest in terms of processing capacity, a term that in AI contexts typically refers to aggregate FLOPs (floating-point operations per second) rather than traditional metrics like rack count or floor space. Anthropic's Claude models, which require substantial compute for both training runs and inference at scale, benefit from dense GPU clusters with high-bandwidth interconnects. The Osaka centre is likely to be optimised for these workloads, with liquid cooling, high-speed InfiniBand or Ethernet fabrics, and modular power distribution designed for future GPU generations.

From a regional perspective, this facility extends Anthropic's geographic reach at a time when data residency, sovereignty, and latency are increasingly factoring into enterprise procurement decisions. Japanese financial institutions, healthcare providers, and government agencies have shown interest in foundation models but often require that sensitive data never leave national boundaries. A Japan-based facility allows Anthropic to offer domestic inference endpoints and potentially on-premises fine-tuning, addressing those concerns without sacrificing model performance.

The move also positions Anthropic alongside OpenAI and Google, both of which have signalled intentions to expand AI infrastructure in Asia. The race is no longer just about model capabilities; it is about ensuring that compute capacity, data pipelines, and compliance frameworks exist where customers are. In that sense, the NTT deal is as much about market access as it is about megawatts.

Financing, Risk, and the AI Capex Cycle

A $6.3 billion project of this nature raises questions about financing structure and risk allocation. Typically, such deals involve a mix of NTT's balance sheet, project finance from Japanese banks, and long-term take-or-pay commitments from the tenant. Anthropic, backed by significant venture and strategic capital, has the financial strength to underwrite a multi-year lease, but the scale of this facility suggests NTT may also be betting on eventual multi-tenancy or resale value if Anthropic's needs evolve.

The broader context is a global AI infrastructure arms race in which capital expenditure by hyperscalers and foundation-model companies is running well ahead of revenue from AI products. Anthropic, like its peers, is building for anticipated demand rather than current utilisation, a strategy that hinges on continued progress in model capabilities, falling inference costs, and widespread enterprise adoption. If those assumptions hold, the Osaka facility will be a cornerstone asset. If demand plateaus or shifts to edge inference, NTT and Anthropic will need to adjust.

What It Means for Japan's AI Ecosystem

Beyond the immediate infrastructure play, the project has symbolic and practical weight for Japan's AI ambitions. The government has prioritised AI and semiconductor self-sufficiency in recent industrial policy, and hosting a frontier-model company's dedicated compute is a tangible outcome of those efforts. It also creates a nucleus for local talent development: NTT will need to hire and train engineers in AI-specific data-centre operations, and Anthropic may station technical staff in Japan to manage the facility and engage with regional customers.

There are risks. Japan's energy grid, while reliable, is not inexpensive, and power costs will be a significant line item in the facility's operating budget. The country also faces demographic headwinds that constrain the labour pool for both construction and operations. And while Japan's regulatory environment is generally supportive, any future restrictions on AI model exports, data flows, or compute access could complicate Anthropic's ability to operate the facility as a seamless extension of its global infrastructure.

Still, the deal underscores a shift: AI companies are no longer treating infrastructure as a back-office concern. Where compute lives, how it is powered, and who controls it are now strategic decisions with implications for product velocity, customer trust, and geopolitical positioning. NTT and Anthropic are placing a multi-billion-dollar wager that Osaka will be a node in that new map.