NTT DOCOMO Business will begin offering GPUaaS, a cloud-based GPU platform for AI model development and training, starting September 30, 2026. The service lets customers start with as little as a single GPU and includes a ready-to-use browser-based development environment. The company expects it to be used not only by AI development firms but also by manufacturers, drug discovery companies, and universities and research institutions.
Designed for Small-Scale AI Experiments
Building an in-house AI development platform has traditionally required securing a substantial number of GPUs up front, which can mean over-investing for small-scale proof-of-concept (PoC) work. GPUaaS addresses this by letting customers start with a single GPU as the minimum configuration. The design is built around testing on a small scale first and scaling up as needed once results look promising.
A Development Environment That Runs Entirely in the Browser
GPUaaS comes with a pre-configured environment built around JupyterLab and Visual Studio Code, both accessible directly from a web browser without installing any dedicated software. The platform also manages AI models, datasets, and verification histories on its own, reducing the setup work required so more time can go toward actually building and testing models.
Closed-Network Access and Hybrid Configurations
On the security side, GPUaaS supports connections through NTT DOCOMO Business's closed corporate network services such as docomo business RINK, allowing access without routing through the public internet. It can also be linked with the company's Smart Data Platform cloud and server services, making it possible to keep AI processing and data storage entirely within a domestic environment. GPUaaS can further be combined with a customer's own dedicated GPU environment in a hybrid setup, so that steady, always-on workloads run on dedicated infrastructure while variable development and testing work runs on GPUaaS.
Who the Service Is Aimed At
NTT DOCOMO Business expects the service to be used by AI development startups and specialist firms, as well as manufacturers, drug discovery companies, and universities and research institutions. It is also positioned for companies that are cautious about entrusting sensitive data to external cloud services and want to build their own private AI environment. The service is designed to fit a staged approach to AI adoption: testing small in the early stages of research and development, then moving to full deployment once results are proven.
Pricing and GPU Models Not Yet Disclosed
As of now, NTT DOCOMO Business has not disclosed the specific GPU models or per-unit pricing for GPUaaS. These details are expected to be announced as the September 30 launch date approaches. In Japan, many companies remain reluctant to send AI training and inference data overseas, and telecom-affiliated carriers have been rolling out domestically hosted AI infrastructure services one after another to meet that demand. GPUaaS is positioned as part of that broader trend.
The service is being rolled out by the Cloud & Network Service Department within NTT DOCOMO Business's Platform Service Division. The company has previously offered cloud-based infrastructure services, and GPUaaS represents an extension of that business into the AI domain.
Summary
GPUaaS combines a pricing structure that starts from a single GPU, a development environment that runs entirely in the browser, and security through closed-network connectivity, positioning it as a cloud-based AI development platform. Once it launches on September 30, 2026, it looks set to become an option for companies and research institutions that want to start AI development without building their own infrastructure, while keeping costs limited during the testing phase.
