Practical Performance Confirmed in Remote Distributed AI Infrastructure Demonstration Between Tokyo and Fukuoka Utilizing 'IOWN APN', Tailored to Workload Characteristics
Four companies—GMO Internet, NTT East, NTT West, and QTnet—have completed a technical demonstration of remote distributed AI infrastructure between Tokyo and Fukuoka using IOWN APN. They confirmed that large language model training experienced only about a 0.5% performance degradation compared to a local environment, indicating its practical viability.
📋 Article Processing Timeline
- 📰 Published: March 30, 2026 at 19:30
- 🔍 Collected: March 30, 2026 at 22:56 (3h 26m after Published)
- 🤖 AI Analyzed: April 22, 2026 at 22:24 (551h 28m after Collected)
GMO Internet Inc. (Headquarters: Shibuya-ku, Tokyo; Representative Director, President and Executive Officer: Tadashi Ito; hereinafter, GMO Internet) of GMO Internet Group, NTT East Corporation (Headquarters: Shinjuku-ku, Tokyo; Representative Director, President: Naoki Shibuya; hereinafter, NTT East), NTT West Corporation (Headquarters: Osaka-shi, Osaka; Representative Director, President: Ryota Kitamura; hereinafter, NTT West), and QTnet, Inc. (Headquarters: Fukuoka-shi, Fukuoka; Representative Director, President: Yoshio Ogura; hereinafter, QTnet) have completed a technical demonstration of remote distributed AI infrastructure between Tokyo and Fukuoka utilizing "APN (All-Photonics Network)" of "IOWN (Innovative Optical and Wireless Network)".
In this demonstration, conducted from November 2025 to February 2026, a real IOWN APN line was laid between Tokyo (storage) and Fukuoka (GPU), and the AI workload performance was measured and evaluated on an AI development platform connecting GMO GPU Cloud's GPUs with large-capacity storage. As a result, for large language model (LLM) training, the performance degradation was limited to only about 0.5% compared to a local environment, confirming that the impact was extremely limited. For image classification tasks involving data loading, it was also confirmed that practical level processing is possible even in a remote environment through optimization of training data, etc., demonstrating that practical AI development in a remote distributed environment is possible with designs tailored to workload characteristics.
Prior to this demonstration, the four companies conducted a preliminary demonstration (Phase 1) in July 2025, testing performance in a simulated remote environment assuming a distance of approximately 1,000 km between Tokyo and Fukuoka, and published the details as a technical report.
Press Release: https://internet.gmo/news/article/88/
Technical Report: https://internet.gmo/news/article/87/
Based on the results of this demonstration, the four companies will continue their efforts towards the practical application of remote distributed AI infrastructure tailored to customer needs.
**[Background and Purpose]**
With the recent spread of generative AI and large language models (LLMs), the demand for AI development platforms has rapidly expanded. Traditionally, GPUs and large-capacity storage have been required to be physically co-located. However, due to data center space constraints and the need for companies to manage data at their own locations, there has been a demand for distributed AI development platforms that transcend geographical limitations. The four companies have been exploring the technical feasibility of connecting GPUs and storage in remote locations using IOWN APN, which features high-speed, large-capacity, and low-latency communication.
**▼Examples of Challenges in Building AI Development Platforms**
[Image of challenges in building AI development platforms]
In this demonstration, conducted from November 2025 to February 2026, a real IOWN APN line was laid between Tokyo (storage) and Fukuoka (GPU), and the AI workload performance was measured and evaluated on an AI development platform connecting GMO GPU Cloud's GPUs with large-capacity storage. As a result, for large language model (LLM) training, the performance degradation was limited to only about 0.5% compared to a local environment, confirming that the impact was extremely limited. For image classification tasks involving data loading, it was also confirmed that practical level processing is possible even in a remote environment through optimization of training data, etc., demonstrating that practical AI development in a remote distributed environment is possible with designs tailored to workload characteristics.
Prior to this demonstration, the four companies conducted a preliminary demonstration (Phase 1) in July 2025, testing performance in a simulated remote environment assuming a distance of approximately 1,000 km between Tokyo and Fukuoka, and published the details as a technical report.
Press Release: https://internet.gmo/news/article/88/
Technical Report: https://internet.gmo/news/article/87/
Based on the results of this demonstration, the four companies will continue their efforts towards the practical application of remote distributed AI infrastructure tailored to customer needs.
**[Background and Purpose]**
With the recent spread of generative AI and large language models (LLMs), the demand for AI development platforms has rapidly expanded. Traditionally, GPUs and large-capacity storage have been required to be physically co-located. However, due to data center space constraints and the need for companies to manage data at their own locations, there has been a demand for distributed AI development platforms that transcend geographical limitations. The four companies have been exploring the technical feasibility of connecting GPUs and storage in remote locations using IOWN APN, which features high-speed, large-capacity, and low-latency communication.
**▼Examples of Challenges in Building AI Development Platforms**
[Image of challenges in building AI development platforms]