日本語 
Japan is making yet another big move towards creating AI-infrastructure of the future after NTT DOCOMO announced the deployment of a network-focused distributed AI cluster based on All-Photonics Network (APN) technology. This project, called GPU over APN, brings together various geographically separated data centers around Japan that will enable GPUs located in different areas to act as one single platform for AI training and inference. Based on the IOWN architecture, this new venture is supposed to provide high-speed connectivity with low latency while not requiring huge central data centers.
This move is timely as corporations all over the world are trying to create infrastructure for their AI systems which can handle increasingly complex large language models (LLMs). As computing loads keep growing, networking technologies become as significant as computation. NTT’s latest move proves once again how important photonics can be for the future design and operation of AI data centers.
Why Photonics Is Becoming Critical for AI
The need for high-performance computing has increased due to the rise in artificial intelligence usage, but it goes further than adding more GPUs. Contemporary applications based on AI require huge amounts of data to be exchanged fast among processors, storage, and data centers. Regular electronic networking technologies are inefficient and form barriers when the AI cluster grows, creating additional delay time and wasting energy.
Also Read: Micronity Names Akihiro Kadohata, as an advisor to “srest”
Using All-Photonics Network developed by NTT, one can use optical communications and exchange data faster while spending less power. With GPU over APN, GPUs distributed in Tokyo, Osaka, Fukuoka, Sapporo, Kanazawa, Kawasaki, Mitaka, Akihabara, and Yokohama can work as if they are all situated within one data center. NTT has proven 100 Gb/s connection by using remote direct memory access (RDMA) and transferring 25 GB of data in two seconds from Fukuoka to Tokyo.
Such an approach allows companies to allocate their AI load among several data centers without any effect on performance.
A Major Boost for Japan’s AI Ambitions
The project is very much in line with the overall strategy of Japan to enhance its capabilities in the AI realm and digital sovereignty.
The government of Japan has recently declared massive funding for sovereign AI models, semiconductors, advanced cloud computing infrastructure, and future communications networks. But having a good AI infrastructure demands not just powerful processors, but high performance networking infrastructure as well.
In addition to these national efforts, the new photonics architecture of NTT will help build a networking infrastructure that would allow Japan to connect all the computational resources together. Rather than using the hyper-scale centralized data centers alone, Japan will develop an infrastructure structure whereby different regional AI data centers work in conjunction to form one single computing environment.
Transforming Enterprise AI and Sustainable Computing
Implications far exceed the realm of telecommunications.
Firms using generative AI models are faced with problems of GPU scarcity, high capital expenses, and fast-growing power usage. Networks of AI clusters built upon optical networking make it possible for firms to use their computing capacities effectively and decrease their reliance on one particular physical location.
Manufacturing industry, finance industry, healthcare industry, logistics, telecommunications, and science all stand to gain from faster AI model training, low-latency inference applications, and increased robustness.
Yet another crucial advantage is sustainability. AI data centers use tremendous amounts of electricity, with networking infrastructure taking up an increasingly large portion of overall power needs. NTT’s photonics technology makes it possible not only to conserve energy but also to distribute workloads among regional facilities which could get their power needs met by renewables.
New Opportunities for Japan’s Technology Industry
The new NTT project promises to open up numerous prospects within Japan’s developing AI infrastructure eco-system.
An increase in distributed AI data centers’ adoption will bring demand for optical networking components, semiconductors, AI accelerators, cloud services, advanced cooling solutions, and high-performance storage technology. Companies in Japan specialized in fiber-optic communication, precision machining, industrial automation, and data centers are well-positioned to take advantage of greater investments into AI infrastructure.
Furthermore, the project opens up possibilities for cloud service providers, system integrators, and software vendors to develop platforms for managing AI workloads on multiple geographically distributed sites. With enterprises relying more and more on hybrid AI architecture, the demand for intelligent workload management, cybersecurity, and network management solutions is anticipated to grow substantially.
The new project will also complement the recent initiatives by Equinix Japan and Sakura Internet for testing photonic interconnection between regional data centers, illustrating that the distributed AI infrastructure is progressing towards commercial implementation.
A New Blueprint for AI Data Centers
The NTT GPU over APN strategy highlights a paradigm shift in the way AI infrastructure will be developed over the coming decade. In contrast to developing AI infrastructure with supercomputing power in just a few enormous hyperscale facilities, AI infrastructure of the future will be made up of multiple interconnected regional data centers connected by ultra-fast optical networks.
This will constitute an inherent competitive advantage for Japan, which has long been a leader in photonics, telecommunications, and precision engineering. Now these areas of expertise are being put to use in the rapidly growing area of the digital economy, thus offering an example of innovation for building an AI infrastructure of the future.
With enterprises increasingly relying on artificial intelligence technologies, innovations like the photonics-powered data center developed by NTT can have far-reaching consequences for the way enterprises develop artificial intelligence. They will make it possible not only to increase performance but also to strengthen Japan’s position as a leader in AI infrastructure development.
English