Addressing Social Challenges with Industry AI Cloud Powered by IOWN

1. The rapid rise in AI utilization

The emergence of generative artificial intelligence (AI) has accelerated the adoption of AI at an unprecedented rate. For instance, ChatGPT amassed 100 million users within 2 months—a stark contrast to Facebook, which took 54 months to achieve the same milestone. This highlights the extraordinary pace at which AI is advancing.

The AI market is projected to grow 20-fold from its 2021 levels, reaching 1.8 trillion dollars (approximately 280 trillion yen) by 2030 [1]. Within businesses, AI adoption has progressed significantly, particularly for improving productivity in general-purpose tasks.

According to a 2024 survey by Japan’s Ministry of Internal Affairs and Communications, more than 90% of companies in the United States have already introduced AI, and about 60% of companies in Japan have done so. While AI is an essential tool for driving digital transformation (DX), its impact on creating new services and fundamentally transforming business models remains limited. This indicates that the full potential of AI-enabled DX has yet to be achieved.

2. The need for AI in specialized tasks

To enhance the value of existing services and create new ones, it is vital to leverage AI in more specialized business operations. However, these specialized tasks often involve diverse factors such as data types, formats, and legal requirements, which vary by industry.

The manufacturing sector, for example, handles computer-aided design data, while the healthcare industry relies on electronic medical records and genetic information. These types of data are often highly sensitive, and companies are cautious about their potential exposure. Strict guidelines, therefore, govern their management, particularly regarding distribution and protection.

In healthcare, sensitive data owned by individuals, medical institutions, and corporations must be securely collected and analyzed in a protected environment. Within the NTT Group, NTT Precision Medicine provides genetic testing services that collect genetic information to analyze disease risks, contributing to advanced healthcare solutions (Fig. 1).

Fig. 1. AI application to specialized tasks: Case studies in the healthcare and medical fields.

At Prime Research Institute for Medical RWD, Inc. (PRiME-R), real-world data are collected and stored, while AI is used to structure electronic medical record data. These structured datasets are then analyzed to support research and development in fields such as pharmaceuticals. At NTT DATA, health-checkup information from employees is stored in the Health Data Bank, which also incorporates personal health records, such as step counts and sleep duration, to provide advice for promoting employee health.

However, these initiatives remain confined to their respective domains. To create new value—such as delivering personalized medicine or accelerating drug discovery—it is essential to aggregate independently collected data and analysis results into a shared platform for broader application (Fig. 2). In developing treatments for rare conditions, for example, it is crucial to efficiently identify patients with relevant medical histories and provide this information to pharmaceutical companies. Traditionally, pharmaceutical firms or their intermediaries have had to individually verify such cases with medical institutions.

Fig. 2. Cross-industry value creation: Case study in medical industry.

By combining and analyzing multiple datasets, it becomes possible to achieve outcomes that were previously unattainable with isolated data. This integration can lead to enhanced productivity across the industry and foster innovation by transforming traditional processes.

3. Co-creation through the Industry AI Cloud

General business processes have begun leveraging AI to enhance efficiency, but introducing AI into specialized operations can significantly increase their added value. By accumulating data onto a shared platform across companies and applying AI, industries can also address social challenges collectively.

This initiative, referred to as the Industry AI Cloud, involves NTT collaborating with various industry partners to foster co-creation. Below are examples of projects within this framework.

4. Toward a sustainable future

As AI adoption accelerates across various domains, increased power consumption will become an issue. AI computational infrastructure, built on large-scale datacenters, consumes significantly more energy than conventional servers. By 2030, it is estimated that energy consumption by Japan’s domestic datacenters alone will exceed the total energy demand of Tokyo in 2022 and exceed Japan’s total energy demand by 2050 (Fig. 3).

Fig. 3. Power consumption forecast for ICT infrastructure in Japan.

To address this growing demand, NTT is actively pursuing research and development to create a sustainable AI-powered society. This effort includes developing tsuzumi, a lightweight, low-power AI model and implementing the IOWN 2.0 energy-efficient computing infrastructure.

In 2023, NTT launched tsuzumi, an original, lightweight LLM optimized for Japanese, and commercial deployment began in March 2024. Depending on the type of data handled in various industries, lightweight models such as tsuzumi are often better suited to specific use cases.

Since its commercialization, tsuzumi has undergone continuous advancements. To meet diverse needs, efforts are underway to enable tsuzumi to process and interpret visual data, such as photos and graphs. Beyond improving its comprehension capabilities, research is focused on enabling tsuzumi to interact with web interfaces and systems as well as executing tasks collaboratively with humans. For example, if a user requests tsuzumi to order a product from an e-commerce site, the model can navigate the website, locate the desired item, complete the purchase, and subsequently access internal company systems to generate the necessary payment documents (Fig. 4).

Fig. 4. Advancement of tsuzumi.

The evolution of tsuzumi continues, with plans to incorporate user feedback into future updates. In a remarkable achievement, tsuzumi recently became the first Japanese LLM to be included in Microsoft’s Models-as-a-Service lineup. This milestone was announced at Ignite, Microsoft’s tech conference held in Chicago, and tsuzumi became available for use starting November 20, 2024.

5. Progress of low-power computing with IOWN

The rollout of IOWN 2.0 is slated for 2025, marking a significant step toward low-power computing. This initiative integrates photonics-electronics convergence (PEC) devices into computing infrastructure to reduce the energy consumption of servers and APN-transmission equipment.

The computing framework for the IOWN era is termed data-centric infrastructure (DCI), which reduces power consumption through the following two key approaches.

(1) Disaggregated computing

In the conventional architecture, servers contain multiple components within a single unit. Disaggregated computing, in contrast, involves separating these components and enabling shared use. This enables optimal combinations of central processing units (CPUs), memory, and other components based on actual needs while powering down unused components to conserve energy. The concept can be likened to breaking down and reorganizing server components for greater efficiency.

(2) PEC

This approach replaces traditional electronic circuits that process electrical signals with optical waveguides, reducing the energy used for electrical wiring (Fig. 5).

Fig. 5. Low-power computing resources through DCI.

6. From the IOWN concept to reality and beyond

As part of the IOWN 2.0 initiative, NTT is advancing the development of DCI-2 (DCI version 2) with the goal of commercial deployment by around 2026. DCI-2 leverages composable disaggregated infrastructure servers, which break down computing resources into board-level components. These components are interconnected using optical switches powered by PEC devices, all optimized by a DCI controller. The target is to achieve a dramatic one-eighth reduction in power consumption.

A major milestone in this effort will be showcased at the Expo 2025 Osaka, Kansai, Japan in April 2025, where NTT plans to implement these PEC-based servers in its pavilion. This marks a significant step toward actualizing the IOWN concept, offering visitors the chance to witness these next-generation servers in action. Commercialization of the server is anticipated by 2026.

Looking further ahead, NTT aims to achieve optical communication between chips by 2028 and extend this innovation to intra-chip communication by 2032, ultimately targeting a 100-fold reduction in power consumption (Fig. 6).

Fig. 6. IOWN from concept to reality and beyond.

By building on IOWN’s cutting-edge technology and harnessing the potential of the Industry AI Cloud, NTT is committed to addressing social challenges in a sustainable and transformative manner.

Reference

[1]	Ministry of Internal Affairs and Communications, “2024 White Paper on Information and Communications in Japan,” Chapter 9, 2024.