Trust in AI deep dive: Sustainability
As AI evolves and becomes more widespread, the power consumption and costs associated with computational processing are increasing exponentially. This is a critical business challenge that no company can afford to ignore, from the perspectives of both sustainability and financial discipline. To continue relying on AI, “sustainability”—the ability to operate smoothly over the long term—is essential.
Featured Content: A Journey Through History—Learning from the Past
Events that marked turning points in history raise questions that still resonate with today’s business leaders: the heliocentric theory, the Renaissance, the Titanic, the Silk Road, and the Age of Discovery. This series uses stories from the past as clues to understand the meaning of “trust” in the age of AI.
“1-Bit Quantization Technology” that makes advanced AI lightweight, low-cost, and easy to use
Technology introduction
1-bit quantization technology is a technique that minimizes the size of AI models while preserving as much of their capability as possible. AI models typically require significant amounts of memory and computational resources, but this technology enables generative AI to run faster, more affordably, and with lower power consumption. Furthermore, it makes it easier to use generative AI in field deployments or on-premises environments without the need for high-performance GPUs.
Why it matters
The value this technology brings to businesses is the ability to use advanced AI with lower costs and fewer constraints. It makes it easier to use generative AI even in environments with limited computational resources—such as on-premises environments, factory equipment, store terminals, and edge devices—and is expected to reduce initial investment and operational costs, reduce power consumption, and lower barriers to production deployment.
Example use case
The technology demonstrated up to 94% memory reduction and enabled large generative AI models that previously required four GPUs to run on a single GPU, opening opportunities for deployment on smartphones, industrial equipment, and other edge devices.
“AI Computing Broker,” middleware for efficient GPU utilization
Technology introduction
AI Computing Broker (ACB) is middleware designed to efficiently share the GPUs required for AI processing and make effective use of limited computing resources. While GPUs are indispensable for AI training and inference, in practice, there are often periods before and after processing when GPUs are not in use. ACB increases utilization by leveraging this idle time to flexibly allocate GPUs to other tasks. A key feature is its ability to facilitate the operation of multiple AI applications with fewer GPUs.
Why it matters
The value this technology brings to businesses is the ability to improve GPU utilization efficiency while keeping AI infrastructure costs and power consumption in check. Rather than continuously adding new GPUs, this technology enables more effective use of existing resources, making it easier to improve the return on investment in AI development and the operation of AI services. In particular, in areas that rely heavily on GPUs—such as R&D, LLM operations, and image and video processing—this results in significant differences in operational costs and deployment speed.
Example use case
AI-driven forecasting services must continuously develop high-accuracy models while also delivering fast responses to users, creating heavy demand for GPU resources. AI Computing Broker dynamically allocates computing resources according to workload requirements, enabling efficient support for both model development and service delivery. As a result, organizations can improve processing efficiency and maintain stable service performance as user demand increases.
“FUJITSU-MONAKA,” made-in-Japan CPU built for high-performance, energy-efficient AI
Technology introduction
FUJITSU-MONAKA*1 is a next-generation, Japan-made CPU based on the Arm architecture. By combining leading-edge 2nm process technology, a 3D many-core architecture, and ultra-low voltage operation technology, FUJITSU-MONAKA is designed to deliver both high processing performance and outstanding power efficiency—approximately twice that of competing CPUs*2. It also adopts Arm SVE2 and PCI Express 6.0 (CXL 3.0), and supports Confidential Computing, helping provide the performance, efficiency, and reliability required for AI, cloud, and enterprise infrastructure. Toward the release of FUJITSU-MONAKA in 2027, Fujitsu plans to begin trials using PoC machines equipped with the new CPU from summer 2026. Fujitsu also plans to release FUJITSU-MONAKA-X, a further successor product, in 2029.
*1 This is based on results obtained from a project subsidized by the New Energy and Industrial Technology Development Organization (NEDO).
*2 Based on Fujitsu's estimated performance projections for products planned for release in 2027. Actual performance may vary depending on usage, configuration, and other factors.
Why it matters
As AI adoption expands, data centers are facing growing constraints around power capacity, cooling facilities, and installation space. FUJITSU-MONAKA applies Fujitsu’s proprietary design technologies, cultivated through the development of world-class supercomputers, to support data processing and AI inference. For mid-sized LLMs of up to around 70 billion parameters, FUJITSU-MONAKA can operate independently, contributing to building AI inference infrastructure, optimizing Total Cost of Ownership (TCO), and strengthening sovereign AI infrastructure.
Example use case
[For data centers] FUJITSU-MONAKA can be used as a next-generation AI inference platform that does not rely solely on GPUs, thanks to its design and technologies optimized for AI inference. With its high energy efficiency, it is easier to deploy in existing data centers and can support AI use cases such as RAG, Q&A, summarization, and business process automation. It can also serve as a general-purpose on-premises infrastructure platform, contributing to TCO optimization through reduced power consumption.
[For national security and highly sensitive domains] By combining the reliability of a Japan-made CPU with Confidential Computing, FUJITSU-MONAKA can be used as a foundation for sovereign AI and secure data processing in highly sensitive areas such as defense and government. It supports AI inference in closed-network and on-premises environments, as well as the modernization of infrastructure that handles confidential data.