NTT Integration and IBM Japan Begin Technical Verification of Next-Generation On-Premise AI Infrastructure Using 'IBM Spyre'
NTT Integration and IBM Japan have jointly launched a technical verification to develop a next-generation on-premise AI infrastructure service utilizing 'IBM Spyre Accelerator for Power', an enterprise AI accelerator. This marks the first such initiative in Japan. They aim to solve power consumption and data security issues in full-scale AI deployment. NTT Integration will act as 'Client Zero' to accumulate practical knowledge.
📋 Article Processing Timeline
- 📰 Published: May 20, 2026 at 00:00
- 🔍 Collected: May 19, 2026 at 15:32
- 🤖 AI Analyzed: May 20, 2026 at 08:18 (16h 46m after Collected)
NTT Integration Corporation and IBM Japan, Ltd. have jointly initiated a technical verification for the development of a next-generation on-premise AI infrastructure service utilizing the 'IBM Spyre Accelerator for Power' (hereinafter IBM Spyre), an enterprise AI accelerator. This marks the first such endeavor in Japan. Through the provision of this service, NTT Integration aims to realize an environment where companies can utilize AI more securely and efficiently within an on-premise environment.
In recent years, while the use of generative AI and large AI models has rapidly expanded, new challenges are emerging as enterprises move beyond PoCs and limited application stages into full-scale operational deployment. One of these challenges is the increasing power consumption of AI inference infrastructure and the power constraints of data centers. While traditional GPU-centric AI infrastructures offer high performance, their significant power consumption and installation requirements pose a challenge for sustainable AI utilization. Furthermore, the introduction and operation of generative AI and RAG require high-level specialized knowledge, often resulting in the concentration of AI infrastructure design and operation tasks among specific IT personnel. Consequently, issues have been raised regarding the difficulty for frontline departments to utilize AI autonomously. Additionally, for companies handling highly confidential core data and business data, there is a persistent need to use AI securely without moving data to external environments, reaffirming the importance of AI utilization in on-premise environments. Particularly in the design departments and production sites of the manufacturing industry, highly confidential intellectual property and sensitive information, such as design drawings, manufacturing process data, and yield information, are strictly managed on-premise. The demand to utilize this data with AI without taking it outside the company has become increasingly strong.
'IBM Spyre' is gaining attention as a key to solving these challenges. IBM Spyre is an AI accelerator specialized for AI inference processing and designed for enterprise use. It features the ability to suppress power consumption and installation requirements while maintaining high inference performance, and is expected to address issues such as power constraints and operational workloads. While the introduction of typical GPU servers often requires renovation works for large-capacity power supplies and the preparation of dedicated cooling facilities, IBM Spyre is designed to be installed as an add-on card to IBM Power11 servers. This allows for the addition of AI inference capabilities without significant modifications to power facilities or air conditioning. Moreover, by combining it with IBM Power servers, AI inference can be completed within the server without moving core business data externally, allowing for the construction of an on-premise AI infrastructure that meets the needs of companies emphasizing security and data sovereignty. Based on these features, NTT Integration and IBM Japan have started technical verification regarding the feasibility of an on-premise AI infrastructure utilizing IBM Spyre.
In this technical verification, the effectiveness as an enterprise AI inference infrastructure will be verified by combining IBM Power11 and IBM Spyre. By coordinating the core business data handled by IBM Power and the AI inference processing on the same infrastructure, a configuration that allows AI utilization without moving data will be verified, aiming to realize an on-premise AI infrastructure that balances high security and operational efficiency. In addition, evaluations will be conducted from the perspective of power consumption and operational workload to verify whether it is an AI infrastructure that companies can implement in actual operations.
NTT Integration positions itself as 'Client Zero' in this initiative and will apply IBM Spyre to actual operations. Verifications will be conducted targeting use cases tailored to daily business, such as the analysis of core business data running on IBM i and internal knowledge searches utilizing RAG. The policy is to accumulate practical knowledge in construction, operation, and post-operation improvements, and reflect the results in future services.
In recent years, while the use of generative AI and large AI models has rapidly expanded, new challenges are emerging as enterprises move beyond PoCs and limited application stages into full-scale operational deployment. One of these challenges is the increasing power consumption of AI inference infrastructure and the power constraints of data centers. While traditional GPU-centric AI infrastructures offer high performance, their significant power consumption and installation requirements pose a challenge for sustainable AI utilization. Furthermore, the introduction and operation of generative AI and RAG require high-level specialized knowledge, often resulting in the concentration of AI infrastructure design and operation tasks among specific IT personnel. Consequently, issues have been raised regarding the difficulty for frontline departments to utilize AI autonomously. Additionally, for companies handling highly confidential core data and business data, there is a persistent need to use AI securely without moving data to external environments, reaffirming the importance of AI utilization in on-premise environments. Particularly in the design departments and production sites of the manufacturing industry, highly confidential intellectual property and sensitive information, such as design drawings, manufacturing process data, and yield information, are strictly managed on-premise. The demand to utilize this data with AI without taking it outside the company has become increasingly strong.
'IBM Spyre' is gaining attention as a key to solving these challenges. IBM Spyre is an AI accelerator specialized for AI inference processing and designed for enterprise use. It features the ability to suppress power consumption and installation requirements while maintaining high inference performance, and is expected to address issues such as power constraints and operational workloads. While the introduction of typical GPU servers often requires renovation works for large-capacity power supplies and the preparation of dedicated cooling facilities, IBM Spyre is designed to be installed as an add-on card to IBM Power11 servers. This allows for the addition of AI inference capabilities without significant modifications to power facilities or air conditioning. Moreover, by combining it with IBM Power servers, AI inference can be completed within the server without moving core business data externally, allowing for the construction of an on-premise AI infrastructure that meets the needs of companies emphasizing security and data sovereignty. Based on these features, NTT Integration and IBM Japan have started technical verification regarding the feasibility of an on-premise AI infrastructure utilizing IBM Spyre.
In this technical verification, the effectiveness as an enterprise AI inference infrastructure will be verified by combining IBM Power11 and IBM Spyre. By coordinating the core business data handled by IBM Power and the AI inference processing on the same infrastructure, a configuration that allows AI utilization without moving data will be verified, aiming to realize an on-premise AI infrastructure that balances high security and operational efficiency. In addition, evaluations will be conducted from the perspective of power consumption and operational workload to verify whether it is an AI infrastructure that companies can implement in actual operations.
NTT Integration positions itself as 'Client Zero' in this initiative and will apply IBM Spyre to actual operations. Verifications will be conducted targeting use cases tailored to daily business, such as the analysis of core business data running on IBM i and internal knowledge searches utilizing RAG. The policy is to accumulate practical knowledge in construction, operation, and post-operation improvements, and reflect the results in future services.
FAQ
What is the purpose of this technical verification?
To evaluate the feasibility of an on-premise AI infrastructure that securely and efficiently utilizes AI without moving data, combining IBM Power11 and IBM Spyre.
What is the difference from existing GPU infrastructure?
It allows building a low-power AI inference environment as an add-on to IBM Power servers without requiring major power or cooling upgrades.
What kind of companies is it particularly suitable for?
It is ideal for companies, such as in manufacturing, that want to strictly manage confidential data like design blueprints on-premise without using external clouds.