Fujitsu has announced the expansion of its strategic collaboration with NVIDIA to deliver a comprehensive AI infrastructure with built-in AI agents and high-performance optimized hardware. The goal is to accelerate enterprise adoption and drive industrial transformation. The initiative — summarized by Fujitsu’s European branch from the corporate press release dated October 3 — highlights a dual approach: a sector-ready AI agent platform and an integrated computing architecture that combines Fujitsu MONAKA CPUs with NVIDIA GPUs via NVIDIA NVLink-Fusion.
This move seeks a delicate balance: to provide companies with the speed and efficiency demanded by the new wave of use cases in healthcare, manufacturing, and robotics, without compromising corporate autonomy in AI usage and governance. Behind the rhetoric, the promise is clear: a platform of agents that learns and improves continuously, delivered as NVIDIA NIM microservices, and an IT infrastructure designed from the silicon up to scale reliably.
“Fujitsu’s strategic partnership with NVIDIA will accelerate AI-driven business transformation,” states Takahito Tokita, Fujitsu’s CEO and President. “By combining the cutting-edge technologies of both companies, we will develop and deliver a full-stack AI infrastructure, starting with sectors like manufacturing, where Japan is a global leader.”
“The AI industrial revolution has begun, and we need to build the necessary infrastructure to propel it,” emphasizes Jensen Huang, NVIDIA’s Founder and CEO. “Together, NVIDIA and Fujitsu are connecting and expanding our ecosystems to forge a powerful alliance for the AI era.”
Two Pillars: AI Agents and Optimized Computing
The collaboration revolves around three lines of work, with two main axes:
1) Self-evolving AI Agent Platform (Leading Industries)
Fujitsu and NVIDIA will co-develop a platform of AI agents designed for multi-company and multi-environment realities. Its purpose is to balance speed and security, enable multi-tenancy, and accelerate sector- and client-specific customization. Technologically, it involves:
- The platform relies on Fujitsu Kozuchi and its AI workload orchestration technology, integrated with NVIDIA Dynamo.
- The autonomous evolution of agents and models leverages NVIDIA NeMo and Fujitsu’s multi-agent technologies, including the Takane Model optimization.
- Customer delivery will be via NVIDIA NIM microservices, promoting faster deployments and optimized inferences.
The focus is not solely on “model layer”: it envisions a human-AI co-creation that integrates human judgment and creativity with accelerated computing for real-world tasks — from digital twins in factories to physical AI in robots for manufacturing or logistics.
2) “Silicon-From-Infrastructure” Approach (NVLink-Fusion + MONAKA)
The second pillar is a state-of-the-art AI infrastructure developed collaboratively:
- Integration of the Fujitsu MONAKA CPU series with high-performance NVIDIA GPUs, interconnected via NVIDIA NVLink-Fusion.
- Platform optimized from the silicon layer with zeta-scale performance goals, aimed at broad industrial adoption.
- Integrated software fusing Fujitsu’s high-speed ARM software technology with the NVIDIA CUDA ecosystem, to provide end-to-end support for HPC + AI.
The ambition is clear: to build a convergent technological foundation that enhances power efficiency, reduces latencies between CPU and GPU, and facilitates orchestration of complex workloads, ranging from multimodal training to low-latency inference in operational environments.
Why Focus on Agents?
The popularity of generative AI has opened doors to increased productivity and creativity, but its real-world deployment faces cost and complexity challenges. AI agents — capable of reasoning, planning, executing, and learning about business flows — are the next value layer: automating tasks that exceed the simple prompt-response of traditional assistants. Through this alliance, Fujitsu and NVIDIA propose:
- Sectorspecific agents (healthcare, manufacturing, robotics) pre-tuned to accelerate time-to-market.
- Continuous learning mechanisms (based on NeMo and Fujitsu’s own capabilities) for supervised self-improvement.
- SaaS microservices via NIM, shortening lab-to-production gaps and simplifying observability and lifecycle management.
In healthcare, this enables administrative process automation, clinical decision support, or image analysis within security frameworks. In manufacturing, digital twins for planning and predictive maintenance. In robotics, physical AI systems that interpret environments and act precisely during operations.
Business Autonomy and Responsible Governance
The announcement underscores a crucial point for corporate fabric: the importance of preserving organizational autonomy in AI usage. The proposal aims for companies to control their data, oversee exposure to external services, and define their governance policies (access, auditing, privacy, security). The multi-tenancy setup and NIM delivery should facilitate policies of isolation, compliance, and observability, especially critical in regulated industries.
An Ecosystem to Turn Promises into Systems
The partnership aims not just at proofs of concept. Among the key initiatives is the creation of a robust ecosystem of partners to expand agent and model use, with joint programs to accelerate transformative use cases. The roadmap envisions starting in Japan and scaling globally: public sector, industry companies, and services.
The underlying goal by 2030 is to establish an AI infrastructure regarded as indispensable for Japanese society’s digital transformation, contributing to competitiveness, growth, and sustainability. Physical AI (perception-decision-action with sensors and actuators) appears here as a leverage for operational automation in a context of labor shortages.
Implications for the Market (and Why Now)
The concept of a “full-stack infrastructure” isn’t new, but the timing is. After the first wave of generative AI, companies demand less friction in deployment, predictable costs, and performance guarantees. The convergence of agents, accelerated computing, and HPC-AI software has the potential to:
- Reduce the time from proof of concept to production environment, thanks to NIM microservices and refined orchestration.
- Optimize TCO (Total Cost of Ownership) via high-speed coupling (NVLink-Fusion), optimized software, and CPU-GPU balance.
- Scale with control, maintaining multi-tenancy and governance adapted to each sector.
For Fujitsu, this enhances its position as a leading integrator in Japan and Europe, backed by its supercomputing expertise and managed services. For NVIDIA, it deepens its solutions layer — agents, NIM, development platforms — built on its hardware foundation, and strengthens its industrial ecosystem where reliability and operability are as critical as TFLOPS.
Looking Ahead: Use Cases, Challenges, and Metrics
Short-term use cases
- Manufacturing: digital twins for line optimization, vision-based QA, and collaborative robotics with physical AI.
- Healthcare: workflow automation and diagnostic support under data governance.
- Robotics: planning and control with robust real-time perception.
Operational challenges
- Governance and security: Ensuring isolation, traceability, privacy, and resilience to incidents.
- Technology integration: Orchestrating each tech stack (Kozuchi, Dynamo, NeMo, NIM) with legacy systems and mission-critical applications.
- Energy efficiency: Improving power efficiency in data centers and industrial edge environments.
Key metrics
- Latency and inference under real load, agent deployment time, automation rate per process, operational SLA, and OEE improvements (Overall Equipment Effectiveness) in manufacturing.
- Compliance and auditability: Ability to demonstrate control and explainability to regulators.
An Industrial Policy and Sustainability Message
Fujitsu anchors the collaboration within its corporate purpose — making the world more sustainable through reliable innovation — aligning it with the SDGs (Sustainable Development Goals). The narrative links productivity and competitiveness with social impact: responsible automation, strengthening industrial fabric, new skills for workers, and better public services.
From a geostrategic tech view, the proposal also advocates “autonomy through partnerships”: collaborating with a global AI acceleration leader (NVIDIA), while ensuring that companies and governments retain control over how, where, and under what rules their agents and data operate.
Frequently Asked Questions
What is NVLink-Fusion, and why is it important in this partnership?
It’s a high-speed interconnect that couples NVIDIA GPUs with CPUs (specifically FUJITSU-MONAKA), reducing bottlenecks between general compute and accelerated workloads. In AI and HPC applications, this bandwidth and low latency are key to sustained performance and power efficiency.
How are AI agents delivered to clients?
As NVIDIA NIM microservices, enabling agile deployment, scaling, and standardized operation (with observability, updates, security). They leverage Kozuchi and Dynamo for orchestration, and NeMo for models and adaptation capabilities.
What does “physical AI” mean in this context?
Refers to AI systems that perceive their environment via sensors, make decisions, and act through actuators (e.g., robots). It’s essential in industrial automation and logistics, requiring real-time perception-logic-action loops.
How is corporate autonomy preserved when using AI?
The platform supports multi-tenancy, governance, and security so each organization can control its data, access policies, and traceability. Delivered as NIM and integrated with high-performance software, it ensures monitoring and compliance, especially in regulated industries.
via: fujitsu