Fujitsu and Scaleway Join Forces to Revolutionize AI Inference with Monaka Processor
Fujitsu and Scaleway are collaborating to introduce a groundbreaking approach to AI inference by leveraging the Monaka processor, which aims to enhance performance, uphold European sovereignty, and boost energy efficiency. This partnership signifies a pivotal step toward diversifying cloud infrastructures, with the goal of providing more sustainable options for large-scale AI deployments.
As the adoption of artificial intelligence accelerates across Europe, the need for transformative cloud infrastructures becomes increasingly urgent. Organizations are now challenged to balance performance, cost predictability, and compliance with sovereignty and sustainability requirements. In this context, central processing unit (CPU)-based architectures are regaining interest, fueled by advancements in applied research and rising energy costs associated with specialized infrastructures.
Announced on December 4, 2025, the strategic alliance between Fujitsu and Scaleway is poised to set new benchmarks in the industry. By opting for the Monaka processor, both companies express a shared commitment to providing European organizations with a credible alternative to proprietary accelerators. Monaka offers up to 1.9 times more energy efficiency and 1.7 times greater economic efficiency for inference workloads while ensuring compatibility with very large AI models. This technical positioning also emphasizes a structured approach to security, integration in heterogeneous environments, and adherence to sovereignty expectations.
Monaka: A Chiplet Architecture Designed for Maximizing Inference
Developed by Fujitsu as part of its ongoing high-performance computing advancements, the Monaka processor is specifically engineered to address the challenges of large-scale AI deployments in production. It features a microarchitecture designed to optimize cloud-based AI processing. The many-core architecture incorporates up to 144 Armv9 cores distributed across four chiplets (36 cores per chiplet). In a standard server configuration, two Monaka sockets can be integrated, resulting in a total of 288 available cores.
This chiplet partitioning, along with a separate SRAM memory die, minimizes the active area of the cores on 2 nm silicon and enhances thermal efficiency. The input/output (I/O) interface includes a DDR5 memory controller (12 channels), PCIe 6.0 connectivity, and support for CXL 3.0, catering to the needs of cloud and high-performance computing (HPC) environments.
The significant level of parallelism, with dozens or even hundreds of cores, is particularly well-suited for AI workloads reliant on large model inference (such as language models, deep learning, and scientific computing). By multiplying the cores, Monaka facilitates fine-grained computation, leverages extensive parallelism in matrix and vector operations, and mitigates memory bottlenecks through tight coupling between cores and high-level cache.
Despite lacking integrated high-bandwidth memory (no HBM), the combination of high-speed DDR5, a sizable SRAM cache, and high-speed connectivity via CXL provides a functional alternative: reduced latency, significant memory bandwidth, and opportunities for memory expansion or external accelerator connections. Thus, Monaka prioritizes scalability, computing density, and energy efficiency over sheer raw memory bandwidth.
A Balanced Performance-to-Power Ratio
This architectural choice makes Monaka particularly well-suited for production AI inference, enabling homogeneous scaling, strong performance per watt ratios, and modularity that simplifies infrastructure sizing according to business needs (cloud, HPC, edge, continuous AI).
Monaka’s native compatibility with major open-source AI libraries (such as TensorFlow, PyTorch, and ONNX) is supported by dedicated hardware accelerations, minimizing the need for cumbersome software abstraction layers. This technical choice allows for the direct execution of AI models and ensures easy portability across various cloud infrastructures. Additionally, the architecture includes advanced hardware security mechanisms to protect both data flows and computational loads, which are critical in multi-sector inference contexts or for handling sensitive data.
For Scaleway, integrating Monaka reflects a strategy to diversify its AI infrastructure portfolio. The aim is to offer an alternative without completely replacing Nvidia’s ecosystem. Monaka stands out as a credible option for specific use cases, including regular AI inference, cloud AI production, scalable deployments, and a European context focused on sovereignty, as well as modular or varied workloads. In these scenarios, the performance/cost/efficiency/governance ratio becomes increasingly favorable. However, for intensive AI applications, such as training massive models or ultra-high-performance use cases that require maximum acceleration or infrastructures already optimized for GPUs, the Nvidia ecosystem remains difficult to surpass. This drives the strategic interest among some players to combine Monaka CPUs with Nvidia GPUs to build hybrid architectures that capitalize on the strengths of both.
Towards a More Responsible and Sovereign AI Deployment Model
The collaboration between Fujitsu and Scaleway emerges against a backdrop of increasing fragmentation in the AI infrastructure market, where technological sovereignty and carbon footprint reduction are becoming key concerns. According to Yann-Guirec Manac’h, head of Hardware R&D at Scaleway, “Our commitment to providing unmatched energy efficiency in the cloud remains central to everything we do, driving us to continuously innovate for a sustainable digital future.” This approach responds to a growing demand for alternatives to traditional acceleration solutions, which are often associated with high energy consumption and extraterritorial dependencies.
Assessing Monaka in inference environments reflects a pragmatic approach to testing this technology’s suitability for a diverse range of use cases while ensuring compliance with data management and cybersecurity regulations. Scaleway’s strategy is grounded in complete transparency, allowing customers to access clear performance metrics, operational costs, and the environmental impact of each deployed option. This model supports the industrialization of AI while aligning with the ecological choices of the digital sector.
Credible Alternatives to Hyperscaler Offerings
A central challenge of this partnership lies in its ability to provide credible alternatives to the offerings of American hyperscalers in a market characterized by extreme concentration of computing power. Scaleway claims full control over the value chain, from operating data centers to delivering value-added services, all within a secure European framework. This control facilitates the swift integration of innovations like Monaka and the adaptation of offerings in line with evolving market needs. For Fujitsu, this partnership opens the door to a broader dissemination of its technologies in the European market while emphasizing its expertise in energy performance.
However, some observers might interpret the Fujitsu–Scaleway partnership as a retreat prompted primarily by GPU shortages and uncertainty over access to proprietary accelerators. Yet a deeper analysis reveals a response that is both pragmatic and strategic, potentially structuring the European cloud and AI ecosystem. In the short term, this initiative may resemble a stopgap solution for stakeholders grappling with the scarcity of Nvidia GPUs or the holdback of specialized chips by major American platforms. The proliferation of export restrictions, prioritization of U.S. domestic markets, and the verticalization of cloud offerings push European providers to explore every alternative path to avoid reliance on inaccessible or prohibitively expensive resources.
In this sense, opting for an optimized CPU architecture that is market-available and developed with an open logic appears to be a timely opportunistic response. However, reducing this movement to a mere fallback strategy would be an oversimplification. The true value of this partnership lies in its potential to initiate a transition toward more sovereign computing models, closely aligned with European standards for compliance, security, and sustainability. Monaka is not merely a backup solution; it facilitates large-scale experimentation with reimagined processors for AI, accelerates the diversification of supply chains, and reduces technological dependency on American giants.
In the medium term, this momentum could contribute to the emergence of a new generation of cloud infrastructures that are less vulnerable to geopolitical tensions, more agile, and tailored to the specific constraints of European organizations. The Fujitsu–Scaleway partnership thus represents a hybrid initiative, both a response to immediate market pressures and a precursor to a structural evolution in the sector. The ability of European stakeholders to transform this “stopgap” into sustainable competitive advantage will depend on their capacity to industrialize these solutions, build a software ecosystem around these new architectures, and demonstrate their value in terms of performance, costs, and digital sovereignty.
This pressure on supply chains affects the entire sector, including players like OpenAI, which is actively exploring the possibility of developing or co-developing its own chips to secure its future needs. The market is bracing for a prolonged period of resource tension, where the ability to innovate and integrate alternatives, such as next-generation AI CPUs, could become a decisive competitive edge for European stakeholders.



