Summary
- Semidynamics will evaluate Nokia's Data Center Fabric technology for its AI inference racks.
- The proposed work covers technical assessment, testing, validation, and possible integration.
- The agreement is exploratory, with no customer deployment or commercial system confirmed.
Barcelona-based processor developer Semidynamics and Finnish networking supplier Nokia have agreed to examine the integration of AI inference accelerators, liquid-cooled racks, and data-centre networking technology.
The memorandum of understanding establishes a programme of technical assessment and potential validation involving Semidynamics’ RISC-V-based AI computing systems and Nokia’s Data Center Fabric networking products.
Semidynamics will evaluate Nokia’s networking technologies for use with its inference racks, while the companies consider whether their systems could form the basis of future integrated technical and commercial offerings.
The agreement does not establish a supply contract, production deployment, or jointly available platform. No customer, equipment capacity, or installation date has been identified.
Semidynamics is developing processors designed around the demands of AI inference, with particular attention to how computing equipment accesses and moves information through memory.
Inference workloads use trained models to process inputs and generate outputs. Their performance depends on processor capability, but it can also be limited by the rate at which model data and temporary information can be accessed.
When computing units wait for information from memory, additional arithmetic capability may not translate directly into higher application throughput. Processor architectures that address memory access attempt to reduce the time spent waiting for data and improve sustained utilisation.
Semidynamics’ proposed systems combine its processor technology with rack infrastructure and liquid cooling. The collaboration with Nokia adds the communications systems needed to connect computing equipment within a larger installation.
Nokia’s Data Center Fabric technology is intended to support network connectivity between computing systems. Within an AI installation, the network carries traffic associated with applications, storage, coordination between processing systems, and other operating services.
Bandwidth and latency requirements depend on how workloads are distributed. Some inference applications may operate largely within a single processing system, while others require communication between multiple machines or shared infrastructure services.
The appropriate network configuration consequently depends on the application architecture rather than a single performance specification. The companies will need to establish how their respective technologies interact under representative workloads.
Testing complete systems
Integrating processor and networking equipment involves more than confirming compatible physical interfaces. The systems must support suitable software, communications protocols, monitoring, management, and operational fault handling.
Large computing installations require consistent behaviour during equipment maintenance and failures. Network redundancy, service recovery, workload placement, and configuration management all influence whether individual hardware components can form a dependable production platform.
Thermal and electrical integration also affect the proposed rack design. Liquid-cooled systems need suitable connections between the computing equipment and the facility’s cooling infrastructure, while electrical distribution must accommodate the load and operating characteristics of the complete installation.
Neither company has disclosed the power requirements, rack densities, cooling capacities, or network configurations that will be tested. Those details will be necessary before the proposed technology can be assessed against particular data-centre facilities.
The collaboration takes place as European suppliers seek to provide a greater share of the computing and communications hardware used in regional AI infrastructure.
Semidynamics’ use of the RISC-V instruction set provides a processor development foundation, while Nokia contributes networking products and expertise. However, the geographical origin of individual technologies does not by itself establish complete control of semiconductor manufacturing, component sourcing, or software dependencies.
Commercial deployment would still require hardware qualification, manufacturing arrangements, customer commitments, and facilities capable of accommodating the resulting systems.
The partners intend to assess whether integration can provide a technically and commercially viable platform for customers developing European AI infrastructure. Their immediate work concerns feasibility and validation, rather than an already established computing service.
Any later deployment will depend on the results of testing and the ability to supply complete equipment that can be operated reliably within production data centres.
