AI Scale-out Networking

About AI Scale-out Networking

The network infrastructure connecting GPUs, accelerators and storage systems plays a key role in enabling data centres to handle AI and high-performance computing workloads. The AI Scale-out Networking programme is designed to compensate the innovators that made these foundational technologies possible and provide licensing clarity to the data centre ecosystem.

The programme addresses three technologies:

InfiniBand: An industry standard, channel-based switched fabric interconnect architecture for server and storage connectivity. With its ultra-low latency, native RDMA support and lossless transmission, InfiniBand supports the world’s most demanding environments, from powerful supercomputers to large AI clusters. The InfiniBand Architecture Specification is published and maintained by the InfiniBand Trade Association.

RoCE v2: RDMA over Converged Ethernet version 2, a protocol that brings remote direct memory access capabilities to standard Ethernet networks. An evolution of the original RoCE specification released in 2010, RoCEv2 has played a key role in enabling hyperscale data centre deployments by providing routing across Layer 3 networks. The RoCE v2 standard is defined within the overarching InfiniBand Architecture Specification published and maintained by the InfiniBand Trade Association.

Ultra Ethernet: An open, interoperable networking technology designed to meet the increasing bandwidth and latency demands of AI and high-performance computing. Developed by the Ultra Ethernet Consortium (UEC) – a group of leading hyperscalers, chipmakers and networking vendors – the specification defines how wired connections can move data at unprecedented speed and scale. The UEC v1.0 specification was launched in June 2025 and v1.0.2 was released in January 2026.

Together, these technologies serve the full spectrum of AI training and inference environments, high-performance computing and cloud infrastructure – essentially anywhere that compute-intensive workloads require fast, reliable and scalable networking.

Sisvel has the track record and expertise to create a successful pool

Sisvel is Europe’s largest and longest-established patent pool operator, with a track record of creating successful licensing programmes around networking technologies.

The company’s Wi-Fi 6 pool secured more than 40 licensees within a three-year period, including industry giants such as Acer, Cisco, HP, Huawei and Netgear. This success led to the January 2026 launch of Sisvel Wi-Fi Multimode, a successor pool covering both Wi-Fi 6 and Wi-Fi 7. Membership expanded to 10 patent owners and deals have been announced with licensees including Sony and HP Enterprise. These were breakthrough programmes in a technology area that has long been known for complex and opaque licensing.

The AI Scale-out Networking programme represents a natural progression from wireless to wired networking technology for the Sisvel team, which has invested considerable resources in understanding the technical and business landscape.

The Sisvel Tech unit – which is unique in the industry – has conducted detailed analysis of the specifications, mapped required and optional features and identified relevant patent portfolios. Meanwhile, our business development team has engaged with stakeholders across the industry to understand the wired networking licensing environment.

This rigorous preparation will enable us to ensure that should we proceed to pool formation, the licence offer that we put together will be compelling to both licensors and licensees.