Via Licensing Alliance sets hourly royalty rates for AI datacenter DRAM

This standardized licensing framework addresses the critical memory bottleneck in AI infrastructure by providing predictable costs for high-bandwidth memory. For the streaming ecosystem, which increasingly relies on AI-driven personalization and efficient cloud transcoding, this move stabilizes the intellectual property landscape for the underlying server hardware. By consolidating essential patents for DDR5 and HBM4 into a single pool, Via reduces the risk of patent litigation that could otherwise slow the deployment of next-generation datacenter capacity. Watch for the participation rate of major memory manufacturers and server OEMs to determine if this becomes the definitive licensing standard for AI-optimized hardware.

Via Licensing Alliance is positioning its new DRAM patent pool within a broader ecosystem of memory-licensing programs that have historically struggled to achieve universal participation. The company, which operates under the Via Licensing umbrella and manages pools across audio, video, and semiconductor technologies, faces competition from established patent administrators. Sisvel International has operated a competing DDR5 and HBM licensing program since early 2025, claiming commitments from Samsung, SK Hynix, and Micron as licensors, though Via's hourly-rate model for server processing units represents a distinct pricing approach aimed at datacenter operators rather than component manufacturers. The divergence in royalty structures between the two pools could create confusion among OEMs and cloud providers evaluating which framework to adopt.

The business case for standardized DRAM licensing has intensified as AI datacenter buildouts accelerate. SK Telecom announced a gigawatt-scale AI Cloud built on NVIDIA DGX SuperPOD architecture in mid-2026, while Deutsche Telekom secured the German federal government's sovereign AI cloud contract, signaling that memory-intensive AI infrastructure is expanding well beyond hyperscalers into telecom and government sectors. Each of these deployments requires high-bandwidth memory covered by JEDEC standards, making patent licensing a material cost consideration. Via's hourly royalty model, which charges per server processing unit rather than per chip, aligns with how cloud providers already meter compute resources and could simplify budgeting for operators running thousands of GPU-accelerated nodes.

On the technical side, the pool's coverage of HBM4 is particularly significant given the memory bandwidth demands of current AI training and inference workloads. Nokia and AWS demonstrated in March 2026 what they called the industry's first agent for network slicing, running on cloud-hosted infrastructure that relies on high-bandwidth memory for real-time processing, illustrating how edge AI workloads are pushing memory requirements beyond traditional datacenter boundaries. Nokia's autonomous networks portfolio, which claims automation rates above 90 percent and service delivery times under four hours, depends on the same DDR5 and HBM architectures that Via's pool now covers. The intersection of patent licensing costs and infrastructure deployment speed will likely determine whether Via's program gains traction among the tier-one operators and cloud providers building out AI capacity through 2027.