Kenyi startup develops DPU chiplet to break Intel virtual RAN monopoly

This development represents the first viable attempt to provide Arm-based infrastructure with the hardware parity needed to challenge Intel’s dominance in the virtual RAN (vRAN) market. For the streaming and telco ecosystem, this shift could reduce proprietary silicon lock-in, enabling more cost-effective and energy-efficient edge processing at the network's L1 layer. If successful, Kenyi’s hardware acceleration would allow vendors like Ericsson to finally port their software stacks to Arm without sacrificing capacity. Watch for the commercial rollout of the 2nm ADP620 platform to see if Arm-based servers can successfully displace x86 in live commercial 5G sites.

The battle for vRAN silicon dominance comes as the overall radio access network market faces significant revenue contraction, falling from $45 billion in 2022 to approximately $35 billion by 2025. Per Light Reading (August 2026), the shift toward virtualization is accelerating as vendors like Samsung abandon purpose-built hardware in favor of software-defined architectures. While Intel has maintained a near-monopoly on vRAN deployments for years, its competitors are pursuing diverse strategies to reclaim market share. Nokia, for instance, has pivoted toward an AI-RAN strategy leveraging Nvidia’s GPUs for Layer 1 acceleration, recently announcing spectral efficiency gains of over 20% through this approach, per a Nokia July 2026 release.

Intel is not standing still, however. In early 2025, the company launched its Xeon 6 SoC with integrated vRAN Boost, which it claims delivers up to a 2.4x capacity gain and a 70% improvement in performance per watt compared to previous generations, per Intel (March 2025). Meanwhile, Samsung has successfully deployed over 53,000 vRAN sites globally as of early 2026, largely powered by Intel infrastructure, though it has recently begun exploring non-accelerated paths using high-core-count AMD EPYC processors for specific multi-cell configurations to reduce hardware dependency.

The entry of Kenyi Technologies introduces a third viable path: the modular chiplet. By disaggregating the accelerator from the CPU, Kenyi aligns with a broader industry trend toward heterogeneous integration. This move is supported by a 2026 MoU with ADTechnology to utilize 2nm process technology, targeting high-performance edge computing and telco workloads. This approach specifically targets the 'lookaside' architecture preferred by Ericsson, which keeps primary functions on the CPU while offloading only the most taxing computations to external silicon, potentially offering a middle ground between Intel's integrated approach and Nvidia's heavy GPU-centric model.