Industry Insights 12 min read

Who Leads the DPU Market? A Deep Dive into Global and Chinese Players

This article examines the highly concentrated DPU market, highlighting the dominant global vendors Nvidia, Broadcom and Intel, and provides a detailed analysis of emerging Chinese manufacturers—including NebulaX, Paratus, xFusion, Chiplet, and K2—covering their architectures, performance claims, and strategic positioning.

Architects' Tech Alliance
Architects' Tech Alliance
Architects' Tech Alliance
Who Leads the DPU Market? A Deep Dive into Global and Chinese Players

The DPU (Data Processing Unit) market is highly concentrated, with international giants Nvidia, Broadcom and Intel holding roughly 55%, 36% and 9% of the domestic Chinese market respectively, according to data from Toubao Research Institute. Chinese DPU vendors are gaining traction, and the article focuses on five of them.

International Leaders

Nvidia leads with its BlueField series, now in the third generation. BlueField‑2 integrates eight 64‑bit ARM A72 cores, a 2VLIM accelerator and a ConnectX‑6Dx smart NIC, delivering up to 200 Gbps single‑port or 100 Gbps dual‑port networking. It supports RDMA/RoCE, DPU Direct, elastic storage, block‑level encryption and malicious‑application detection, freeing a single DPU from 125 CPU cores. BlueField‑3, released in 2022, adds Ampere‑based GPUs and AI acceleration, while BlueField‑4 is expected in 2023 with a claimed 600× performance boost (up to 75/400 TOPS and 400 Gbps).

Chinese Vendors

Star Cloud (NebulaX D1055AS) : Launched in July 2021, this is China’s first fully hardware‑accelerated DPU. It is a PCIe card targeting bare‑metal, VM and container workloads, providing network and storage offload to improve performance, reduce host CPU usage and simplify IaaS operations. The control plane runs on a generic CPU + Linux stack with OVS, EBS‑Client and other services, supporting out‑of‑band management and easy integration with OpenStack or Kubernetes.

Dayu Smart Chip (Paratus 2.0) : The second‑generation product, released in October 2022, combines an ARM SoC with an FPGA. It adds 10 Gbps/25 Gbps/100 Gbps interfaces and leverages FPGA‑based packet processing for high‑performance forwarding. Software openness is emphasized through VirtIO support, enabling full OVS offload, storage initiator offload, NVMe‑over‑TCP/RDMA, and cloud‑management plugins with a dedicated BMC for automated deployment.

Yunmai Chiplink (xFusion50) : Unveiled on 31 May 2022, xFusion50 is the first domestically developed multi‑scenario RDMA smart NIC. It features a programmable congestion‑control algorithm, HyperDirect technology for GPU‑Direct RDMA (bypassing the CPU), and full network/storage offload, supporting cloud VPC, distributed storage (NVMe‑oF) and VirtIO‑based elastic networking.

Chiplet Source : Offers a fully proprietary DPU built on an NP‑SoC architecture that mixes a general‑purpose ARM core with a RISC‑V‑based NP data‑plane core. The chiplet modular design targets 400 Gbps+ performance, low power consumption and cost efficiency, and enables easy integration of third‑party dies via die‑to‑die interconnect.

China Science Yushu (K‑Series) : Developed the K1 (first generation, taped out in 2019) and K2 (taped out early 2022) DPU chips, and is now working on the third‑generation K2 Pro. Their HADOS platform provides a unified software‑defined accelerator stack, high‑bandwidth network protocol processing, heterogeneous compute scheduling, and a virtualized hardware platform that supports full network, compute and storage virtualization.

Domestic cloud providers such as Huawei, Alibaba, Baidu and Tencent also design or acquire DPU solutions for data‑center, storage and security workloads, further intensifying competition.

Overall, the DPU landscape is shifting from a few global incumbents to a more diversified ecosystem, with Chinese companies emphasizing open software stacks, programmable hardware (FPGA, chiplet) and specialized offload capabilities to capture niche market segments.

Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

HardwareNetworkingMarket analysisAI accelerationDPUData Processing Unit
Architects' Tech Alliance
Written by

Architects' Tech Alliance

Sharing project experiences, insights into cutting-edge architectures, focusing on cloud computing, microservices, big data, hyper-convergence, storage, data protection, artificial intelligence, industry practices and solutions.

0 followers
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.