What Drives Kunlun Chip’s 10‑Year Rise in China’s AI Hardware Market?

Introduction

Kunlun Chip celebrated its first anniversary at Baidu Tech Salon’s "Intelligent Chip" session, where four experts unveiled a ten‑year journey of Chinese AI silicon development and presented the latest achievements of two chip generations.

Development History

Founded in June 2021, Kunlun Chip’s core team dates back to 2011, initially building AI accelerators on FPGA. By 2017 the self‑designed Kunlun XPU architecture was mature and presented at Hot Chips. In 2018 the project, then called Baidu Kunlun, shifted from FPGA to AI ASIC, leading to two generations of chips reaching mass production by 2021.

Company Advantages

Large‑scale deployment and engineering experience : Over 20,000 chips have been deployed across various sectors, proving the technology’s industrial viability.

Stable core team and deep technical depth : A balanced hardware‑software team with extensive academic and engineering contributions.

Deep scene understanding and holistic product view : Originating from Baidu, the team leverages real‑world AI workloads to guide architecture decisions.

Fully self‑designed architecture and top‑tier software stack : 100% in‑house XPU design combined with a comprehensive software ecosystem.

Mission and Vision

The mission "Make computing smarter" is interpreted as delivering faster, stronger, and more efficient computation to empower industries. The vision is to become a epoch‑making, globally leading intelligent computing company.

Founding Team

CEO Ouyang Jian, former Baidu chief architect, leads a team with extensive front‑line AI scenario experience, providing strong R&D capabilities.

Honors

The team has published pioneering papers in international computer architecture conferences, including four Hot Chips papers, and both chip generations have received China Chip "Outstanding Technology Innovation Product" awards.

Product Roadmap

Two generations of chips have been mass‑produced (first generation and 7nm second generation). Development of a third generation using 4nm process is underway.

Product Advantages

Massive deployment: over 20,000 chips in production.

Fully self‑designed architecture.

Flexible software stack enabling rapid development and cross‑hardware migration.

Second‑Generation Chip (Kunlun XPU‑R)

Built on a 7nm process, the second‑generation chip features the XPU‑R architecture with high compute density, hardware virtualization, and distributed system capabilities. It supports major AI frameworks such as PaddlePaddle, PyTorch, TensorFlow, and ONNX, and can be deployed in public cloud, private cloud, and edge environments.

Performance tests show the R200 inference board delivers 1.5‑2× higher throughput than leading 150W GPUs, delivering tangible cost‑saving and efficiency gains for customers.

Application Scenarios

The chips are deployed in:

Internet services : Baidu search and related products, powering billions of daily queries.

Smart cities : City brain projects such as Yichang’s environmental monitoring.

Bio‑computing : Protein structure prediction and gene analysis leveraging flexible XPU design.

Smart industry : Quality inspection and other Industry 4.0 use cases, achieving up to 6.5× ROI over traditional solutions.

Additional sectors include finance, transportation, emergency response, and AI‑driven data centers.

Continuing the "China Chip" Vision

By combining large‑scale deployment, self‑reliant technology, and industry‑focused innovation, Kunlun Chip strives to create domestic AI silicon that delivers real economic value and positions China as a leader in intelligent computing.