Initiation: Built for compute

Research evidence excerpt

Initiation: Built for compute

Macquarie Equity Research Biren Technology

Investment thesis

Standout on high computational power

Biren Tech, founded in 2019 in Shanghai, began R&D on its first GPGPU chip in 2020. Its chip

portfolio leverages relatively high computational power and other specs, and based on our

estimates, its first chip BR106 (tape-out in 2021) already was equipped with 500+ TFLOPS

of compute power (based on FP16). Biren's newer chip products BR166 (mass production in

2H2025) has been met with solid domestic customer demand.

Chip interconnect and computing clusters

BR166 is the first domestic product to package two computing dies using 2.5D chiplet

technology. Besides, riding on its SuperPod technology, Biren has delivered 2,048-card

computing clusters to the domestic customers, and we believe larger-scale intelligent

computing clusters are being developed.

Software ecosystem

The company has established the "BIRENSUPA" software stack, consisted of multiple layers

(drivers, libraries, programming platform, machine learning framework, and solutions). This

software system has achieved CUDA-compatibility, and is capable of adapting to domestic

large models to maximise Biren's GPGPU capabilities.

Why we are confident

We like Biren's GPGPU product portfolio skewed towards higher computational power, chip

interconnect, and large-scale computing clusters. Moreover, the company's focus on the

domestic supply chain is likely to bear fruit to facilitate its new product launch, in our view.

Company background

Biren Tech is a leading

domestic GPGPU player in Biren Tech is one of China’s leading domestic GPGPU firms, leveraging its self-developed

China, leveraging its self- hardware chips and proprietary BIRENSUPA software ecosystem, which is CUDA-compatible.