BirenTech wins 174M yuan bid for Kunlun Lingyi AI compute project Package 1

Shanghai Biren Technology Co., Ltd. has won the bid for Package 1 of the Kunlun Lingyi Intelligent Computing Center Project (Phase I) with a bid amount of RMB 174 million, becoming the sole winning supplier for that package.

The total procurement budget for the project is approximately RMB 626 million, divided into two packages:

• Package 1: Budget of approximately RMB 174 million
• Package 2: Budget of approximately RMB 452 million

Biren Technology secured Package 1 with a comprehensive evaluation score of 89.71 points, ranking first overall. According to the contract requirements, Biren must complete delivery within 45 calendar days after signing.

Notably, Package 2 was declared invalid due to fewer than three suppliers submitting response documents. The failure of a package with a budget exceeding RMB 450 million due to insufficient participating vendors reflects the current concentration and scarcity of the domestic AI computing power supply chain.

Core Product: Biren Bili Series GPUs

The core product in Biren Technology's winning bid is the Bili series GPU. According to public information, the Bili GPU is based on a classic GPGPU architecture, with its core computing unit—the SPC stream processor—integrating dedicated tensor computing units called T-Cores.

The company has developed a three-generation product matrix:

• BR106: Mass production began in 2023
• BR110: Designed for edge and cloud inference scenarios
• BR166: Adopts chiplet design

The next-generation flagship training and inference chip, BR20X, is targeted for release in 2026, supporting data formats such as FP8 and FP4, with larger memory and high interconnect bandwidth.

Interconnect and Scale-Out Capabilities

Biren Technology has developed its proprietary BLink protocol for direct GPU-to-GPU connectivity, achieving a single-channel bidirectional bandwidth of up to 64 GB/s and supporting up to 8-card point-to-point direct interconnection.

At the scale-out level, the company adopts an OCS (Optical Circuit Switching) solution. As of the end of 2025, Biren has successfully delivered a 2048-card optical interconnect GPU supernode cluster, achieving a linear acceleration ratio of 95% for thousand-card clusters.

Software Ecosystem

Biren Technology has achieved Day 0-level adaptation for major large language models, covering leading domestic model developers including:

• Zhipu AI (GLM)
• DeepSeek
• Alibaba (Qwen)
• Moonshot AI (Kimi)

Its BIRENSUPA software platform supports mainstream frameworks such as PyTorch, TensorFlow, and PaddlePaddle, as well as inference frameworks like vLLM.

Editor: Lin Zi, Chip Intelligence