Chinese AI chipmaker Biren seeks up to $624m from HK IPO

Shanghai Biren Technology, a Chinese AI chipmaker, has started bookbuilding for its Hong Kong IPO and targets to raise up to US$624 million.

The company, which develops graphics processing units (GPUs), is offering 247.7 million shares at HK$17 (US$2.2) to HK$19.60 (US$2.5) each, and plans to begin trading on January 2, 2026.

Biren’s IPO has attracted 23 cornerstone investors, including major asset managers, mutual funds, insurers, and hedge funds, who have committed to buy US$372.5 million worth of shares with a six-month lock-up.

Founded in 2019, Biren reported 336.8 million yuan (US$47.8 million) in sales in 2024 and 58.9 million yuan (US$8.3 million) in sales for the first half of 2025, but its net loss for the same period rose 32.3% to nearly 9 billion yuan (US$1.28 billion) year-on-year.

🔗 Source: South China Morning Post

🧠 Food for thought

Implications, context, and why it matters.

Biren’s IPO pushes a growth bet with unproven manufacturing at scale

Biren posted 336.8 million yuan (~US$47 million) in 2024 revenue, and its H1 2025 net loss rose 32.3% year-on-year to nearly 9 billion yuan.
The flagship BR100 GPU was built on Taiwan Semiconductor Manufacturing Company (TSMC)’s 7nm process 1. Production moved to Semiconductor Manufacturing International Corporation (SMIC) after US export controls barred TSMC manufacturing in October 2022 1. That puts yield rates (the share of chips that work), capacity, and volume targets in doubt.
Biren says the BR100 set a “global computing power record” 1. Many Chinese AI firms still stick with Nvidia’s CUDA (its proprietary GPU programming platform) despite export limits 2.

Asian cloud and systems integrators can win CUDA migration work as Chinese firms hedge Nvidia risk

With export limits on Nvidia chips 2, enterprises running AI will need services to move applications from CUDA to the BIRENSUPA platform 1 or Huawei’s options. That shift boosts demand for kernel optimization (tuning low-level code), benchmarking, and machine learning operations (MLOps) tools for fragmented Chinese hardware.
In December 2023, Biren and Infinigence AI, an AI developer in China, nearly doubled Large Language Model (LLM) training performance 1. Independent software vendors with cross-platform AI frameworks for Nvidia and Chinese chips can serve risk-averse customers seeking to limit supply chain exposure.
Clouds hosting Biren GPUs can stand out with hybrid setups for Chinese and Western chips, appealing to multinationals needing China-compliant AI capacity without dropping Nvidia’s ecosystem 2.