China’s Domestic GPU Players Line Up for the New Year

Biren Technology, a general-purpose GPU unicorn valued at 20.9 billion yuan, has just passed its listing hearing at the Hong Kong Stock Exchange. It is now close to ringing the bell as it races to become Hong Kong’s “first domestic GPU stock.”

Founder Zhang Wen, a Harvard Law School doctoral graduate, launched the company in 2019. Before that, he was widely known as president of SenseTime.

The company secured angel funding just three months after its founding. Over six years, it has raised no fewer than 10 rounds, backed by well-known investors including Qiming Venture Partners, IDG Capital, Hillhouse Venture Capital, and Gree Venture Capital.

Including Moore Threads and MetaX, which recently listed on China’s A-share market, three domestic GPU companies have delivered positive capital-market news this month.

What Does Biren Technology Do?

Biren Technology is an AI infrastructure computing company that develops general-purpose GPU chips and intelligent computing solutions, providing full-stack support from cloud to edge for AI training and inference.

Its core product, an integrated intelligent computing solution, consists of two parts:

The first is a hardware system based on its self-developed GPGPU architecture, designed for AI workloads, especially LLMs, and supporting multimodal fusion and large-scale model training.

Its GPGPU chips are compatible with mainstream third-party software platforms and can also be integrated into industry-standard hardware form factors such as PCIe cards and OAM modules.

PCIe cards, OAM modules, GPGPU servers, and server clusters in different configurations together make up Biren Technology’s GPGPU hardware system.

There is no question that GPGPU chips are the critical component in this system.

So far, Biren has developed two chips, the Bili 106 and Bili 110. The Bili 106 is used for training and inference, with support for virtualization and a hardware security engine. The Bili 110 is used for edge inference, is optimized for energy efficiency, and targets scenarios such as industrial control and robotics.

Biren has also launched the higher-performance Bili 166 chip by co-packaging two Bili 106 chip dies, using chiplet technology and die-to-die interconnect technology.

Under its future roadmap, Biren plans to launch the Bili 20X series, developed on its second-generation architecture, for cloud training and inference, with commercialization expected to begin in 2026. It also plans the Bili 30X and Bili 31X chips, focused on cloud-edge collaboration, with launches around 2028.

The second product is its self-developed BIRENSUPA software platform, built on top of its GPGPU and used to develop AI applications.

The platform uses a layered architecture covering drivers, libraries, programming platforms, machine learning frameworks, and solutions:

At the base layer, drivers enable hardware recognition and cross-architecture data transfer, while virtualization software supports the partitioning of physical devices to secure AI applications.

The programming platform integrates the BIRENSUPA model, a self-developed compiler, and libraries such as suDNN and suBLAS, while maintaining compatibility with mainstream architectures and improving development efficiency.

The framework layer supports open-source frameworks such as PyTorch and TensorFlow, and provides a high-performance inference engine as well as Elastic Toolkit and HGCT modules to support large-scale cluster training and heterogeneous collaboration.

The solutions layer includes the BIRENCUBE cloud management platform and Model Zoo, offering ready-to-use AI deployment and pretrained model services so customers can put applications into production quickly.

Biren Technology also integrates its hardware system and software platform with servers, storage, networking equipment, and other hardware infrastructure provided by partners to form intelligent computing clusters, which it offers to customers as an integrated solution.

The company’s products mainly serve AI large-model training, data analytics, intelligent computing, and related use cases.

Its customers include large enterprises in high-computing-power industries such as AI data centers, telecoms, fintech, and the internet, including nine Fortune China 500 companies, such as the three major telecom operators.

In the first half of this year, Biren sold 2,216 Bili 106 chips and 22 Bili 110 chips, both up from the same period in 2024.

Biren Technology operates under a “fabless model,” meaning it does not build its own factories. Instead, it focuses more of its energy on R&D and design, while outsourcing chip and hardware system production to third parties.

That has also laid the groundwork for improving Biren Technology’s financial metrics.

How Do the Financials Look?

From 2022 to 2024, Biren Technology’s revenue rose from 500,000 yuan to 337 million yuan. Revenue in the first half of this year was 58.9 million yuan, up about 50% year on year.

Breaking down its revenue structure, the company’s main sources of revenue include:

Product sales, including intelligent computing solutions and agency fees

Support services or extended warranty services

Leasing revenue from intelligent computing clusters

Intelligent computing solutions are the core revenue source. This revenue was formally recognized starting in 2023, while all revenue in 2022 came from agency fees. Intelligent computing clusters also began generating leasing revenue in the first half of this year.

On profitability, from 2022 to 2024 the company reported gross profit of 500,000 yuan, 47.4 million yuan, and 179.2 million yuan, respectively. The corresponding gross margins were 100%, 76.4%, and 53.2%.

In the first half of this year, Biren’s gross profit was 18.77 million yuan, with a gross margin of 31.9%.

Gross margin volatility is itself a normal feature of early commercialization. Biren explained the decline in its prospectus as follows:

In 2022, its core business had not yet begun generating revenue, so the 100% gross margin did not reflect the performance of its main business.

Subsequent fluctuations in gross margin were consistent with changes in the gross margin of its intelligent computing solutions, mainly due to shifts in the product sales mix driven by customer-specific needs.

In 2023, the company provided one major customer with a server cluster solution. In 2024, revenue mainly came from PCIe card sales. In the first half of this year, entry-level products contributed more revenue.

At the same time, Biren Technology remains loss-making. In 2022, 2023, 2024, and the first half of this year, the company recorded losses for the period of 1.474 billion yuan, 1.744 billion yuan, 1.538 billion yuan, and 1.6 billion yuan, respectively.

Adjusted net losses over the same periods were 1.038 billion yuan, 1.051 billion yuan, 767 million yuan, and 552 million yuan, respectively.

On the expense side, R&D spending is by far the largest item.

From 2022 to 2024, Biren’s R&D expenses were 1.018 billion yuan, 886 million yuan, and 827 million yuan, respectively. In the first half of this year, R&D spending was 572 million yuan, up 44% year on year.

The company has 657 R&D employees, accounting for about 83% of total staff. Among them, 78% hold master’s degrees or higher from well-known universities.

As of the end of the first half of this year, the company had 1.285 billion yuan in cash and cash equivalents.

Who Is Leading the Company?

Harvard PhD Founder Builds a 20.9 Billion Yuan Unicorn

Biren Technology was founded in 2019 by Zhang Wen, who holds an MBA from Columbia University and a doctorate in law from Harvard Law School.

He previously served as chairman and CEO of Shanghai Dingyu Hengrui Equity Investment Fund Management Co., Ltd., and as CEO of Enraytek Optoelectronics Technology, a company under SMIC founder Zhang Rujing. He later served as president of SenseTime, where he led the establishment of SenseTime’s headquarters in Shanghai.

Early members who joined Biren also include current CTO Hong Zhou, COO Zhang Linglan, and general manager and legal representative Xiao Bing, all of whom joined less than six months after the company was founded.

Hong Zhou was previously engineering director at U.S. graphics chip company S3, Inc. He also served as a chief architect at NVIDIA and as chief architect at Futurewei Technologies, Huawei’s U.S. research center.

Before joining Biren, Zhang Linglan was a GPU SoC architect at AMD and a senior R&D manager at Samsung Electronics’ U.S. R&D center.

Xiao Bing may also be familiar to many readers. He was previously vice president of business development at SenseTime and has held China-region executive roles at several overseas-listed companies, including IBM and Petuum.

This core team has made Biren a long-running favorite in the capital markets.

Before the IPO, the company had completed 10 funding rounds. It secured angel financing just three months after its founding, at a post-money valuation of about $130 million, or roughly 900 million yuan.

After completing its Pre-B+ round one year after founding, Biren’s post-money valuation reached 7.15 billion yuan, making it a GPU unicorn. Eighteen months after launch, its cumulative financing exceeded 4.7 billion yuan, setting a fundraising record for a domestic chip startup.

Its investor lineup is packed with marquee names, including Qiming Venture Partners, IDG Capital, Walden International China, Ping An Group, Hillhouse Venture Capital, Gree Venture Capital, Songhe Capital, Yunhui Capital, Guosheng Capital, and China Merchants Capital.

By August this year, Biren’s implied post-money valuation had reached 20.9 billion yuan.

Before the IPO, chairman and CEO Zhang Wen directly held an 8.68% stake.

Domestic GPU companies have recently been crossing the IPO finish line together. Moore Threads and MetaX both listed on the A-share market recently and surged on debut.

With the earlier-founded Biren Technology set to claim the title of Hong Kong’s “first domestic general-purpose GPU stock,” what kind of first-day showing will it deliver?