Baidu's AI chip company targets $50 billion IPO

Kunlunxin heads for a Hong Kong listing

Baidu's semiconductor unit Kunlunxin is reportedly targeting an IPO on the Hong Kong Stock Exchange at an implied valuation of $50 billion, according to The Information. The company was established as an internal unit within Baidu before being spun off as a separate entity with the goal of driving China's push for homegrown AI hardware.

The listing represents one of the most ambitious capital raises in the Chinese semiconductor segment in years, arriving at a time when US export restrictions continue to shut Chinese players out of Nvidia's most advanced products.

Prospective investors are reportedly being asked to commit to semiconductor purchases as a condition of participating in the IPO.

P800: Powerful enough to capture Chinese market share?

Kunlunxin's third-generation chip, the P800, was unveiled in April 2025 and is the company's strongest card heading into the IPO. The chip is manufactured on a 7nm process, likely from SMIC, and delivers around 345 TFLOPS of FP16 performance — a figure that, according to analysis from Guosen Securities, places it on par with Huawei's Ascend 910B and Nvidia's A100.

Compared with Nvidia's H100 — launched in 2022 and delivering up to 989 TFLOPS dense FP16 — the gap remains significant. Expert analyses cited in available research material estimate that the P800 offers around one-third of the H100's computational performance. The H100 also uses HBM3 memory with bandwidth of up to 3.35 TB/s, while Kunlunxin's chips primarily rely on GDDR6 according to available information, which provides substantially lower memory bandwidth.

Architectural strengths and scalability

Despite the raw performance deficit, the P800 has characteristics that are relevant for large-scale deployment in China. The chip physically separates communication and compute units, enabling parallel data transfer and calculation. Kunlunxin states that this delivers over 90 percent scaling efficiency in clusters of more than 5,000 units.

Baidu has reportedly already deployed clusters of 30,000 P800 chips for training large language models, including the company's own Ernie 5.1. The chip has also been used to train DeepSeek-style models with hundreds of billions of parameters. CUDA compatibility is a deliberate priority, intended to lower the barrier for companies looking to migrate away from Nvidia infrastructure.

Market climate and investor appetite

The timing of the IPO is not without its challenges. The Fear & Greed Index for global markets stands at 15 out of 100 as of June 30, 2026, indicating widespread risk aversion among investors. In such an environment, pricing a growth company operating in a geopolitically sensitive segment is no easy task.

It has also not been confirmed by Kunlunxin itself whether or when a listing will formally take place. The Information is a credible industry publication, but the details come from unnamed sources and should for now be treated as unconfirmed.

What it means for the global AI chip market

A successful IPO at close to a $50 billion valuation would place Kunlunxin among the largest semiconductor companies in Asia by market capitalisation, and send a clear signal that China's domestic AI hardware industry is maturing faster than many Western analysts had anticipated.

For Nvidia, this does not represent an immediate threat scenario in Western markets, where the H100 and eventually the Blackwell architecture remain dominant. But in the Chinese domestic market — where export restrictions have effectively shut Nvidia out of its most advanced products — Kunlunxin and Huawei's Ascend could together fill a vacuum that grows larger with each passing quarter.

The source of this report is Yahoo Finance, which is relaying information from The Information.