Baidu's AI Chip Subsidiary Kunlun Chip Files for Hong Kong IPO, Targeting $50 Billion Valuation While Requiring Investors to Purchase Chips Alongside Shares
Taylor Wilson
Baidu's AI chip subsidiary Kunlun Chip is preparing a Hong Kong listing at a target valuation of $50 billion — roughly 40% above Baidu's own market cap; unusually, investors subscribing to the IPO must also commit to purchasing 3 to 7 times their subscription amount in Kunlun chips.
A $50 billion valuation — how can a subsidiary be worth more than its parent?
Kunlun Chip (昆仑芯) is targeting a $50 billion valuation. Baidu's current market cap sits at roughly $36 billion — the subsidiary would be priced about 40% higher.
This means → the market is being asked to believe Kunlun's AI chip business, on its own, is worth more than all of Baidu.
The IPO is still in the regulatory-approval stage with no fixed timeline; earlier reports indicated a fundraise of up to $2 billion, but the final size has not been disclosed.
"Buy shares, but first buy chips" — how unusual is this requirement?
Kunlun Chip is asking IPO investors to commit to purchasing chips worth 3 to 7 times their share subscription.
In plain terms = if you invest $100 million in stock, you must also spend $300–700 million on chips — and buying chips means you need to operate your own data centers.
The targeted investors are mainly local-government-backed funds with AI and semiconductor investment mandates.
This reflects the intensity of competition in China's AI chip market: chip companies are bundling equity with hardware commitments to lock in shipment volumes.
What is Kunlun Chip's technical edge?
Founded in 2011, Kunlun Chip is one of China's longest-established AI chip companies. Its products are compatible with Nvidia's CUDA software ecosystem — the programming toolkit that lets developers write code once and run it on GPUs.
This means → developers can migrate some workloads from Nvidia hardware with relatively low switching costs.
Baidu has already shifted part of its new Ernie large-model training onto Kunlun chips; the key release of Ernie 5.1 was trained on the platform.
Why is Kunlun Chip missing from the government's "safe and reliable" list?
China published its first "safe and reliable" certification list for domestic AI training and inference chips in May this year. Kunlun Chip was not included.
Companies on the list include Huawei, Moore Threads, Metax, and T-Head (Alibaba's chip arm).
In plain terms = without this certification, state-owned enterprises and government projects cannot procure Kunlun chips — yet SOEs are precisely the core customer segment Kunlun is targeting.
Kunlun Chip is reportedly seeking to shift fabrication to SMIC (中芯国际) in order to qualify in subsequent certification rounds.
What does the product line and customer base look like?
Kunlun Chip focuses on the inference chip market: the P800 series handles inference and fine-tuning, the M100 series began supporting large-scale inference this year, and the M300 series is planned to support both training and inference by 2027.
Three target customer segments: state-owned enterprises, major internet companies, and large AI model developers; Tencent is already a key external customer.
If the listing succeeds, Kunlun Chip will become the sixth Chinese AI chip design company to go public since December 2025.
What are the key uncertainties hanging over this IPO?
First, whether the valuation holds — can the market accept a subsidiary priced 40% above its parent?
Second, whether the missing government procurement certification can be resolved before listing — this directly determines whether SOE orders materialize.
Third, whether the "buy shares plus buy chips" model can attract enough capital — the high barrier narrows the investor pool significantly.
Content is for reference only, not financial advice.