The Information reported June 28 that Baidu's chip arm Kunlunxin is targeting a $50 billion valuation for its Hong Kong listing, and the unusual condition attached to the deal tells you almost as much as the valuation itself.
Kunlunxin isn't just pitching itself as an investment. It's packaging capital and procurement into the same handshake. According to The Information, prospective IPO investors are being asked to commit to buying its chips as part of participating in the deal. You don't see that often in a conventional IPO, and you shouldn't treat it as a small procedural wrinkle. It suggests Kunlunxin still has to prove that customers outside Baidu will buy its semiconductors at the scale a $50 billion valuation requires.
The valuation has moved fast enough to give anyone pause. South China Morning Post recently put Kunlunxin's target at about $14.7 billion, while TrendForce cited HK$100 billion, around $12.8 billion, in May. The Wall Street Journal reported on June 3 that Morningstar analysts estimated the chip unit could be worth HK$400 billion to HK$500 billion, or roughly $51 billion to $64 billion. The new number is not coming from nowhere, but the speed of the reset matters. Either investor appetite has run well ahead of the operating record, or Kunlunxin is using the current AI chip market to price itself like a strategic national asset. Frankly, it is asking for both readings at once.
Biren Technology helps explain the mood. The Chinese AI chip designer raised about $717 million when it listed in Hong Kong in January 2026, then surged 119% intraday, according to market reports at the time. That kind of debut changes the room for every Chinese semiconductor company behind it. But Biren came to market as a focused chip startup. Kunlunxin is different. It is a Baidu-controlled unit trying to convince outside buyers that it can become more than the in-house silicon arm of China's dominant search company.
Kunlunxin's strongest proof point is the P800, a third-generation XPU that Baidu unveiled at its developer conference in April 2025. Baidu said then that it was running a 30,000-chip cluster on the P800, and Chinese reports later described the chip as capable of standalone deployment for DeepSeek V3 and R1 at their full 671 billion parameter scale. That's a real technical marker, not a brochure phrase. A domestic chip that can run China's most watched open models without leaning on Nvidia hardware gives cloud providers, state enterprises and government-linked buyers a reason to listen.
The roadmap is doing work too. Baidu has said the M100, aimed at inference, is scheduled for 2026, with the M300 training chip following in 2027. Those dates matter because inference is where spending keeps compounding once models are deployed, while training chips remain the harder and more politically sensitive prize. If Kunlunxin can hit both steps, it becomes a supplier in the part of the market Beijing most wants to localize. If it misses, the valuation starts to look like a bet on policy rather than performance.
Baidu still owns 57.67% of Kunlunxin, which grew out of an internal chip effort that began more than a decade ago. The company confidentially filed with the Hong Kong Stock Exchange in January 2026, with CICC, Citic Securities and Huatai Securities as lead underwriters, and it has also started STAR Market listing guidance in Shanghai. The dual track is not decorative. Hong Kong gives Kunlunxin access to international capital, while Shanghai gives it the domestic institutional stamp that can matter when state-linked buyers choose suppliers.
The export-control backdrop is the reason this story has teeth. US restrictions have limited Chinese access to Nvidia's most advanced AI chips for years, and the market has only become messier as Washington, Beijing and Nvidia work around successive rules on parts such as H20 and H200. Recent reporting from the Financial Times also showed prices for banned Nvidia systems climbing sharply on China's black market. You don't need a grand theory to see what that does. It pushes Chinese buyers toward domestic alternatives even when those alternatives are still catching up on software, availability and ecosystem support.
The investor-as-customer structure is the sharpest tell. Baidu has been Kunlunxin's anchor customer from the beginning, but the public market will care about third-party sales. Asking investors to buy chips as well as shares makes commercial sense if Kunlunxin can pull it off, because it turns capital into demand before the first public earnings call. It also creates a tension that shouldn't be waved away. A shareholder who is also a committed customer may not look at procurement terms the same way a normal outside investor would.
At $50 billion, Kunlunxin would sit near US chip names such as Marvell Technology and far above the usual public-market range for many Chinese semiconductor firms. That is a bold place to price a company still proving it can sell beyond its parent. The IPO market will probably answer before the end of 2026, but the real test is more basic than the listing pop. Can Kunlunxin turn China's forced chip independence into repeat customers who would still buy its hardware without an IPO allocation attached?
Also read: Google DeepMind is losing the researchers who built its best work and the damage is already showing, Suno's Spark artist grants come with a permanent silence clause that says more about the lawsuit than the music, and AI coding agents are turning code review into the next startup risk
