Baidu is preparing to spin its in-house AI chip unit Kunlunxin into a Hong Kong IPO at a target valuation of $50 billion, according to a report from The Information picked up by Reuters. The listing would carve out one of China's most strategically important silicon programs as a standalone public company at a valuation that rivals the largest pure-play AI chip designers globally.
The $50 billion target is the number that matters. It places Kunlunxin in the same conversation as the handful of non-Nvidia AI accelerator companies that public markets treat as serious bets, and it implies Baidu's parent company has been carrying a chip business on its balance sheet that the market values at a substantial multiple of what Baidu itself trades at by segment.
Kunlunxin designs AI accelerators that Baidu uses internally to train and serve its own models, including its Ernie family of large language models, and that the company has increasingly positioned as a domestic alternative for Chinese customers locked out of the top tier of Nvidia hardware by US export controls. A Hong Kong listing is the natural venue: it gives Chinese institutional capital direct access to the equity, sidesteps the US-listing friction that has dogged Chinese tech for half a decade, and lets Beijing's regulators sign off without the political theater a New York filing would invite.
Key facts
- 01Kunlunxin, Baidu's AI chip unit, is targeting a $50 billion valuation in a Hong Kong IPO, per The Information.
- 02The listing would carve out Baidu's in-house silicon arm as a standalone public company.
- 03A $50 billion mark would rank Kunlunxin alongside the largest pure-play AI chip designers globally.
The timing tracks with a broader push across Chinese tech to monetize semiconductor assets while domestic AI demand is running hot. Cloud providers, internet platforms, and state-linked buyers in China are scrambling for any accelerator that can substitute for restricted Nvidia parts, and Kunlunxin is one of the few in-house programs at a hyperscaler-scale parent that has shipped silicon at volume.
For Baidu, the spin would do two things at once. It would unlock a valuation the market has not been giving Baidu credit for inside the consolidated structure, where the chip business sits buried under search, autonomous driving, and a cloud unit that has struggled against Alibaba and Tencent. It would also raise primary capital for Kunlunxin to fund the next generations of its accelerators, where the spending requirements scale fast with each process-node jump.
The $50 billion figure should be read as a target, not a price. IPO targets in Hong Kong frequently compress between filing and pricing, particularly for hardware companies whose comparable trading multiples can shift sharply on a single Nvidia earnings call. The final mark will depend on what the order book looks like when the deal goes out, what the macro setup is in Hong Kong, and what Beijing signals about the strategic importance of domestic AI silicon between now and the listing window.
There are also open questions the report does not resolve: how much of Kunlunxin Baidu intends to float versus retain, who the cornerstone investors are, what the customer mix looks like outside Baidu itself, and how dependent the unit remains on its parent for revenue. A chip business that sells 80% of its output back to the company spinning it out is a very different investment from one with a diversified external customer base, and the prospectus when it lands will be the document that matters.
The competitive frame is also worth naming. Kunlunxin sits in a domestic Chinese accelerator market that also includes Huawei's Ascend line, Cambricon, Biren, Moore Threads, and several smaller programs, all chasing the same set of Chinese cloud and enterprise customers that can no longer buy top-tier Nvidia parts. A $50 billion public-market valuation, if it holds, would set a clear pecking order in that field and pull capital toward whoever the listing implicitly anoints.
A spinout at this scale signals that the in-house chip programs at Chinese hyperscalers have stopped being cost centers and started being assets the market will pay for separately. That is the same pattern playing out at OpenAI, Google, and Apple on the US side, where custom silicon programs are increasingly treated as load-bearing pieces of the AI stack rather than internal infrastructure. If Kunlunxin prices anywhere near $50 billion, expect Alibaba's T-Head and other Chinese in-house silicon units to face pressure to follow — and expect Nvidia's competitive map inside China to look meaningfully different a year from now.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
