Baidu's search engine runs on its own chips. So does its Ernie (文心一言) chatbot. On the first day of 2026, the company moved those chips toward their own balance sheet: a confidential filing that would turn an internal division into a separately listed company, with Baidu still holding the controlling stake.
The move is a spin-off (分拆上市), not a sale. But it puts a number on a piece of Baidu investors have never priced directly — and raises a question about why a chip unit good enough for Baidu's own AI needs outside money.
Key takeaways
- Baidu (9888.HK / BIDU) said on 2 January 2026 that Kunlunxin had confidentially filed a listing application with the Hong Kong Stock Exchange on 1 January, paving the way for a spin-off and separate listing, according to Reuters.
- After the proposed spin-off, Kunlunxin is expected to remain a Baidu subsidiary, with Baidu retaining a controlling stake (控股权). Offering size and structure have not been finalized.
- Reuters had previously reported that Kunlunxin completed a fundraising valuing it at RMB 21 billion (≈US$3.0B / HK$23.4B).
- Kunlunxin began in 2012 as an internal business unit developing AI chips (人工智能芯片) for Baidu. It now operates independently but still sells mainly to its parent.
- The company has expanded external sales over the past two years, Reuters reported — the detail that makes a standalone listing plausible.
- Hong Kong raised about US$36.5 billion from 114 new listings in 2025, its strongest year since 2021 and more than triple 2024's US$11.3 billion, according to LSEG data cited by Reuters.
What the filing actually says
The announcement is deliberately thin. Baidu confirmed a confidential filing (秘密递表), which in Hong Kong means the application is not public and the financials stay hidden until a later stage. No prospectus, no revenue, no offering size. What is public is the intent: a separate listing, with Kunlunxin remaining inside the group afterward.
A spin-off in which the parent keeps control is about visibility and capital access, not cutting the business loose. Baidu is not walking away; it is giving the unit its own shareholders and cost of capital while keeping the strategic link.
What Kunlunxin makes
Kunlunxin develops AI chips for Baidu's own computing needs — search, cloud and the Ernie family of large models (大模型). It started life in 2012 as an internal team, which explains why its largest customer has historically been its own parent.
The interesting change is external sales. Reuters reported that Kunlunxin has broadened beyond Baidu over the past two years. Secondary reporting cited by TrendForce puts external customers at roughly 40% of the business, spanning internet companies, smartphone makers, telecom operators and state-owned enterprises. That figure comes from media reports rather than the confidential filing, so treat it as indicative, not audited. Even without it, the direction is clear: a unit that sells only to its parent is a cost centre, while one with outside customers is a business.
Why separate it now
Two forces sit behind the timing.
The first is export controls. Washington has restricted China's access to advanced semiconductors, and Beijing has promoted domestic alternatives. Several Chinese AI chip designers have already listed. A separately funded Kunlunxin can raise money for chip development without competing with Baidu's other capital priorities.
The second is the Hong Kong window. With 2025 fundraising at its strongest since 2021, Hong Kong has become the default venue for Chinese technology companies seeking offshore investors. A listing there gives Kunlunxin a dollar-denominated currency for acquisitions, a standalone equity incentive plan for engineers, and a market price that, in principle, reflects chip economics rather than Baidu's advertising cycle.
For Baidu, there is a defensive logic: chip development is expensive and slow. Ring-fencing it shows investors what the infrastructure layer is worth while shielding group earnings from that spending.
The valuation, handled carefully
The US$3 billion number is the one most likely to be repeated without context. It comes from Reuters' earlier report of a completed fundraising, not from a public market and not from the confidential filing. A private-round valuation is a price agreed between a company and a small group of investors; it is not a market consensus and it carries no liquidity.
There is also no public financial base to check it against. Because the filing is confidential, there are no disclosed revenue, margins or shipment figures in the sources reviewed here. Any price-to-sales or price-to-earnings comparison with a listed peer would be guesswork. The only defensible statement is the narrow one: Reuters reported a private valuation of about RMB 21 billion (≈US$3.0B / HK$23.4B) after a funding round.
What changes after a spin-off
The practical differences are concrete. A standalone balance sheet can fund tape-outs and capacity without asking the parent each time. A separate share price is a currency to pay engineers and acquire small design teams. And outside customers often prefer an independent supplier to one owned by a potential competitor.
For Baidu, the trade-off is dilution of a strategic asset in exchange for capital and clarity. It keeps control, so the relationship does not break; but it does invite outside shareholders who will judge the chip unit on its own returns rather than on its service to the group.
That is the real stake for ordinary readers. This is less a story about one valuation than about whether China's internet giants can turn internal infrastructure into standalone businesses that survive a public market's scrutiny.
Honest limitations
The filing is confidential, so no prospectus, financial statements or offering terms are public. All facts come from Baidu's statement as reported by Reuters and earlier Reuters reporting on the valuation; I have not seen the filing. The roughly 40% external-sales figure is secondary media reporting cited by TrendForce, not a company disclosure. The 2012 founding date and the "remains a subsidiary" language are from Reuters. No listing has occurred; offering size, price and timing are undecided, and a private valuation is not a public price. This is not investment advice.
What readers can do now
- Watch for the application to become public: the first real financials will show whether external revenue is growing faster than Baidu's own orders.
- If you follow China's chip sector, treat the RMB 21 billion (≈US$3.0B / HK$23.4B) figure as a reported private valuation, not a benchmark for listed peers.
- For anyone comparing AI chip listings, note the structural difference here: this is a controlled spin-off of an existing unit, not a first-time standalone company.
