Alibaba has confirmed what the market had been pricing for weeks. On the evening of 10 November the group said it plans to spin off T-Head, its chip subsidiary, for a separate listing on the main board of the Hong Kong Stock Exchange.
The filing itself is older than the announcement. T-Head submitted a confidential A1 application form to the Hong Kong exchange on 11 September, after the exchange confirmed the group could proceed with a spin-off under Practice Note 15 of the listing rules.
What the filing contains
Under the proposal, T-Head would issue H shares through a Hong Kong public offer and an international placing. Alibaba has not fixed the size of the offer or the timing, though Chinese media report the unit is aiming to list next year and could raise between USD 2 billion and USD 3 billion.
T-Head would remain a subsidiary of Alibaba after the spin-off, with the final stake still undetermined. The group’s stated rationale is that separate listings let investors value the parent and the chip business independently, and let T-Head raise equity and debt on its own terms.
The scale of the business is already substantial. In March, Alibaba chief executive Eddie Wu said T-Head’s annualised chip revenue had reached the RMB 10 billion level. By April, the TrueWave series of AI chips had shipped more than 560,000 units. T-Head’s PPU was one of the highest-volume domestic GPU chips of 2025, and by early August the TrueWave customer count had passed 650.
In August the group reorganised its financial reporting, folding the cloud intelligence group and T-Head into a single AI cloud and compute services segment. That segment posted RMB 48.437 billion of revenue in the quarter, up 45 per cent year on year. Alibaba does not break out T-Head’s contribution.
Product cadence ran ahead of the paperwork
On 22 September, at the Yunqi conference, Wu introduced the TrueWave V900. Alibaba claims the chip delivers three times the performance of the previous M890 and was the most powerful China-designed AI chip at launch. Volume production is scheduled for the first quarter of next year.
What the announcement sequence reveals is that T-Head had already filed in Hong Kong before the V900 took the stage. The commercial story was being told to customers and developers while the capital story was already in a regulator’s inbox.
T-Head was founded in 2018 and initially served Alibaba’s internal workloads. It now spans TrueWave AI accelerators, the Yitian server CPU, ICN interconnect silicon, the Panmai smart network card and the Zhenyue storage controller, which makes it one of very few Chinese firms designing compute, storage and networking core silicon at the same time.
The comparison that matters is Kunlunxin
Baidu’s chip unit filed confidentially in January and has now been waiting for more than ten months. In May it also completed the tutoring filing for a STAR Market listing, running Hong Kong and Shanghai in parallel. In early November the market briefly speculated that its Hong Kong IPO would slip, and Baidu denied it, saying the company remains in the confidential submission stage.
The two are close enough in shape to price against each other. Both design AI accelerators, both are owned by internet platforms with captive demand, and both are seeking offshore capital. Once both prospectuses are public, investors will be able to compare products, customers and revenue directly, which is precisely what a confidential filing delays.
The broader backdrop has not been friendly. Global AI and semiconductor equities went through a sharp correction, with the Philadelphia Semiconductor Index falling roughly 12 per cent on 5 November. By 9 November some domestic compute names had bounced, with Enflame up more than 17 per cent intraday and Cambricon up more than 6 per cent, trailed by MetaX and Moore Threads.
Overseas, the same logic is running in the opposite direction for the incumbents. Alphabet’s third-quarter results credited TPU systems as a driver of cloud revenue growth, and on 9 November Google added two more systems to its eighth-generation TPU line, bringing that generation to four systems aimed at training and inference. Google’s AI infrastructure chief said the TPU line now has close to 100 external customers. Last October, Anthropic said it would use up to one million TPUs under an expanded Google Cloud agreement worth tens of billions of dollars.
Tencent and ByteDance are also accelerating in-house silicon, though for now the emphasis is on development and internal deployment rather than merchant sales.
Why the Hong Kong route
Hong Kong gives a Chinese chip designer three things Shanghai does not currently offer as easily. It provides an offshore currency for international investors, a listing venue where specialist semiconductor funds can take a position without onshore quota friction, and a pricing reference that is legible to suppliers and customers outside China.
It also transfers the valuation risk. Once T-Head trades separately, Alibaba’s own multiple stops being a blend of e-commerce, cloud and silicon, and the chip unit has to defend its own number every quarter in public.
Editor’s note: This is an adapted translation of the original Sohu Tech report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://www.sohu.com/a/1099958696_116132.