Skip to content

Business

Alibaba discloses Hong Kong IPO filing for AI chip unit T-Head

Edited and reviewed by AI Ping Editorial Team

3-line summary

  1. Alibaba disclosed plans on November 10 to list its AI chip subsidiary T-Head separately in Hong Kong.
  2. T-Head filed confidentially on September 11. Hong Kong listing approval and a mainland regulatory filing remain.
  3. Alibaba expects T-Head to remain a subsidiary. The offer size and first trading date are undisclosed.
PracticalOffice workersDevelopers
A fictional AI accelerator against a Hong Kong harbor skyline
AI-generated conceptual image: A fictional AI accelerator set against Hong Kong’s harbor. · AI Ping

Alibaba disclosed on November 10 that it is pursuing a separate Hong Kong listing for T-Head, its AI and data-center chip design subsidiary. The Hong Kong Stock Exchange cleared Alibaba to proceed with the proposed spin-off, and T-Head submitted a confidential Form A1 listing application on September 11. The November notice publicly confirmed that filing.

If the listing goes ahead, investors would be able to trade shares in the chip business separately from shares in Alibaba Group. Alibaba owned all of T-Head at the time of the announcement and expects the company to remain a subsidiary after the listing. The size and structure of the offering and Alibaba’s post-listing stake are still open. The notice gives no offer price or first trading date. The exchange records the announcement at 8:36 p.m. on November 10 in Hong Kong.

Hong Kong listing approval and a mainland filing remain

T-Head applied to list H shares, the shares of a mainland Chinese company traded in Hong Kong, on the exchange’s Main Board. The proposal calls for a public offering in Hong Kong and a placement with institutional and professional investors. The exchange lists Form A1 as the application for companies seeking a new Main Board listing of equity or debt securities.

The exchange’s Practice Note 15 governs separate listings of businesses that are already part of a listed group. It requires the parent company to submit its spin-off proposal to the exchange. In this case, the Listing Committee still has to approve the listing and trading of T-Head’s H shares. The plan also requires a filing with China’s securities regulator and final decisions by the two companies.

The exchange granted a waiver of the rule that would ordinarily give Alibaba’s existing shareholders an assured opportunity to receive or apply for shares in the spin-off. Alibaba cited the cost of complying with securities rules across jurisdictions and restrictions on transferring existing shares. Separately, the exchange proposed scrapping the assured-entitlement rule in a September 21 consultation open until November 30.

An open paper folder beside a closed one
AI-generated conceptual image: Two fictional folders represent the application and the review that follows. · AI Ping

The chips behind Alibaba Cloud computing services

T-Head’s chip portfolio covers AI accelerators, server processors, storage and networking. Alibaba Cloud’s September roadmap identified the Zhenwu accelerators for training and inference, Yitian server CPUs, Panmai networking chips and Zhenyue storage controllers. The latter components move data between servers and manage reads and writes to storage. In its August results for the June quarter, Alibaba said more than 650 external customers in over 20 industries used chips in the Zhenwu family, including the M890, through Alibaba Cloud services.

For cloud customers, one way to access the hardware is through a server instance. In August, Alibaba CEO Eddie Wu said the company had launched a Supernode instance powered by the M890 at commercial scale. He said the instance was already running Kimi K3 and Qwen3.8-Max for external customers. Those hosted model services use the hardware for inference, the calculations that produce a response to a user’s request.

Alibaba Cloud announced the next accelerator, the Zhenwu V900, at its Apsara conference on September 22. It described the chip as a processor for model training and inference, with 216 GB of memory and 1,200 GB per second of bandwidth between chips. An upgraded server design pairs it with T-Head networking and storage controller chips. The company’s announced target for volume production and commercial release was the first quarter of 2027.

A fictional rack server with heatsinks and power cables
AI-generated conceptual image: A conceptual view inside a data-center server that houses accelerators. · AI Ping

A separate route to capital

Alibaba’s board said the listing could give investors a way to value T-Head apart from the rest of the group and let the chip designer raise equity or borrow in debt markets on its own. Alibaba expects to retain its dual primary listing in Hong Kong and New York if the proposal goes ahead.

Sources