T-Head reportedly plans 2027 Hong Kong IPO raising up to $3 billion

By Nathan Brooks

Alibaba's chip design unit, T-Head, could raise $2 billion to $3 billion in a Hong Kong IPO planned for 2027, according to International Financing Review, a Reuters service, citing sources. The lower end equals HK$15.6 billion.

IFR did not name banks or provide a valuation, and Alibaba has not announced a listing. The money, in other words, currently has a better headline than a prospectus.

What does the listing plan include?

Bloomberg reported in January that Alibaba was preparing to take T-Head public. The company would first restructure the unit into a business partly owned by employees; the restructuring was at an early stage, with timing and valuation unclear.

T-Head was founded in 2018 and designs the Zhenwu line of AI accelerators used in Alibaba's cloud. The proposed listing would expose that chip operation to public-market scrutiny, although the report gave no revenue split or production capacity.

The 2027 date remains a reported target, not a committed offering timetable. Without a valuation, named banks or financial forecasts, the offering's public ownership structure and economics remain unsettled.

What does the V900 contain?

Alibaba unveiled the Zhenwu V900 at its Apsara Conference in Hangzhou in September. The accelerator handles both training and inference, and its disclosed specifications are:

• Memory: 216GB.
• Chip-to-chip bandwidth: 1,200GB/s.
• Data formats: FP8 and FP4.

Alibaba claims the V900 delivers three times the performance of its predecessor, the M890. No independent benchmark validates those figures. Alibaba says the chip will enter mass production and go on sale in the first quarter of 2027.

Alibaba Group CEO Eddie Wu (Wu Yongming) called the chip "China's most powerful AI chip by computing performance." The national ranking is a company claim, not an independent comparative result.

The two 2027 markers are not equally firm. The IPO remains an IFR report based on sources, while the V900 production schedule comes from Alibaba. One sets a possible financing date; the other sets a product deadline.

How far does the cluster scale?

T-Head says more than 1,000 V900 chips can work as one system through its ICN Switch interconnect chips. A cluster can scale to as many as 500,000 accelerators on Alibaba Cloud's network, according to the company. Neither number establishes current installed capacity or shipment volume.

What has reached customers?

Servers using the M890 are already in large-scale commercial deployment, Alibaba says. The Zhenwu family has served more than 650 enterprise customers, also on the company's account, while Alibaba uses the accelerators in its own cloud. The report does not attach revenue, wafer capacity or unit shipments to either claim.

An earlier T-Head PPU accelerator was described by Chinese state broadcaster CCTV in September 2025 as comparable to Nvidia's H20. That is a separate competitive claim from the V900 performance figure, and neither has independent benchmark support in the report.

Which claims are commercial now?

For buyers, the useful distinction is between what the chip contains and what a deployed system can deliver. The 216GB of memory and 1,200GB/s bandwidth describe the V900 itself; the ICN Switch and 500,000-accelerator figure describe system scaling. Neither establishes how many V900s Alibaba has built or shipped, or what they cost.

The Zhenwu customer count establishes deployment breadth, but not T-Head's external revenue. The report does not separate internal Alibaba Cloud use from sales to outside enterprises, leaving the independence of demand untested.

What would investors be buying?

The proposed listing would give public investors exposure to a chip unit already deployed through Alibaba's cloud. Employee ownership would sit inside the new public structure, according to Bloomberg's early-stage account. Neither report supplied an offering structure or financial forecasts.

For 2027, the decisive questions are whether T-Head can complete the Hong Kong listing, reach V900 mass production in the first quarter and turn a 500,000-accelerator network claim into commercial deployments; pricing and independent benchmarks remain undisclosed.