About TrendForce News

TrendForce News operates independently from our research team, curating key semiconductor and tech updates to support timely, informed decisions.

[News] Alibaba Begins AI Model Training with T-Head Chips, Targets Full Domestic Adoption by 2027 amid New U.S. Curbs


2026-11-20 Semiconductors editor

Alibaba is accelerating its shift toward domestically developed AI chips as tighter U.S. export restrictions threaten Chinese companies’ access to advanced computing power. According to Nikkei, CEO Eddie Wu said the company is already training its next-generation AI model using proprietary chips developed by its semiconductor unit, T-Head. As the report notes, Alibaba aims to use domestic chips for all its AI training workloads, including pre-training and post-training, by 2027.

To support the transition, Alibaba plans to significantly expand T-Head’s production capacity over the next two years as its three-generation GPU roadmap advances faster than expected. According to The Paper, citing Wu’s remarks during the earnings call, the Zhenwu M890 has been fully deployed in 2026, with the V900 set to enter mass production in the first quarter of 2027, followed by the J900 in 2028.

Meanwhile, Alibaba is reassessing its broader AI infrastructure investment strategy amid tightening U.S. export restrictions. According to 21Jingji, Wu said the changing regulatory environment could prompt the company to lower its long-term data center capacity target of more than 20GW and reduce overall capital expenditure, even as it increases investment in domestic chips and related supply chains.

Amid the potential adjustment to its infrastructure spending plans, Alibaba expects its proprietary chip strategy to strengthen its competitive position in China’s increasingly constrained computing market. According to National Business Daily, Wu warned that 2027 and 2028 could see the most severe computing capacity shortages in China, potentially supporting pricing and gross margins. He added that proprietary chips and growing scale could give Alibaba procurement and supply chain advantages, improving profitability compared with relying on externally sourced chips. As T-Head’s computing supply chain scales up, Alibaba aims to achieve positive cash flow in the second half of 2028.

Alibaba is also pursuing a separate Hong Kong listing for T-Head, having confidentially filed an IPO application in September. According to The Paper, citing sources, the chipmaker could go public in 2027 and raise US$2 billion to US$3 billion, although the timing and offering size remain unconfirmed.

Chip Supply Constraints and Evolving AI Investment Plans

Even as Alibaba advances its chip roadmap, production capacity remains a potential bottleneck. As Nikkei reports, Wu said Alibaba would prioritize its proprietary chips for training its own foundation models and those of close partners, while relying on other Chinese chip suppliers to support inference workloads. Still, international expansion remains part of Alibaba’s longer-term strategy. Nikkei reports that Wu raised the possibility of eventually exporting the company’s proprietary chips, although no specific timeline was disclosed.

According to Alibaba‘s official earnings release, the company reported a 213% year-on-year increase in fiscal second-quarter net income to RMB 64.4 billion (US$9.7 billion), while revenue reached RMB 271.5 billion. Cloud revenue surged 53% year-on-year to RMB 61 billion, driven by strong demand for AI and cloud computing services. Meanwhile, quarterly capital expenditure rose 60% to RMB 50.48 billion as Alibaba continued investing in AI infrastructure.

Read more

(Photo credit: Alibaba)

Please note that this article cites information from Nikkei, The Paper, National Business Daily, 21Jingji, and Alibaba.



Get in touch with us