Chinese cloud company Alibaba's chip unit T-Head has announced a new AI chip that can handle both training and inference workloads

Dubbed the Zhenwu M890, the chip apparently offers three times the performance of the current Zhenwu 810E processor.

Alibaba
– Getty Images

Revealed at Alibaba's Cloud Summit conference on Wednesday, May 20, the new chip will have 144 GB of GPU memory and an interchip bandwidth of 800 GBps.

T-Head's vice president, Gao Hui, further noted that the company will release the Zhenwu V900 in the third quarter of 2027, followed by Zhenwu J900 in the third quarter of 2028.

During Alibaba Group's earnings call last week, much weight was put on its chip unit for its success in the AI race.

CEO Eddie Wu said during the call that its T-Head unit means the company can "actually also sell AI servers leveraging those chips to other computing centers, or we can co-build computing centers with others," and that the company has "achieved scaled mass production" and that 60 percent of compute capacity is now serving external customers.

"As the only AI cloud provider in China capable of delivering self-developed AI chips at scale, we've secured autonomy over our compute supply chain while providing customers with highly competitive AI inference and training services," Wu said. "In an environment of compute scarcity, this structural advantage is favorable to our revenue growth and gross margin improvement."

Overall, the cloud unit for Alibaba brought in more than $6bn, a 38 percent growth Year-on-Year (YoY).

Since then, competitor Baidu has also released its quarterly earnings.

Baidu, known best for its Chinese-language search engine, has also seen significant growth on the back of AI.

Yanhong Li, Baidu's CEO, chairman, and cofounder, said its GPU cloud offering has seen a revenue growth of 184 percent YoY, reaching RMB 8.8 billion ($1.29bn)

This was, according to CFO Haijian He, due to the higher margin of GPU clouds than CPU clouds, driven by "stronger demand, tighter supply chain, higher technical barriers and pricing power."

Baidu also has its own chip offering - the Kunlunxin AI chips.

Dou Shen, the president of Baidu’s AI Cloud Group, conceded that the Chinese AI chips “are still catching up with the most advanced global products in certain frontier training scenarios,” but added: "We have seen remarkably strong enterprise demand for AI infrastructure, both training as well as inference.

"Inference is showing particularly strong momentum, which is a pretty healthy signal. It tells us that customers have moved beyond training models and are now running AI across more parts of their business at an accelerating pace.”