* Alibaba unveils Zhenwu M890 AI chip

* Chip is three times faster than previous generation

* Multi-year roadmap includes V900, J900 chips

BEIJING/SINGAPORE, May 20 (Reuters) - Alibaba Group ( BABA )

on Wednesday unveiled a new AI chip, the Zhenwu M890,

as the Chinese technology giant intensifies efforts to build

domestic alternatives to Nvidia ( NVDA ) processors amid

tightening U.S. export curbs.

The chip, developed by Alibaba's semiconductor design

subsidiary T-Head, delivers three times the performance of its

predecessor, Zhenwu 810E. It is purpose-built for the emerging

wave of AI "agents" - software systems that can carry out

complex, multi-step tasks with limited human oversight.

Alibaba said the new processor is well-suited to handle the

heavy memory and communication demands of agent workloads, where

models must retain long stretches of context and coordinate with

one another in real time.

The company also outlined a multi-year chip roadmap, saying

it would follow the M890 with a successor called the V900 in the

third quarter of 2027, and a further chip, the J900, in the

third quarter of 2028. The V900 is expected to deliver another

roughly threefold performance gain over the M890, Alibaba said,

signalling a sustained cadence of in-house silicon upgrades.

The plan underscores China's growing efforts to produce

locally developed AI chips as Washington bans the sale of the

most powerful U.S. processors to Chinese customers, and follows

a similar announcement by Huawei last year.

Hangzhou-based Alibaba last year pledged to spend more than

380 billion yuan ($53 billion) on cloud and AI infrastructure

over three years, its largest-ever commitment to the sector.

The investment reflects a broader bet across China's

technology industry that demand for AI computing power will

continue to surge as enterprises adopt agent-based applications.

Alibaba unveiled the chip at its annual Alibaba Cloud

Summit, alongside a new server system, the Panjiu AL128, which

packages 128 of the accelerators into a single rack.

The system is available immediately to Chinese enterprise

customers through Alibaba Cloud's domestic model platform, known

as Bailian.

T-Head said it has shipped more than 560,000 Zhenwu units to

date, with over 400 external customers across 20 industries,

including automakers and financial services firms, having

deployed the chips.

Alibaba also announced Qwen 3.7-Max, the latest version of

its flagship large language model, which it said is engineered

for advanced coding and long-running agent tasks. The company

said the model can operate continuously for up to 35 hours

without performance degradation.