Pingtouge shifts its AI chips from internal use to external sales, expanding into the broader market

In September 2025, a brief frame in China’s national news broadcast showed a performance comparison between a domestic AI chip and Nvidia hardware. Industry insiders quickly identified the chip as a long-secret PPU (Parallel Processing Unit) developed by Alibaba’s semiconductor arm, T-Head. That single second effectively ended years of quiet development.

On January 29, 2026, T-Head officially unveiled its high-end AI chip, “Zhenwu 810E,” positioned against Nvidia’s H20, announcing shipments of several hundred thousand units. A new AI chip is reportedly in the pipeline.

T-Head was born out of Alibaba’s need for “computing power independence.” Alibaba launched the DAMO Academy in 2017 and formally established T-Head in 2018 after integrating chip assets. Its first AI chip, Hanguang 800, debuted in 2019. Development of its PPU architecture began quietly in 2020.

The Zhenwu 810E features a proprietary parallel architecture, 96GB of HBM2e memory, and 700GB/s interconnect bandwidth, targeting AI training, inference, and autonomous driving.

For years, T-Head mainly supplied chips internally to Alibaba, seen as a cost center. Now, with shipments in the hundreds of thousands and over 400 external clients—including State Grid and XPeng—its role is shifting toward a commercial growth engine.

Reports in early 2026 suggest Alibaba may pursue an independent listing for T-Head. Analysts expect Chinese AI chipmakers to gain share in 2026, with T-Head projected at around 5%. However, advanced process constraints remain a challenge, and its long-term capital story will depend on ecosystem strength and execution.