Alibaba’s T-Head is moving toward a public listing of its AI chip unit, reflecting rising investor interest in AI silicon amid limits on Nvidia in China.

The IPO timing is not set and the restructuring is in early planning, with no valuation yet established.

T-Head has secured contracts, including deploying Pingtouge AI accelerators in a northwestern China data center for China’s second-largest wireless carrier.

Pingtouge accelerators will operate in the carrier’s data center alongside MetaX Integrated Circuits and Biren Technology hardware.

The PPU, a rumored GPU-style accelerator, is said to deliver competitive performance with NVIDIA’s H20/A100 line, featuring 96GB HBM2e, up to 700GB/s inter-chip bandwidth, PCIe 5.0 x16, and about 400W, with strong 2025 market demand.

Ownership aims to retain top engineers from XuanTie RISC-V and Hanguang AI units, emphasizing talent retention in the restructuring.

The move mirrors a broader industry shift toward hyperscalers building in-house silicon, with potential implications for software compatibility and global supply chains.

T-Head designs across the stack—from data center chips to AI accelerators and IoT—competing in a market where Nvidia faces export restrictions in China.

T-Head has broadened into storage and edge/IoT chips, aiming to cover a full data-center chip stack with products like Zhenyue 510 SSD controllers and Yuzhen edge chips.

Nvidia’s H200 and other accelerators remain global benchmarks, with CEO Jensen Huang planning a China-focused trip to ease access for its chips.

Long-term questions include sustaining rapid tech progress under export controls and whether T-Head can become a leading silicon provider for a sovereign AI cloud in Asia and the Middle East.

Caixin reports T-Head’s PPU was the top-selling domestic GPU in 2025, underscoring its strong market position.