Dreame Xinji (追觅芯际) unveils three AI chips, claims per-chip compute up to 2,000 TOPS

Lead

Dreame (追觅科技)’s chip arm, introduced as Xinji ChuanYue (芯际穿越), has unveiled a slate of frontier processors covering smartphones, autonomous driving and robotics, it has been reported. The announcement came at the AWE 2026 Chip Industry Summit where the company framed the push as part of a broader strategy to move from consumer devices into mobility and “robotic” compute. The splashy headline? A single driving SoC reportedly delivers up to 2,000 TOPS of AI throughput.

Product details

The company introduced a flagship mobile AI SoC named Chi Xiao 01 (赤霄 01) that reportedly embeds a self‑developed NPU architecture with about 200 TOPS of equivalent AI compute, aimed at low‑latency multi‑turn dialogue and on‑device reasoning alongside integrated graphics features such as smart super‑resolution, dynamic resolution reconstruction and frame‑rate upscaling. For vehicles, Dreame showcased a highly integrated cabin‑and‑drive chip built on a claimed 2 nm node with the 2,000 TOPS figure and design goals to meet L4 autonomous driving compute needs; the chip is described as tightly coupled with Dreame’s self‑developed full‑stack driving software, fusing vision, LiDAR and a world model. The firm also listed plans for a personal “super AI” computer, a general‑purpose robot SoC and a space compute center; the driving chip is reportedly at tape‑out or sample testing stage.

Geopolitical and market context

China has accelerated domestic chip and AI hardware development amid Western export controls and strained trade ties, and Dreame’s pitch fits that national priority: build stack‑level capability from silicon to sensors to models. That said, some technical claims invite scrutiny — reported 2 nm processes and multi‑thousand TOPS numbers raise questions about foundry access, packaging and power budgets, and where such chips would be manufactured given ongoing restrictions on advanced node exports. Commercial timing and production scale remain unclear. Can a consumer‑appliance veteran scale into cars, robots and space compute at this performance level? The market — and geopolitical realities — will help decide.