DeepSeek revenue run-rate hits $1B, shifts training to Huawei chips

DeepSeek's business has scaled sharply: The Information reports an annualized revenue run-rate of $1 billion, and the company has engaged CITIC Securities to prepare for a possible listing on China's STAR Market. That would formalize DeepSeek's transition from a research-driven disruptor to a commercial AI provider with IPO ambitions.

The more strategically significant disclosure is on hardware. CEO Liang Wenfeng framed training AI models on domestic chips as one of the firm's biggest moves, with Huawei expected to supply training silicon as early as Q4. DeepSeek reportedly directs over 70% of its compute to training — an aggressive posture that, if executed on Huawei hardware, would demonstrate a viable non-NVIDIA training path and blunt the effect of US export controls.

This fits the week's larger China-AI narrative alongside Alibaba's Zhenwu V900 chip and trillion-parameter Qwen plans: Chinese labs are pursuing full-stack independence, from silicon to frontier models. Success would reshape the compute geopolitics that currently center on NVIDIA.

The caveats are real. Chinese AI stocks fell on reports of a regulatory probe into DeepSeek and Moonshot, injecting uncertainty into the IPO storyline. And a shift to Huawei training chips is easier announced than proven — training frontier models on less-mature silicon and software stacks has historically carried efficiency penalties. Watch for concrete evidence that DeepSeek can train a competitive model end-to-end on domestic hardware, and for how the probe affects listing timing.