TL;DR
Huawei now owns 50 to 62% of China's AI chip market as Nvidia collapses to near zero share, while DeepSeek simultaneously launches a model, files for a $75 billion IPO, and orders 160,000 Huawei chips for a gigawatt datacenter, cementing a fully domestic AI stack that bypasses American technology entirely.
What happened
- Huawei projects 50 to 62% share of China's AI accelerator market in 2026, targeting 750,000 Ascend units shipped, while Nvidia's share has fallen to near zero from a prior 95% (Jensen Huang's own acknowledgment).
- DeepSeek ordered 160,000 Huawei Ascend 950DT chips (~$2.56 billion) for a 1-gigawatt inference datacenter in Ulanqab, Inner Mongolia, a raw chip count exceeding xAI's Colossus and Google's TPU Virgo clusters.
- DeepSeek launched V4.1-Flash on September 10, its first natively multimodal model, with aggressive sub-cent-per-million-token pricing and a 60% cache-hit discount, retiring V4-Pro on September 14.
- DeepSeek tapped CITIC Securities as lead underwriter for a STAR Market IPO targeting a ~$75 billion valuation, following a June funding round above $50 billion that included Tencent ($1.4 billion) and CATL ($700 million).
- A joint NSA, CISA, and FBI advisory (AA26-251A, September 8) named DeepSeek, Alibaba, Moonshot AI, MiniMax, StepFun, and Z.AI for industrial-scale distillation of Claude, GPT, Gemini, and Grok since late 2024.
Why it matters
- Vertical integration is complete: DeepSeek's V4 architecture was designed for Huawei Ascend silicon, its IPO funds domestic compute buildout, and its pricing undercuts every rival, creating a closed-loop stack with zero American dependency.
- Nvidia's China business is effectively gone: a collapse from 95% to near zero share is not a market setback, it is an exit, and it removes a major revenue stream while validating China's years-long chip localization push.
- The distillation row is now a sanctions trigger: US Treasury put sanctions and Entity List actions explicitly on the table for firms conducting at-scale capability extraction, turning model provenance into a hard compliance question for any enterprise deploying Chinese open-source models.
- Anthropic quantified the damage: 190 million illicit exchanges detected across five campaigns, with ~151 million linked to Alibaba alone between May and July 2026, and DeepSeek accused of routing 12 million customer exchanges to Claude in 14 days without user disclosure.
- Market reaction was immediate: MiniMax and Z.AI each dropped over 8% in Hong Kong trading after V4.1-Flash launched, signaling that DeepSeek's pricing power is now a systemic competitive threat to every Chinese AI peer.
What to watch next
- US sanctions and Entity List decisions targeting named distillation actors, which would escalate the tech war beyond chip export controls into direct penalties on Chinese AI labs.
- DeepSeek's IPO timeline and STAR Market approval, which would give the company public capital to accelerate compute buildout and widen its pricing moat against both domestic and Western rivals.
- Huawei Ascend 950DT delivery and performance benchmarks for the Ulanqab datacenter, since chip count alone does not confirm parity with Nvidia-based clusters and actual throughput will determine whether China's domestic silicon can sustain frontier inference at scale.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.