Huawei's chairman says its computing goal is to become Nvidia
Two infrastructure stories today: the ambition inside Huawei's compute business finally goes on the record, and a security vendor puts a hard number on whether detection rules actually fire.
Huawei's supervisory board chairman Guo Ping told new employees that the company's target in ICT and computing is to become the Nvidia of the field — and that its real advantage is not owning a frontier model, but running everyone else's. The remarks come from roughly 10,000 characters of Q&A minutes that Huawei published on its internal Xinsheng Community forum on September 14, and which Chinese tech media picked up this week. "For the ICT business and computing, Huawei's goal is to become Nvidia," he said, adding that the company's core strength "is not having its own large model, but supporting customers in building excellent ones — ensuring any global model can run efficiently on our Ascend, Kunpeng, SuperNodes and clusters."
The framing matters because it inverts the story the West usually tells about Huawei's AI business. Guo was blunt that Huawei started early in large models and never took a dominant position, and that Anthropic, Moonshot's Kimi and DeepSeek are the ones who broke through — an admission that heavy spending by the biggest labs has not, by itself, produced results. He cites DeepSeek founder Liang Wenfeng's argument that rapid iteration in the AI era can erode moats like Nvidia's CUDA ecosystem, which is precisely the opening Huawei is building for. Its Ascend 950 series closes part of the gap through architectural work rather than process leadership — Guo concedes Huawei faces harsher fabrication constraints than rivals — and the company's answer is system-level: clusters and SuperNodes instead of single-chip specs. DeepSeek is already testing that thesis in production, with a planned 160,000-chip Huawei cluster in Inner Mongolia.
On strategy, Guo said Huawei's focus is connectivity and computing with no expansion beyond it, and that in-house models are justified only where the business demands them: terminals and smart driving, where open weights come without the continued training support customers need, and Huawei Cloud, which he says cannot stay competitive without one. He also repeated that Huawei does not chase unit volumes against Apple, and that sanctioned companies normally end up either "in jungle" or "in prison" — Huawei is trying to carve out a third path.
UltraViolet Cyber launched Equinox, a service that inventories the detection rules and log sources a security team already runs, maps them against MITRE ATT&CK, and reports which ones actually work — a job that has traditionally taken detection engineers weeks and now finishes in under 30 minutes. The interesting part is the second framework. UltraViolet says it is the first to produce a full coverage map and scorecard against MITRE ATLAS, the companion knowledge base for adversary behavior aimed at AI and machine learning systems — the attack surface that every company now wiring agents into production has added without adding detection coverage for it.
Nothing deploys automatically: engineers on the company's Threat Intelligence and Detection Engineering team review each recommendation and backtest it against the customer's own telemetry before tuning, and a separate log-source gap report puts a number on how much extra coverage each new data feed would buy. "Every SOC team has heard the question 'are we actually covered?'" said Dan Gittis, director of the TIDE team. He calls confirming that a detection can fire on live data "the difference between a coverage dashboard and coverage you can defend in an audit." Equinox ships with the company's managed SOC service at no extra cost and is sold standalone to teams running their own; Splunk and Microsoft Defender support is still in development.
What to watch: whether "any model runs on Ascend" survives contact with the next generation of frontier models trained on CUDA-first tooling.
If Huawei gets a fair shot at being the non-Nvidia option, do you expect labs outside China to take it? Tell us in the comments.
