Nvidia CEO Jensen Huang has openly praised Huawei’s latest AI innovation, calling the Chinese giant’s newly unveiled Atlas 950 Supernode “undeniably competitive” and capable of rivaling Nvidia’s Grace Blackwell systems. In a September 2025 Bloomberg interview, Huang acknowledged Huawei’s evolution, stating the company’s AI clusters have become “quite formidable” despite U.S. sanctions, marking a candid admission of Beijing’s closing gap in the global AI race. For tech executives, investors, and policymakers searching Nvidia Huawei AI Supernode competitive, Jensen Huang Huawei comments, or US-China AI rivalry 2025, Huang’s remarks—delivered amid China’s reported ban on Nvidia’s RTX Pro 6000D chip—underscore the escalating competition, with Huawei claiming its Supernode offers world-class computing without U.S. tech.
As Huawei pushes self-reliance with its Ascend chips, Huang’s balanced view—disappointment in restrictions but respect for rivals—highlights the stakes in a market where Nvidia holds 80-90% share, yet faces eroding dominance in China.
Huawei’s Supernode: A ‘Formidable’ Leap in AI Clusters
Unveiled on September 18, 2025, at Huawei’s Connect Conference in Shanghai, the Atlas 950 Supernode aggregates 8,192 Ascend chips into a high-performance cluster, enabling ultra-fast interconnects for large-scale AI training. Huawei claims it’s the “world’s most powerful” by computing capability, with a SuperCluster scaling to over 500,000 chips—surpassing Nvidia’s setups in raw node density.
Huang, in his Bloomberg discussion, equated its performance to Nvidia’s Grace Blackwell, noting Huawei’s innovations in system design overcome U.S. export limits on advanced nodes. The Supernode’s architecture—linking SuperPods and SuperClusters—targets telecom, manufacturing, and AI labs, with Huawei delivering 300+ Atlas 900 units to 20 customers already.
| Feature | Huawei Atlas 950 Supernode | Nvidia Grace Blackwell Equivalent |
|---|---|---|
| Chip Count per Node | 8,192 Ascend Chips | Comparable Multi-GPU Cluster |
| Scalability | 500,000+ in SuperCluster | DGX SuperPOD (Thousands of GPUs) |
| Interconnect | Ultra-Fast Domestic Links | NVLink/NVSwitch |
| Target Use | AI Training, Large Models | Similar High-Performance Computing |
Huang’s Comments: Disappointment Meets Respect
Huang’s praise follows a pattern: In May 2025, he called Chinese rivals “formidable” as Huawei’s CloudMatrix matched Grace Blackwell specs. On September 18, responding to China’s Nvidia ban, he expressed “disappointment” but patience, citing larger US-China agendas.
This comes after Nvidia lobbied U.S. lawmakers in May 2025 on Huawei’s AI rise, warning bans could boost demand for Huawei chips if optimized models like DeepSeek R1 train on them. Huang’s tone balances competition with realism: “The Chinese competitors have evolved.”
Broader Rivalry: China’s Self-Reliance vs. U.S. Dominance
Huawei’s Supernode launch—timed before a Xi-Trump call—advances Beijing’s Made in China 2025 goals, with a roadmap for Ascend 950PR by Q1 2026 matching Nvidia/AMD timelines. Despite U.S. curbs since 2019, Huawei’s domestic processes yield clusters rivaling Nvidia’s, potentially narrowing the AI gap.
Implications:
- For Nvidia: China sales (20% revenue) face pressure; Huang’s “patient” stance hints at diversification.
- Global AI Race: Huawei’s 300+ deployments signal viable alternatives, risking U.S. lead if export controls backfire.
- Market Reaction: Nvidia shares dipped 2% post-announcement, reflecting concerns. CNBC
Conclusion: Huawei’s Supernode Forces Nvidia’s Hand
Nvidia CEO Jensen Huang’s nod to Huawei’s AI Supernode as undeniably competitive is a wake-up call: Sanctions spur innovation, and China’s AI ambitions are no longer dismissible. As Huawei scales to 500,000-chip clusters, the US-China tech duel intensifies—Huang’s respect masks strategic urgency. For those tracking AI chip rivalry 2025, the next Ascend vs. Blackwell showdown could redefine global computing. Will Huawei close the gap, or will U.S. edges prevail? The clusters are computing.
