Huawei’s Bold 3-Year Sprint to Catch Nvidia in AI Chips

Roadmap reveals new Ascend chips and supernodes as China doubles down on semiconductor sovereignty

At the Huawei Connect conference in Shanghai, the Chinese tech giant publicly shared for the first time a detailed three-year roadmap aimed at narrowing (and possibly closing) the gap with Nvidia in AI chip technology. New AI accelerators, improvements in memory technology, and large supercomputing clusters (“supernodes” or “superpods”) are all in its plans.

What Huawei Is Rolling Out

• Huawei targets launching four new Ascend chips between 2026–2028: two variants of Ascend 950 in 2026, followed by the Ascend 960 in 2027 and Ascend 970 in 2028.
• It is also introducing high-bandwidth in-house memory (HBM) tech to reduce dependency on U.S. and Korean suppliers, one of the biggest bottlenecks in AI hardware.
• Huawei’s “Atlas 950” and “Atlas 960” supernodes are planned to support large clusters of those Ascend chips, delivering scale via “supernode” architecture with high-speed interconnectivity.
• The company also claims its upcoming platforms may outperform Nvidia’s GB200 NVL72 system in certain metrics. Whether in raw compute or bandwidth, Huawei is signalling performance gains.

Strategic & Market Implications

• China’s push for chip self-sufficiency: With increasing U.S. export controls and global trade tensions, Huawei’s strategy is part of a broader national goal to reduce reliance on foreign semiconductor technologies.
• Competitive pressure on Nvidia and GPU suppliers: If Huawei delivers on its roadmap, Nvidia’s dominance in AI training hardware may face more concerted challenges not only in China but in markets where cost, supply chain control, and nationalism matter.
• Supply chain ripples: Suppliers of HBM, packaging, interconnects, cooling, and power delivery will be under the spotlight. Delays or restrictions in any segment could slow Huawei’s progress.
• Investor sentiment: Markets, especially in China, will likely reward Huawei’s chip development efforts. Meanwhile, Nvidia investors may re-price risk or see more volatility given rising opposition.

Risks & What to Watch Closely

• Manufacturing & process node constraints: Even with strong design roadmaps, Chinese fabs (like SMIC) remain limited in certain advanced node capabilities. Export controls, equipment access, and material supplies are key risk points.
• Software, ecosystem, and co-design: Performance isn’t only about chip hardware. Nvidia benefits significantly from its mature software stack, interconnect sys-
  tems (e.g. NVLink), toolchains like CUDA, and its developer & research ecosystem. Huawei will need to build or emulate similar ecosystems to compete meaningfully.
• Export restrictions and geopolitical risk: U.S. sanctions and controls over chip exports might constrain Huawei’s access to cutting-edge processes, advanced packaging, and high-speed interconnect tech. Any escalation in restrictions could disrupt roadmap timelines.
• Benchmark vs reality: Claims of outperforming Nvidia in some metrics are promising, but benchmarks can be selective. Volume production, power efficiency, thermal management, yield, and reliability are harder to match.

In summary, Huawei’s three-year push reflects an aggressive bet: by rolling out new Ascend chips, developing in-house high-bandwidth memory, and building massive supernodes, the company aims to rival Nvidia more directly. The core theme: China is accelerating its AI chip ambitions from talk to roadmap, but closing the gap won’t be easy, and success depends on execution, supply chain control, and ecosystem strength.