Huawei has once again captured the world’s attention by introducing a new chip strategy positioned as a direct challenge to Nvidia’s dominance. At the Huawei Connect 2025 event in Shanghai, the Shenzhen-based company released an ambitious roadmap for its Ascend AI processor line and announced the development of in-house high-bandwidth memory. Analysts see this as part of China’s broader push to reduce dependence on global supply chains long controlled by the United States, South Korea, and Japan.
Huawei executives emphasized the company’s plan to release a new Ascend chip every year, with the goal of doubling computing power with each release. The roadmap includes the Ascend 950, scheduled for 2026, the Ascend 960 for 2027, and the Ascend 970 for 2028. “We are committed to accelerating innovation cycles so that AI becomes more accessible,” said Zhang Ping’an, CEO of Huawei Cloud, addressing thousands of conference participants.
Breakthrough in memory technology
The centerpiece of Huawei’s new strategy is the claim of developing its own high-bandwidth memory (HBM). This technology has long been dominated by Samsung and SK hynix. Huawei revealed that its in-house HBM can reach 128 GB of capacity with transfer speeds up to 1.6 terabytes per second. These specifications were shared alongside the announcement of the Ascend 950PR, the first variant expected to integrate this module.
The move drew wide attention because global HBM supply remains tight, and U.S. export restrictions have made it a bottleneck for China. If Huawei succeeds in producing HBM at scale, the company could lower costs while securing domestic AI chip supply chains. However, market analysts stress that such performance claims need to be independently verified, and questions remain about whether China can manufacture HBM consistently without access to foreign technology.
Scaling compute power to a global level
Huawei’s vision extends beyond single chips to large-scale distributed systems under its Atlas brand. The Atlas 950, expected to debut in late 2025, will reportedly integrate up to 8,192 Ascend chips into one logical computer. Even more ambitious, the Atlas 960 targeted for 2027 is said to support up to 15,488 chips. This supernode architecture is designed to train massive AI models, an area currently dominated by Nvidia’s DGX systems.
The company also highlighted its software ecosystem. Huawei relies on CANN as its low-level acceleration SDK and MindSpore as its machine learning framework. Both aim to ensure compatibility with industrial-scale AI workloads. Yet compared to Nvidia’s CUDA ecosystem, Huawei’s software is still seen as less mature. The company’s biggest challenge will be building a global developer community and ensuring model portability to encourage international adoption.
Market reactions and geopolitical context
Huawei’s announcement triggered immediate responses in the semiconductor market. Shares of Chinese optical component suppliers surged on expectations of becoming part of the Atlas SuperCluster supply chain. At the same time, global investors viewed the strategy as deeply political, reinforcing Beijing’s ambition for self-sufficiency in AI computing. The United States is expected to tighten restrictions on software exports and lithography tools to hinder the full realization of Huawei’s plan.
India’s IT industry also voiced concerns over a new U.S. policy that imposes a steep H-1B visa fee, which could complicate the talent pipeline for chip and AI sectors. Against this backdrop, Huawei’s bold roadmap looks even more significant, as it could reshape global competition. Some analysts suggest that if Huawei manages to mass-produce HBM, the cost of training large models could drop sharply, offering Asian AI startups an alternative to Nvidia.
Huawei is placing a massive bet in the middle of geopolitical tensions and the global tech race. The company is not only aiming to match Nvidia in performance but also to build an independent ecosystem outside U.S. control. Whether this strategy succeeds remains uncertain, especially given the challenges of scaling production and gaining global market trust. What is clear, however, is that the announcement signals a new phase in the semiconductor race.
Read more analysis on the global AI industry at Olam News and follow updates from trusted international sources to understand the broader business and political impact.
Discover more from Olam News
Subscribe to get the latest posts sent to your email.
