Key Takeaways
- Huawei has accelerated the launch of its next-generation Ascend 960DT AI chip to Q1 2027, aiming to challenge Nvidia (NVDA)'s dominance in the Chinese market.
- Executives claim the software divide between Huawei’s ecosystem and Nvidia’s CUDA is narrowing through architecture redesigns and "software-hardware co-optimization."
- The company unveiled the Atlas 960 SuperPoD, a computing cluster capable of linking up to 100,000 chips to offset individual hardware performance gaps compared to U.S. rivals.
- Major Chinese AI labs, including DeepSeek, are already deploying massive quantities of Huawei hardware, with orders reaching at least 160,000 Ascend chips for new data centers.
Huawei Technologies used its annual summit in Shanghai on Thursday to signal a major escalation in the global AI hardware race. Rotating Chairman Wang Tao announced that the company is pulling forward the commercial availability of its flagship Ascend 960DT chip to the first quarter of 2027, several months ahead of its original schedule. A second variant, the Ascend 960PR, is slated for delivery in Q3 2027.
The announcement highlights Huawei's growing confidence in its ability to provide a viable domestic alternative to Nvidia (NVDA). While Huawei executives acknowledge that their individual chips still trail Nvidia’s top-tier offerings in raw processing power, they argue that the "effective gap" is closing. Guo Ping, chairman of Huawei’s supervisory board, noted that the company is innovating through chip architectures and software optimization, drawing a parallel to how Apple (AAPL) eventually dominated the processor market despite not being an early leader.
Central to this strategy is Huawei's UnifiedBus technology and its SuperPod clusters. By using proprietary interconnects that Huawei claims are significantly faster than existing standards, the company can string together up to 1 million AI processors to function as a single computing entity. This "system-level" approach is designed to mask the limitations of individual chips produced under restrictive U.S. export controls, allowing Chinese firms to train frontier AI models using entirely domestic infrastructure.
Market adoption appears to be accelerating alongside the roadmap. DeepSeek, a prominent Chinese AI lab, is reportedly building a massive data center in Inner Mongolia featuring at least 160,000 Huawei Ascend chips. While many Chinese firms still rely on smuggled or older Nvidia (NVDA) silicon for initial model training, the shift toward Huawei for "inference" and large-scale operations suggests that the software ecosystem—long considered Nvidia's "moat"—is becoming less of a barrier for Chinese developers.
The timing of the reveal is also being viewed as a geopolitical statement, occurring just days before a scheduled meeting between U.S. President Donald Trump and Chinese President Xi Jinping. As the U.S. maintains strict curbs on advanced semiconductor technology, Huawei’s rapid iteration serves as a benchmark for China’s push toward technological self-reliance. Nvidia shares saw a modest decline of nearly 1% in pre-market trading following the news, as investors weighed the potential for a permanent loss of market share in the world’s second-largest economy.
Ed Liston is a senior contributing editor at TheStockMarketWatch.com. An active market watcher and investor, Ed guides an independent team of experienced analysts and writes for multiple stock trader publications.