Huawei pulled its Ascend 960DT AI chips launch forward by nine months, targeting Q1 2027 instead of Q3 2027, in a direct challenge to Nvidia’s dominance in large-scale AI computing. The announcement, made at Huawei Connect 2026 in Shanghai, arrived one week before a scheduled meeting between President Trump and President Xi Jinping in Washington on September 24.
US export controls have blocked Chinese buyers from accessing top-tier Nvidia GPUs, including the H100. Huawei is moving to fill that gap with domestic silicon and a proprietary cluster architecture announced at the conference.
The Chip and the Claim
Huawei’s rotating chairman put a specific number on the acceleration, and it is a substantial one.
Rotating and acting chairman David Wang announced the updated timeline at the Huawei Connect 2026 keynote, per Huawei’s official press materials.
“Ascend 960DT will be available in Q1 2027, three quarters ahead of the company’s original roadmap.” David Wang, rotating and acting chairman, Huawei Technologies, Huawei Connect 2026 keynote.
Wang described the Ascend 960DT as delivering roughly double the performance of its predecessor. According to The Robotics Media, the Atlas 960 SuperPoD pairs 4,096 Ascend 960 accelerator cards with approximately 8 exaflops of FP8 compute and 1 petabyte of HBM memory, roughly twice the raw compute of the Atlas 950 it replaces.
Architecture and Scale
Huawei’s new interconnect fabric is the infrastructure bet underlying all of its cluster ambitions.
Huawei’s new Peerium Computing Architecture is built around its proprietary UnifiedBus interconnect. It links processors, memory, storage, and networking hardware into a unified fabric designed to reduce protocol-translation overhead across the system.
The current Atlas 950 SuperCluster already connects up to 256,000 accelerator cards, per the Huawei Connect keynote. Huawei’s stated roadmap targets one million NPUs in a single logical system, a scale that underscores the growing demand driving investment in AI-Powered Websites and tools built atop this expanding hardware base.
China tech analyst Rui Ma noted on X, as reported by Huawei Central, that earlier Huawei roadmaps described the Atlas 960 SuperPoD scaling to 15,488 chips. The Huawei Connect announcement centered on a 4,096-chip configuration, suggesting a phased rollout rather than immediate deployment at full scale.
“I think it’s futile to stop China’s development in semiconductors because the stakes for self-sufficiency are just too high at this point.” , Rui Ma, China tech analyst, via Huawei Central.
Geopolitical Stakes
The announcement landed at a moment when AI hardware has become as much a diplomatic variable as a product category.
Trump has argued the US must maintain its AI lead over China. Huawei rotating chairman Eric Xu has countered, according to Quartz, that China must accelerate AI development to catch up before it can fully assess the risks of more powerful systems.
If Huawei delivers the Peerium and Ascend ecosystem at scale, the question shifts from whether China can build competitive AI clusters to whether the global AI data centers and hardware stack will remain unified at all.
