Huawei will ship its next-generation Ascend 960DT AI chip in the first quarter of 2027, two quarters earlier than a previously planned Q3 2027 launch, the company said Thursday at its Huawei Connect conference. The move is Huawei's most direct attempt yet to close the gap with Nvidia in the accelerator market, and it lands one week before a scheduled September 24 meeting between Trump and Xi Jinping in Washington, DC. David Wang, Huawei's rotating and acting chairman, delivered the updated timeline on stage.
A Huawei spokesperson framed the shift as a deliberate cadence change rather than a one-off pull-in. The company is now positioning the Ascend line on a yearly performance-doubling track, a schedule that maps closely onto how Nvidia has publicly described its own accelerator roadmap.
“The Ascend 960DT is expected to be ready in Q1 2027. Ascend 960 chips are launching ahead of schedule, doubling performance and advancing year by year.”— Huawei spokesperson, Huawei
The chip announcement came bundled with a systems story. Huawei is trying to lash hundreds of thousands, and eventually millions, of AI chips into a single logical computer under what it calls its Peerium Computing Architecture. The architecture leans on UnifiedBus, Huawei's in-house fabric for linking processors with memory, storage, and networking hardware.
Key facts
- 01Huawei pulled the Ascend 960DT launch forward to Q1 2027 from Q3 2027, a two-quarter acceleration.
- 02The Atlas 950 SuperCluster can connect up to 256,000 accelerator cards under Huawei's new Peerium Computing Architecture.
- 03The newly announced Atlas 960 SuperPoD system covers 4,096 chips, down from a previously stated 15,488 Ascend 960 chips.
- 04The announcement landed one week before the September 24 meeting between Trump and Xi Jinping in Washington, DC.
- 05David Wang, Huawei's rotating and acting chairman, unveiled the timeline at Huawei Connect on Thursday.
The first products built on Peerium are the Atlas 950 SuperPoD and the Atlas 950 SuperCluster. Huawei says an Atlas 950 SuperCluster can connect up to 256,000 accelerator cards, a scale figure clearly aimed at frontier-training buyers inside China who can no longer procure top-end Nvidia parts.
The accelerated chip timeline came with an awkward downgrade elsewhere on the roadmap. China tech analyst Rui Ma noted on X that Huawei had previously said its Atlas 960 SuperPoD would scale to 15,488 Ascend 960 chips, while this week's announcement referred to a system with 4,096 chips. That is roughly a quarter of the original ambition, and it suggests the interconnect and packaging work is not tracking the silicon.
“The chip itself is coming WAY earlier, but the SuperPoD they announced is much smaller than what they originally laid out.”— Rui Ma, China tech analyst
Eric Xu, Huawei's rotating chairman, said in the company's statement that Huawei is using Peerium to build larger AI computing systems designed for both training and inference. Xu also argued, per the Financial Times, that China needs to accelerate AI development to catch up with the United States, and that Chinese companies need to advance further before they can fully understand and address the risks posed by more powerful AI systems.
Huawei has continued to advance its chip ambitions despite US restrictions on China's access to advanced semiconductor technology. Ma's read is that the restrictions are unlikely to change the trajectory at this point.
The political backdrop is direct. Trump has pushed back against calls from AI industry leaders to slow the development of the technology over safety concerns, arguing that the US needs to maintain its lead over China. A Huawei roadmap that pulls a competitive accelerator forward by six months lands in that debate with obvious force, and does so days before the two heads of state sit down.
The open questions are on the systems side. A Q1 2027 chip is only useful at frontier scale if the SuperPoD fabric, cooling, and software stack can hold together across tens of thousands of parts. The shrink from 15,488 to 4,096 chips in the announced Atlas 960 SuperPoD configuration is the tell — Huawei is comfortable committing to the silicon date but is walking back the maximum system size it will name in public.
Huawei's pitch to Chinese buyers is now straightforward: a yearly Ascend cadence, a proprietary interconnect, and a stated ceiling of 256,000 accelerators per cluster. For Nvidia, the near-term revenue impact inside China is already capped by export controls, so the strategic question is whether Peerium and UnifiedBus mature fast enough to make Huawei a credible seller outside China by the end of the decade. On the chip date, Huawei is now betting yes. On the system scale, the numbers say the harder work is still ahead.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
