Illustration for: Huawei Speeds Up AI Chip Launch To Challenge Nvidia

Huawei Speeds Up AI Chip Launch To Challenge Nvidia

Huawei pulled forward the launch of its next-generation Ascend 960DT AI chip to early 2027, three quarters ahead of schedule, as China pushes to close the AI compute gap with Nvidia.

By the Numbers

Q1 2027
New Ascend 960DT target
Q3 2027
Original target
4,096
SuperPoD chips demoed
15,488
SuperPoD chips previously claimed
Ascend 970/980, 2028-29
Next chip series
TC
By the AI Desk
Edited by Trace Cohen · Early-stage VC & angel · Founder, New York Venture Partners
1 min read
ShareXLinkedInEmail

THE RUNDOWN

1

Huawei moved the launch of its next-generation Ascend 960DT AI chip up to Q1 2027 from a planned Q3 2027, announced Thursday at its Huawei Connect conference in Shanghai.

2

The company unveiled its Atlas 960 SuperPoD compute cluster, but analysts flagged the demonstrated system scaled to 4,096 Ascend 960 chips versus the 15,488-chip figure Huawei had previously claimed the architecture would reach.

3

The announcement lands days before an expected Trump-Xi meeting in Washington on September 24, adding a geopolitical layer to a chip roadmap update that would otherwise be a routine industry disclosure.

4

Huawei's roadmap now extends to Ascend 970 and 980 chip series in 2028 and 2029, positioning the company as China's primary answer to a Nvidia export-control regime it cannot legally buy around.

TC

The VC Read · Trace's Take

Trace Cohen

The 4,096-versus-15,488-chip gap is the number to actually underwrite, not the accelerated launch date -- announcing a faster timeline costs Huawei nothing, but shipping a cluster at 26% of a previously claimed scale is the kind of detail that tells you how far off full-scale production still is. Anyone diligencing China-adjacent chip supply chains should be asking Huawei directly why that number moved, not taking the roadmap slide at face value.

Analysis

Huawei moved up the launch of its next-generation Ascend 960DT AI chip to the first quarter of 2027, three quarters earlier than its previously stated Q3 2027 target, the company announced Thursday at its Huawei Connect conference in Shanghai, according to TechCrunch. The acceleration is Huawei's clearest public signal yet that it intends to compete on release cadence, not just raw specification, against Nvidia.

Huawei also introduced its Atlas 960 SuperPoD compute cluster at the event, describing it as a faster successor to a model launched only months earlier. TrendForce reported that analysts flagged a discrepancy worth taking seriously: Huawei had previously said the Atlas 960 SuperPoD architecture would scale to 15,488 Ascend 960 chips, but this week's demonstrated system referenced only 4,096 chips -- a fraction of the earlier claim, and a gap Huawei did not address on stage.

Huawei also introduced its Atlas 960 SuperPoD compute cluster at the event, describing it as a faster successor to a model launched only months earlier.

Pulse has tracked Huawei's chip ambitions before, including the Kirin 9050 Pro, and the pattern is consistent: Huawei announces aggressive specifications and timelines under US export-control pressure, then ships hardware that narrows but does not close the performance gap with Nvidia's current-generation parts. Huawei's roadmap now extends to Ascend 970 and 980 chip series in 2028 and 2029, a multi-year commitment to a homegrown compute stack that China has pursued more urgently since US-led restrictions cut off access to the most advanced Nvidia chips and the EUV lithography tools needed to manufacture comparable silicon domestically.

The announcement's timing is not incidental: it came days before an expected meeting between US President Donald Trump and Chinese President Xi Jinping in Washington on September 24, where chip export policy is likely on the agenda. A roadmap acceleration announced this close to a bilateral summit reads as a negotiating signal as much as a product update -- Huawei demonstrating self-reliance progress gives Beijing leverage at the table, regardless of whether the 4,096-chip SuperPoD actually matches Nvidia's current Blackwell-generation clusters on delivered performance, which independent benchmarks have not yet confirmed either way.

ShareXLinkedInEmail

More on

Nvidia

Key Sources

2 sources

Reported by TechCrunch · Analysis by Value Add Pulse.

← Back to Pulse

THE WIRE in your inbox— Tech, startup & VC news with Trace's take. Free, no spam.