Huawei Unveils New AI Chips to Challenge Nvidia

Huawei is accelerating its next generation of artificial-intelligence chips, setting up a more direct challenge to Nvidia as China races to build a domestic AI computing ecosystem that is less dependent on U.S. technology.

Huawei has unveiled an expanded roadmap for its Ascend AI processors and large-scale computing systems, including new chips scheduled to arrive beginning in early 2027. The announcement puts one of China’s largest technology companies squarely into an increasingly competitive global battle over the hardware powering generative AI.

At Huawei Connect 2026, the company said its Ascend 960DT processor is now expected in the first quarter of 2027, several months earlier than originally planned. The Ascend 960PR is expected to follow in the third quarter. Huawei says development has progressed faster than expected and that the 960 generation will roughly double computing performance compared with its previous generation.

The company also outlined plans for Ascend 970 chips in 2028 and Ascend 980 chips in 2029, signaling that Huawei intends to move toward an annual release cycle for its AI processors.

Huawei Is Going After More Than the Chip

The strategy isn’t simply about producing an individual processor capable of competing with Nvidia’s GPUs.

Huawei is increasingly focused on connecting enormous numbers of processors into integrated AI computing systems.

The company unveiled its Peerium Computing Architecture, which Huawei says can allow as many as one million processors to operate together as a single computing system. The architecture uses Huawei’s UnifiedBus technology to connect processors and share computing resources across extremely large clusters.

That approach could be particularly important because Nvidia’s advantage isn’t based solely on the performance of an individual GPU. Nvidia has spent years developing a broader ecosystem that includes GPUs, networking, servers and its widely adopted CUDA software platform.

Huawei is attempting to build its own alternative stack.

Demand Is Already Outpacing Huawei’s Supply

The push comes as Huawei says demand for its AI infrastructure inside China is exceeding what the company can currently manufacture.

Huawei rotating chairman Eric Xu said production cannot presently satisfy domestic demand, limiting Huawei’s ability to expand the products internationally. More than 1,000 Huawei AI computing systems have already been deployed, according to the company.

That demand is significant because China represents one of the world’s largest markets for AI infrastructure.

U.S. export controls have restricted China’s access to some advanced American semiconductor technology, including high-end Nvidia products. That has increased the incentive for Chinese companies to develop domestic alternatives.

Huawei has emerged as one of the most prominent companies attempting to fill that gap.

The Bigger Battle: Huawei vs. Nvidia’s AI Ecosystem

Nvidia remains the dominant supplier of GPUs used to train and operate many of the world’s largest AI models.

But Huawei’s latest announcements demonstrate how quickly an alternative AI hardware ecosystem is developing inside China.

Huawei’s strategy appears to emphasize scale as much as individual-chip performance. By combining thousands—and eventually potentially far more—Ascend processors, Huawei can attempt to compensate for limitations created by manufacturing constraints and restricted access to leading-edge semiconductor production technology.

The company has already demonstrated that philosophy with its Atlas SuperPoD systems.

Huawei previously announced the Atlas 950 SuperPoD, scheduled for the fourth quarter of 2026, which can incorporate as many as 8,192 Ascend 950DT processors. Huawei claims a fully configured system can deliver 8 EFLOPS of FP8 computing performance and 16 EFLOPS using FP4. Those are Huawei’s own performance claims and should be treated as such until independently benchmarked across comparable workloads.

Software May Be the Harder Challenge

Building competitive silicon is only one piece of the AI infrastructure race.

Nvidia’s CUDA software platform has become deeply embedded across AI research, cloud computing and enterprise development. Developers have spent years building applications and tools around Nvidia hardware.

Huawei therefore needs to convince developers that its Ascend ecosystem can provide not only competitive computing capacity but also the software, reliability and development tools necessary to run increasingly complicated AI workloads.

That transition could take considerably longer than developing a faster processor.

IEEE Spectrum has noted that matching Nvidia involves much more than raw chip performance, including memory systems, interconnect bandwidth, software ecosystems and the ability to manufacture hardware at scale.

China’s AI Hardware Race Is Getting Crowded

Huawei isn’t alone.

Alibaba, Baidu and other Chinese technology companies are investing heavily in domestic AI processors as China attempts to reduce its reliance on imported technology.

Alibaba, for example, unveiled its Zhenwu V900 AI chip this week and said the processor delivers roughly three times the performance of its predecessor. The company expects mass production in early 2027 and says the processors can operate in extremely large clusters.

That means Nvidia isn’t facing a single Chinese competitor. An entire domestic semiconductor ecosystem is emerging.

The AI Chip Race Is Entering a New Phase

For years, the AI infrastructure story largely revolved around one question: How many Nvidia GPUs can companies obtain?

That is beginning to change.

Google has its TPUs. Major cloud providers are developing custom accelerators. Alibaba is developing its own silicon. And Huawei is building an increasingly ambitious combination of processors, networking technology and massive AI computing clusters.

Nvidia continues to hold substantial advantages, particularly around its mature software ecosystem and leading-edge hardware. Huawei’s new products do not by themselves demonstrate performance parity across real-world workloads.

But Huawei’s roadmap makes one thing clear: China is investing heavily in building an AI computing stack that can operate with substantially less dependence on American technology.

And as AI models become larger and require increasingly massive computing clusters, the competition may no longer be determined by who builds the fastest individual chip.

It may increasingly come down to who can connect the most computing power together—and make the entire system operate efficiently.

Source: Huawei, Reuters, Associated Press, TechCrunch and IEEE Spectrum.