Huawei is rapidly expanding its artificial intelligence hardware ambitions, offering a clear look at how China is trying to build a more self-reliant semiconductor industry despite restrictions on access to advanced foreign chips and manufacturing technology.
At Huawei Connect 2026 in Shanghai, the company unveiled new AI infrastructure and accelerated the timetable for its next-generation Ascend 960 chips. Huawei says the Ascend 960DT, designed primarily for AI model training, will be available in the first quarter of 2027, while the Ascend 960PR, aimed at inference workloads, is expected in the third quarter of 2027. The company says both are expected to deliver roughly twice the computing performance of the current generation.
The announcements are significant because Huawei’s strategy goes beyond developing individual processors. The company is building an entire AI computing ecosystem around its Ascend chips, networking technology, storage systems, software stack, and large-scale computing clusters.
That broader approach illustrates an important shift in China’s semiconductor push: the objective is increasingly not simply to produce a domestic alternative to Nvidia GPUs, but to create an integrated technology stack that can support China’s growing AI industry.
Huawei Accelerates Its Ascend AI Chip Roadmap
Huawei’s latest roadmap shows how quickly the company is attempting to advance its AI processors.
The Ascend 960DT was originally expected later in 2027, but Huawei has moved its planned availability forward to Q1 2027. The company also says it intends to maintain a roughly one-generation-per-year development cycle, with Ascend 970 planned for 2028 and Ascend 980 for 2029.
The roadmap is important because AI computing is advancing rapidly, with demand for processors increasing as companies train increasingly large models and deploy AI agents at scale.
Huawei is therefore attempting to shorten the development cycle between generations while improving not only raw compute performance but also memory bandwidth, memory capacity, and interconnect capabilities.
Those improvements matter because modern AI systems depend on far more than processor speed.
The Strategy Goes Beyond a Single AI Chip
One of the most important developments in Huawei’s 2026 strategy is its emphasis on SuperPoDs and SuperClusters.
Instead of relying on the performance of one processor, Huawei is connecting large numbers of Ascend processors into tightly integrated computing systems.
The company says its upgraded architecture can scale to systems containing as many as one million NPUs, although that figure describes a broader computing architecture rather than a single commercially available chip. Huawei has also introduced the Atlas 960E SuperPoD and upgraded its UnifiedBus interconnect technology as part of this strategy.
This approach addresses one of the major challenges facing Chinese AI chipmakers.
If domestic processors cannot individually match the most advanced Nvidia accelerators, manufacturers can attempt to compensate by connecting larger numbers of chips efficiently.
That makes networking and system architecture increasingly important parts of the AI semiconductor competition.
Why Networking Matters in China’s AI Chip Race
Training advanced AI models requires thousands or even tens of thousands of processors to communicate rapidly.
If communication between processors becomes a bottleneck, simply adding more chips does not necessarily produce proportional increases in computing performance.
Huawei is therefore investing heavily in UnifiedBus, its interconnect architecture designed to link computing, memory, storage, and networking components.
Huawei says its next-generation SuperPoD architecture can support up to 4,096 nodes and a unified memory pool of up to 256 TB. The company also says it has deployed more than 1,000 Atlas 900 A3 SuperPoDs and that its Atlas 950 SuperPoD is seeing large-scale commercial use.
These are company-reported figures, so they should be understood in that context. Nevertheless, they demonstrate where Huawei believes it can differentiate: not necessarily by making a single chip that matches every capability of the world’s most advanced AI accelerators, but by optimizing the complete computing system.
U.S. Export Controls Are Accelerating China’s Localization Push
Huawei’s semiconductor strategy is closely connected to years of U.S. restrictions on China’s access to advanced semiconductor technology.
Export controls have limited Chinese companies’ ability to obtain some of the world’s most advanced AI accelerators and semiconductor manufacturing equipment. These restrictions have also strengthened incentives for Chinese companies to develop domestic alternatives.
This creates a complicated situation.
Restrictions can make it harder for Chinese companies to obtain cutting-edge components, but they can simultaneously strengthen incentives to develop domestic alternatives.
Huawei’s expanding Ascend ecosystem is one of the clearest examples of that dynamic.
Rather than abandoning advanced AI hardware ambitions because of restricted access to foreign technology, Huawei is attempting to redesign parts of the technology stack around what can be produced and integrated within China’s broader semiconductor ecosystem.
HBM Has Become a Major Bottleneck
Developing an AI accelerator is only one part of the challenge.
Modern AI processors require high-bandwidth memory, or HBM, to move huge amounts of data between memory and computing units. The global HBM market is dominated by companies such as SK Hynix, Samsung, and Micron, while restrictions on advanced HBM exports to China have created additional supply challenges for Chinese AI chipmakers.
Recent reporting has highlighted the impact of HBM shortages on Chinese AI accelerator prices. Reuters reported in September 2026 that Huawei had raised the indicated price of its upcoming Ascend 950DT accelerator card to more than 250,000 yuan, citing people familiar with the matter. The report attributed the price pressure partly to the global HBM shortage and difficulties obtaining advanced memory.
This highlights an important reality about China’s semiconductor push.
Even if Chinese companies can design increasingly capable AI processors, they still need access to advanced memory, packaging, manufacturing equipment, substrates, networking components, and other parts of the supply chain.
Building a competitive AI chip ecosystem therefore requires progress across multiple semiconductor technologies simultaneously.
Huawei Is Also Building a Domestic Software Ecosystem
Hardware is only half of the AI accelerator equation.
Nvidia’s GPUs benefit from a mature software ecosystem, particularly CUDA, which has become deeply integrated into AI development workflows.
Huawei is attempting to reduce that software disadvantage by expanding its own developer ecosystem around Ascend and its CANN software platform.
Huawei says Ascend now supports more than 90 major open-source projects, including PyTorch, Triton, vLLM, and veRL. The company also says Ascend has become an officially supported PyTorch accelerator backend.
The importance of this effort should not be underestimated.
Developers are less likely to switch hardware platforms if doing so requires rewriting large portions of their software. A growing collection of frameworks, libraries, tools, and optimized applications can therefore make domestic AI chips considerably more useful.
Huawei is attempting to turn Ascend from a processor product into a broader developer platform.
China’s AI Chip Market Is Becoming More Competitive
Huawei is not the only Chinese company developing AI accelerators.
Cambricon, MetaX, Iluvatar CoreX, Alibaba and other domestic players are also developing chips or AI computing technologies.
Recent reports show that demand for Chinese AI accelerators is rising. Reuters reported that Huawei and other Chinese chipmakers increased prices amid HBM shortages, while domestic companies were attempting to supply the growing AI infrastructure market.
Alibaba has also unveiled new AI accelerator technology, illustrating how China’s major technology companies are investing in alternatives to foreign AI hardware.
The result is an increasingly competitive domestic ecosystem rather than a market centered entirely on one supplier.
That competition could help China develop expertise across chip design, manufacturing, networking, software, and AI infrastructure.
Huawei Says Domestic Demand Is Already Strong
Huawei’s latest comments also suggest that domestic demand is becoming a major driver of its AI hardware business.
At Huawei Connect 2026, rotating chairman Eric Xu said the company could not currently produce enough AI computing equipment to satisfy Chinese demand and therefore was limiting international expansion. Xu also claimed that Ascend’s market share in China had surpassed Nvidia’s, while acknowledging that reliable market-share data was difficult to obtain.
That claim should be treated as Huawei’s own estimate rather than an independently verified market measurement.
Nevertheless, the company’s comments highlight an important development: Chinese AI customers are increasingly becoming a built-in market for domestic accelerators.
That gives Chinese chipmakers an opportunity to improve their products through large-scale deployment inside the domestic technology sector.
Can China Close the AI Semiconductor Gap?
China’s progress should not be confused with complete semiconductor self-sufficiency.
Advanced chip manufacturing remains highly complex, and restrictions on leading-edge manufacturing equipment and advanced memory continue to create obstacles.
At the same time, analysts have documented China’s progress in using older-generation lithography equipment to produce increasingly advanced chips. An April 2026 analysis from the American Enterprise Institute argued that China’s large installed base of deep-ultraviolet lithography equipment could support substantial production of advanced logic chips, although the technical and economic limitations remain significant.
Huawei is consequently pursuing several solutions at once: improving chip architecture, developing domestic software, strengthening networking, optimizing memory systems, and building larger computing clusters.
The strategy is less about overcoming one technological barrier and more about reducing dependence across the entire AI computing stack.
What Huawei’s AI Chip Push Means for the Semiconductor Industry
Huawei’s latest AI chip roadmap demonstrates how China’s semiconductor strategy is evolving.
The focus is moving from simply producing individual domestic components toward building integrated systems capable of supporting large AI models and data centers.
The Ascend 960 roadmap, SuperPoD architecture, UnifiedBus networking, domestic software development, and investment in AI infrastructure all form parts of the same strategy.
This could have consequences well beyond China.
As domestic AI chips improve, Chinese technology companies may have more options when purchasing computing hardware. Global AI infrastructure could also become increasingly divided between technology ecosystems built around different combinations of processors, software, networking standards, and supply chains.
For Nvidia and other international semiconductor companies, the emergence of stronger Chinese alternatives could create a more competitive market in regions where those products can legally be sold.
For China, meanwhile, the bigger objective is resilience.
Final Thoughts
Huawei’s new AI chip roadmap offers a useful snapshot of China’s broader semiconductor ambitions in 2026.
The company is accelerating the development of its Ascend AI processors, expanding its software ecosystem, investing in high-speed networking, and connecting large numbers of chips into SuperPoD and SuperCluster systems.
The strategy reflects a basic reality of modern AI: performance increasingly depends on the entire computing system rather than a single processor.
Huawei still faces significant challenges, particularly around advanced manufacturing, HBM supply, production capacity, and access to leading-edge semiconductor technology. But the company is responding by building a broader domestic technology stack designed to reduce reliance on foreign suppliers.
The next major milestone will be the planned Ascend 960DT launch in Q1 2027. Its real-world performance, availability, manufacturing scale, software compatibility, and ability to support large AI deployments will provide a clearer indication of how far China’s AI semiconductor industry has advanced.
For now, Huawei’s progress shows that China’s semiconductor push is no longer focused solely on catching up chip by chip. It is increasingly about building an entire AI computing ecosystem from processors and memory to networking, software, and large-scale infrastructure.
