Huawei is moving faster than expected on its next generation of artificial intelligence chips, putting even more pressure on the race to build alternatives to Nvidia’s dominance in AI computing.
At its annual summit in Shanghai on September 17, Huawei rotating chairman Wang Tao said the company plans to launch its Ascend 960DT chip in the first quarter of 2027, several months earlier than the previously expected late-2027 commercial launch.
A second version, the Ascend 960PR, is expected in the third quarter of 2027.
The announcement is significant because Huawei is no longer simply developing chips for the Chinese market. The company is increasingly presenting its AI hardware as an alternative to US-made semiconductors, with ambitions that extend beyond China.
Huawei is speeding up its roadmap
Huawei has laid out a multi-year plan to develop increasingly powerful AI accelerators.
The company’s current Ascend 950 family is being followed by a sequence of new generations:
- Ascend 960DT: planned for the first quarter of 2027
- Ascend 960PR: planned for the third quarter of 2027
- Ascend 970: planned for 2028
- Ascend 980: planned for 2029
Huawei has said each generation should roughly double the computing power of its predecessor.
That kind of development schedule shows how seriously the company is approaching the AI chip market. It also reflects the broader pressure on Chinese technology companies to build domestic alternatives to advanced US semiconductors.
The bigger bet is not just the chip
One of the most interesting parts of Huawei’s strategy is that it is not relying entirely on making a single chip as powerful as Nvidia’s latest processors.
Instead, Huawei is increasingly focused on the entire computing system.
The company is developing its cluster-based SuperPod technology, which is designed to connect large numbers of Ascend chips and allow them to work together.
Huawei says the system will eventually be able to connect up to 100,000 Ascend chips in a cluster.
The networking layer is based on Huawei’s UnifiedBus protocol, which is intended to improve the speed at which data moves between chips.
This matters because modern AI systems are not built around one processor working alone. Training and running large AI models requires thousands of processors, fast memory, high-speed networking and software that can coordinate everything efficiently.
That creates another battleground beyond raw chip performance.
Why system-level performance matters
Nvidia has built a powerful position in AI partly because of its ability to combine GPUs, networking, software and complete data center systems.
Huawei is pursuing a similar system-level approach.
The company argues that improvements in networking, packaging and system architecture can help compensate for disadvantages at the individual chip level.
Morgan Stanley analyst Charlie Chan recently highlighted this broader shift, pointing to technologies such as:
- Multi-die chip design
- Advanced packaging
- Rack-scale architecture
- Optical networking
- Hardware and software co-optimization
The idea is straightforward: the fastest individual chip does not necessarily determine the performance of an entire AI data center.
How efficiently thousands of chips communicate and operate together can become just as important.
DeepSeek is becoming an important customer
Huawei’s push is also gaining relevance because of demand from China’s AI industry.
DeepSeek is planning to deploy at least 160,000 Huawei Ascend 950DT chips at a data center being built in Inner Mongolia to operate its AI models.
That is a major potential deployment for Huawei’s hardware and could provide an important real-world test of how its chips perform at scale.
But there is an important distinction.
Chinese AI companies are not completely independent of Nvidia yet.
DeepSeek and other frontier AI developers continue to rely heavily on Nvidia hardware for training their models, while domestic chips are increasingly being used for other workloads such as inference.
That highlights the challenge Huawei still faces.
Nvidia remains difficult to replace
Huawei’s progress does not mean Nvidia has suddenly lost its technological advantage.
The company’s AI chips continue to trail Nvidia’s most advanced offerings, particularly at the high end.
Nvidia’s latest Blackwell processors are also not directly available in China because of US restrictions.
This has created an unusual market situation.
China has enormous demand for AI computing, but access to Nvidia’s most advanced processors is restricted. At the same time, domestic companies such as Huawei are trying to fill the gap with locally developed hardware.
The result is creating a strong incentive for Chinese companies to accelerate development of domestic alternatives.
Production could be the biggest challenge
Developing a competitive AI chip is only one part of the problem.
Huawei also needs to manufacture enough of them.
The company is already facing production constraints as demand for AI computing rises in China.
Morgan Stanley estimates that China’s AI computing market could reach 646 billion yuan, or about $96 billion, by 2030.
Huawei has already raised the price of its Ascend 950DT chip by 60% this summer, citing tight component supply.
That gives an indication of how difficult the supply side of the market has become.
If Huawei wants to challenge Nvidia at scale, it will need not only competitive technology but also reliable access to manufacturing capacity, advanced packaging, memory and other components.
Memory is another important piece
AI accelerators depend heavily on high-performance memory.
Chinese AI chipmakers have been cut off from the latest AI memory products from major suppliers including SK Hynix, Samsung Electronics and Micron Technology as a result of US export controls.
Huawei has responded by developing more components internally.
The Ascend 950 chips are the first Huawei AI accelerators to use memory chips designed in-house by the company, although Huawei has disclosed limited details about its production capabilities.
This is part of a much bigger effort to reduce dependence on foreign technology.
The challenge is that building an entire semiconductor ecosystem domestically is considerably harder than designing a processor.
Huawei is also working on optical technology
Another part of Huawei’s strategy involves improving how data moves between AI chips and servers.
The company is developing near-packaged optics, or NPO, technology designed to increase data transmission speeds compared with conventional pluggable optical modules.
Huawei expects to ship its first NPO modules in the coming quarters.
This may sound like a supporting technology, but it is becoming increasingly important as AI clusters grow.
The more processors that are connected together, the more data has to move between them. Faster and more reliable connections can therefore have a direct impact on overall system performance.
The global ambition is becoming clearer
Huawei’s AI chip strategy is not limited to replacing Nvidia inside China.
The company is also exploring international markets, including Malaysia and Egypt, for AI infrastructure projects.
That gives the semiconductor competition a broader dimension.
For years, the AI chip market has been heavily concentrated around US technology. Huawei is now trying to build an alternative ecosystem that includes processors, networking, memory, optical technology and software.
If those technologies can be deployed outside China, the competition could eventually extend beyond the Chinese market.
But that will depend on technology, production capacity, software compatibility, pricing and the geopolitical environment.
US restrictions are shaping the market
Huawei’s acceleration cannot be separated from the restrictions that have been placed on China’s access to advanced semiconductor technology.
US export controls have limited Chinese companies’ access to some of the world’s most advanced AI processors and memory products.
Those restrictions were intended to limit China’s ability to develop cutting-edge computing capabilities.
At the same time, they have increased the incentive for Chinese companies to develop alternatives.
Huawei is at the center of that effort.
The company is effectively trying to solve several problems at once: build more capable AI chips, manufacture them at scale, develop domestic memory, improve networking and create a complete technology stack that can operate without relying as heavily on US suppliers.
The real test will be scale
Huawei’s accelerated Ascend 960 roadmap is an important development, but the launch date alone will not determine how successful the strategy becomes.
The bigger questions will come after the chips are available.
Can Huawei manufacture them in large quantities?
Can its chips deliver competitive performance across large AI clusters?
Can its software ecosystem support increasingly sophisticated AI models?
Can Chinese companies shift more of their training workloads onto domestic hardware?
And perhaps most importantly, can Huawei build an ecosystem that customers are willing to use at scale?
These questions matter because Nvidia’s advantage has never been based solely on the performance of an individual processor. Its hardware, networking, software and developer ecosystem have all contributed to its position in AI computing.
Huawei is now trying to compete across those same layers.
A changing AI chip landscape
The AI semiconductor race is becoming more complicated.
It is no longer simply a contest between one chipmaker and another. It is increasingly a competition between complete computing ecosystems.
Huawei’s accelerated launch of the Ascend 960DT shows that the company wants to move quickly.
Its SuperPod architecture, UnifiedBus networking, optical technology, internally designed memory and multi-year chip roadmap all point toward the same goal: building a domestic AI computing platform that can operate at enormous scale.
For Nvidia, the Chinese market remains constrained by US export rules, while Huawei faces its own technological and manufacturing challenges.
For China’s AI industry, the stakes are even higher.
The ability to build and operate large AI systems increasingly depends on access to computing power. Huawei is betting that by combining faster chips with better networking, packaging and system design, it can close some of the gap with the world’s leading AI hardware.
The next few years will show whether that strategy can move from ambitious roadmap to large-scale reality.