Huawei is preparing two new artificial-intelligence processors for 2027 as the Chinese technology giant pushes deeper into the market for large-scale AI computing and seeks to build an alternative hardware stack around its Ascend platform.
Huawei rotating chairman David Wang said on September 17 that the company plans to launch the 960DT in the first quarter of 2027, followed by the Ascend 960PR in the third quarter. The announcement, reported by Reuters, adds a fresh milestone to Huawei’s longer-term effort to expand AI compute despite restrictions on access to some advanced semiconductor technologies.
Two Chips, Two Steps in Huawei’s 2027 Roadmap
The two processors sit inside Huawei’s broader Ascend strategy for AI accelerators. The company has not disclosed a complete set of specifications for the newly announced 960DT and 960PR in its latest comments, so performance comparisons with Nvidia or other accelerator platforms would be premature.
Huawei’s earlier public roadmap provides useful context. At HUAWEI CONNECT 2025, the company said the Ascend 960 would be designed to roughly double several key capabilities of its 950 generation, including computing power, memory capacity, memory-access bandwidth and the number of interconnect ports. Huawei’s published roadmap at the time placed the Ascend 960 in the fourth quarter of 2027. The latest announcement indicates that Huawei is now planning two 960-series products within 2027 rather than treating the generation as a single launch.
In that 2025 presentation, Huawei said the Ascend 960 was targeting 2 PFLOPS of FP8 compute and 4 PFLOPS of FP4 compute, alongside support for its proprietary HiF4 data format. Those figures should be treated as Huawei’s previously stated targets rather than confirmed specifications for the newly announced 960DT and 960PR.
Huawei’s technical roadmap also shows why the company is focusing heavily on interconnect technology. As AI models become larger, the performance of a single accelerator is only part of the equation; the ability to connect thousands of processors with low latency and high bandwidth becomes critical.
UnifiedBus Is Central to the Strategy
Wang said Huawei’s UnifiedBus technology will be a key part of its next generation of large AI computing systems. The interconnect is intended to allow large numbers of chips to operate together as a more tightly integrated computing system.
This is important because modern AI infrastructure is increasingly built around clusters rather than individual accelerators. Training and serving large models require processors to exchange enormous amounts of data, and communication between chips can become a bottleneck if the interconnect is not fast enough.
Huawei has been developing this architecture as part of its SuperPoD approach. The company describes a SuperPoD as multiple physical machines that are deeply interconnected so they can operate as a single logical computing system. Its SuperPoD architecture is built around UnifiedBus and is intended to scale AI computing beyond the limits of individual servers.
According to Wang’s latest comments reported by Reuters, Huawei has already shipped more than 1,000 AI computing systems to over 370 customers. That does not establish parity with the largest global AI accelerator suppliers, but it does show that Huawei’s strategy extends beyond designing chips to deploying complete computing systems.
Why Huawei Is Building More Than Just a Chip
The bigger story is Huawei’s attempt to control more of the AI-computing stack. A competitive accelerator needs more than raw compute. Memory bandwidth, networking, software, system design and developer support all affect how efficiently an AI model can actually run.
Huawei’s approach is therefore closer to an infrastructure platform than a conventional processor roadmap. Ascend accelerators sit alongside SuperPoD systems, networking technologies and Huawei’s software ecosystem. The objective is to make a large collection of locally developed components work together closely enough to provide a viable alternative for customers building AI infrastructure.
This strategy has become more important as access to leading-edge semiconductor technologies has become a geopolitical issue. Huawei has been subject to extensive U.S. export restrictions, limiting its access to certain advanced chips, manufacturing technologies and equipment. Rather than relying solely on smaller individual processors, the company has increasingly emphasized architectural improvements and connecting more available compute resources together.
HBM Is Still a Major Constraint
Huawei’s push is also happening against a difficult memory-supply backdrop. Reuters reported earlier in September that Chinese AI-chip companies, including Huawei, had raised prices amid a shortage of high-bandwidth memory, or HBM.
HBM is essential to modern AI accelerators because it can feed processors with data at extremely high bandwidth. As AI models and inference workloads grow, memory capacity and bandwidth can become just as important as arithmetic performance.
That makes Huawei’s longer-term efforts around memory and interconnects strategically significant. In its 2025 roadmap, the company described higher-bandwidth memory technologies for its 950 generation and positioned UnifiedBus as a way to connect much larger pools of compute.
Huawei Is Challenging a Market Dominated by Nvidia
Nvidia remains the benchmark against which virtually every high-end AI accelerator is judged. Its advantage is not limited to GPU hardware: the company has built a large software ecosystem, networking portfolio and developer community around its CUDA platform and data-center products.
Huawei therefore does not need to win simply by producing a faster chip. Its more realistic strategic objective is to create a sufficiently capable, integrated alternative that Chinese organisations can deploy at scale, particularly when access to foreign hardware is constrained.
The result is a different kind of competition. Nvidia’s ecosystem has been built around global developer adoption and a huge installed base, while Huawei is combining silicon, networking and complete AI systems with a strong emphasis on domestic technological independence.
What This Means for India
There is no direct India investment or product-launch announcement attached to Huawei’s September 17 chip disclosure, so the immediate effect on Indian consumers is likely to be limited. The significance for India is instead tied to the country’s rapidly expanding AI infrastructure ambitions.
Indian cloud providers, enterprises and AI startups increasingly need access to large-scale compute. A more competitive global accelerator market could eventually give infrastructure operators more choices beyond the dominant U.S.-led hardware ecosystem.
India is also building its own semiconductor and AI infrastructure capabilities. The government’s semiconductor push now covers not only chip fabrication and packaging but also equipment, materials, design and research. At the same time, Indian enterprises are moving from AI experimentation toward larger production deployments. That makes developments in alternative accelerator ecosystems relevant to India’s long-term compute strategy, even when the hardware itself is not being sold locally.
The 2027 Timeline Matters
Huawei’s announcement is notable partly because it puts two new AI processors on a relatively specific schedule. The 960DT is planned for the first quarter of 2027, while the Ascend 960PR is expected in the third quarter.
But product announcements are not the same as mass availability. Actual impact will depend on manufacturing yields, memory supply, system-level performance, software compatibility and how many customers can deploy the processors at scale.
The company’s own deployment figures suggest that Huawei already has an installed base on which to build. If the 960 generation can deliver substantial gains while maintaining compatibility across Huawei’s wider infrastructure stack, it could strengthen the company’s position in China’s rapidly expanding AI-computing market.
A Wider Shift in AI Hardware
Huawei’s roadmap arrives at a moment when the AI hardware market is becoming more strategically important and more fragmented. Nvidia remains dominant, but AMD, custom accelerators from major cloud providers and Chinese chip designers are all competing for parts of the workload.
The competition is also moving beyond the traditional question of who has the fastest accelerator. AI inference is becoming a larger business, model architectures are evolving, and the cost of moving data between processors is becoming a major factor in the economics of large deployments.
That is why technologies such as UnifiedBus deserve attention. If AI systems continue to scale from thousands to hundreds of thousands or even millions of processors, the interconnect between those processors becomes a fundamental part of the computer itself.
The Bottom Line
Huawei’s announcement of two new 2027 AI chips is another sign that the global AI-compute race is shifting from individual accelerators toward complete, interconnected systems. The 960DT is planned for early 2027, followed by the Ascend 960PR later in the year, with UnifiedBus positioned as a core technology for connecting large numbers of chips.
For Huawei, the challenge is not simply to produce competitive silicon. It must turn its chips, memory, interconnects, software and SuperPoD systems into a dependable platform that customers can deploy at scale. If it succeeds, the result could be a more divided global AI hardware market — and more meaningful alternatives for countries and companies looking to diversify their compute infrastructure.
