Huawei will launch two new AI chips in 2027, the company's rotating chairman David Wang announced on Thursday, accelerating the Chinese tech giant's attempt to build a domestic alternative to Nvidia's dominant AI accelerators.
According to Reuters, the 960DT is slated to arrive in the first quarter of 2027, followed by the Ascend 960PR in the third quarter of the year. The announcement lands amid intensifying US export controls on advanced chip technology destined for China — restrictions that have cut Nvidia off from much of the Chinese market and turned Huawei into the default domestic supplier for AI compute inside the world's second-largest economy. For a wider view of the industry, follow our latest AI developments.
The dual launch is the most concrete milestone yet in a chip strategy Huawei has been laying out publicly since last year, and it arrives with unusual timing: the company used the moment to detail not just new silicon, but the interconnect architecture meant to make thousands — eventually millions — of Chinese-made chips behave like one machine.
A two-chip roadmap for 2027
The two new Ascend devices extend a roadmap Huawei has been publicizing since last year, when it first laid out a multi-generation plan for AI silicon. Wang, speaking in the company's customary rotating-chairman role, said the 960DT would launch in the first quarter of 2027 with the Ascend 960PR following in the third quarter, per the Reuters report.
The unveilings drew international attention within hours. Bloomberg framed Huawei's latest hardware as China's best answer to Nvidia's AI chip reign, while the South China Morning Post described the announcements as the latest step in Huawei's effort to boost China's AI computing power and curb its reliance on US suppliers.
UnifiedBus and the path to million-processor systems
Arguably the more strategically significant piece of the announcement is not a single chip but the plumbing between them. Huawei's UnifiedBus technology is designed to link very large numbers of AI chips so they operate as one cohesive, high-power computing system — the same architectural bet Nvidia makes with NVLink and its rack-scale platforms, and one that has become essential as frontier models outgrow what any single GPU can deliver.
Huawei says UnifiedBus already underpins 11 semiconductor models, enabling systems that can potentially integrate up to one million AI processors — the kind of supernode and supercluster scale that frontier model training increasingly demands. The company also says it has shipped more than 1,000 AI systems to over 370 customers, an early but tangible footprint for its cluster business as Chinese labs and state-backed buyers seek domestic alternatives.
Why it matters for Nvidia — and for Washington
The context for the announcement is as important as the chips themselves. Sweeping US export restrictions have progressively cut Nvidia off from the Chinese market, and Beijing has pushed domestic procurement in response. Tom's Hardware reported late last year that China's government-approved AI hardware supplier list includes Cambricon and Huawei — with Nvidia left off entirely.
The commercial effect is already visible in Nvidia's own numbers and statements. Nvidia chief executive Jensen Huang said in May that his company had "largely conceded" China's AI chip market to Huawei, according to CNBC — a striking admission for a firm that once dominated it. Huawei, meanwhile, told investors it expects roughly $12 billion in AI chip revenue this year, Tom's Hardware reported, as Nvidia's China market share slid toward zero.
Huawei is also talking up the demand side of the equation. At its customer conference this week, the company forecast that billions of AI agents will dominate internet traffic by 2035, per Reuters — a projection that doubles as the business case for the massive computing systems it is racing to build at home.
Chinese labs are already buying in
Domestic demand is materializing at flagship scale. Bloomberg reported this month that DeepSeek plans a major order of Huawei's Ascend 950DT chips — more than 160,000 of them — to power a new data center, one of the clearest signals yet that China's most prominent AI lab intends to train on domestic silicon rather than stockpiled Nvidia hardware.
That order matters for the 2027 roadmap, too. Every generation of Ascend chips depends on convincing Chinese model developers that the hardware, software tools and cluster reliability are production-grade. A successful 960-series launch would give Huawei three consecutive generations of credible accelerators — and give DeepSeek, Z.AI, Moonshot and other Chinese labs a reason to stay domestic.
What to watch as 2027 approaches
Announcements are easier than silicon. Between now and launch, the questions to watch are production capacity at Chinese fabs, manufacturing yields for advanced nodes under sanctions, and the maturity of Huawei's software ecosystem for developers weaned on Nvidia's CUDA — historically the hardest moat to cross.
If the 960DT and Ascend 960PR ship roughly on schedule, 2027 would mark the point at which China's homegrown AI hardware moves from stopgap to strategic alternative, with UnifiedBus-scale clusters as the centerpiece. For Nvidia, which remains dominant everywhere else in the world, the one big market it once led has become the first it must fight to re-enter.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
