Huawei Unveils the Ascend 960: Its Answer to Nvidia in AI Chips

By Carlos Montiel | Enterprise AI Specialist
Leer en español →
Published: 2026-09-17 | By: Carlos Montiel | Reading time: ~5 minutes

Huawei unveiled the Ascend 960, its next-generation AI chip, at the company's annual summit in Shanghai. It's the Chinese company's most serious attempt yet to close the gap with Nvidia in training and inference compute — and it lands as the company accelerates its entire AI silicon roadmap.

What Huawei showed in Shanghai

Huawei's rotating chairman, Wang Tao, presented the Ascend 960 to the audience at the company's annual summit, with commercial availability projected for 2027. Huawei's stated goal is for its Ascend chips to run any relevant AI model, both inside China and beyond its borders — a direct message to customers who today depend on Nvidia GPUs that China cannot freely buy because of US export controls.

The roadmap is accelerating, not just the flagship chip

More telling than the Ascend 960 itself is the pace: according to industry reports, the Ascend 960DT — a variant in the same family — was moved up three quarters on Huawei's calendar, and is now expected in the first quarter of 2027 instead of its original date. That acceleration suggests Huawei is under pressure — internal or market-driven — to prove it can iterate on AI silicon at Nvidia's pace, not just ship an isolated product every couple of years.

What we actually know: - Chip unveiled: Ascend 960, commercial availability estimated for 2027 - Ascend 960DT: moved up ~3 quarters, now expected Q1 2027 - Stated goal: run any relevant AI model, inside and outside China - Confirmed adoption of the prior generation: DeepSeek plans to deploy at least 160,000 Ascend 950DT chips in a data center in Inner Mongolia - Regulatory backdrop: US export controls limit China's access to top-tier Nvidia GPUs
The real risk isn't just chip performance: even if the Ascend 960 doesn't match the performance-per-watt of Nvidia's latest GPUs, DeepSeek's confirmed adoption of the previous generation (Ascend 950DT) shows China is building a software and model ecosystem that runs natively on non-Nvidia silicon. Over time, that means some models and inference stacks will be optimized first — or exclusively — for Huawei's ecosystem, running in parallel to CUDA's.

What this means for a company that uses or buys AI

For companies in Guatemala and Latin America, this announcement doesn't change infrastructure decisions today — access to the Ascend 960 will remain limited outside China, and its software ecosystem (CANN, instead of CUDA) has far less maturity and third-party support in the region. But it is a signal worth watching over the medium term: if the fragmentation of the AI chip market into two ecosystems (Nvidia/CUDA on one side, Huawei/CANN and other Chinese accelerators on the other) deepens, some open-weight models developed in China could start being optimized for non-Nvidia hardware first. Companies that depend on those models for specific use cases should track which hardware they're trained and officially certified on, rather than assuming everything runs equally well on any GPU.

Takeaway for enterprise AI decision-makers: the "Nvidia with no real competition" narrative is weakening faster than it seemed to be a year ago. That's good news for AI compute buyers over the medium term — more competition usually means better pricing and less single-vendor dependency — though in the short term, ecosystem fragmentation can complicate decisions about porting models between cloud providers.
Carlos Montiel
Enterprise AI Solutions Architect
Specialist in LLMs, Agents, and Orchestration
guatemalia.com/en/#contacto · info@guatemalia.com

Need to implement AI in your company?

Carlos Montiel is an enterprise AI solutions architect. He implements LLMs, Agents, RAG, and orchestrators for companies across Guatemala and Latin America. Get in touch for a consultation.

Contact Carlos Montiel

info@guatemalia.com