DeepSeek and Huawei Partner to Build Open-Source CUDA Alternative for Ascend AI Chips

InfraTop News3 sources·Sep 30

chips-and-hardware compute-infrastructure geopolitics-and-export-controls deepseek huawei tilelang ascend

Summary

Updated Sep 30 — Nvidia's CEO has publicly warned the DeepSeek-Huawei software partnership represents a significant threat to US AI technological leverage.

• DeepSeek and Huawei jointly develop open-source programming tools for Ascend AI chips
• TileLang launched as open-source high-level CUDA alternative targeting full Ascend chip performance
• Joint 128-chip Ascend 950 supernode solution optimises AI training compute and communications
• Partnership aims to build China's independent, self-controlled GPU software ecosystem

Adjust signal

Updates

Sep 30

Type	Key Point	Context
Market Impact	Nvidia CEO Jensen Huang warns DeepSeek-Huawei shift threatens US AI leverage	Nvidia CEO Jensen Huang publicly stated that DeepSeek shifting to Huawei hardware and software represents a significant risk to U.S. technological leverage in AI — the first explicit acknowledgment from Nvidia's top executive of the competitive threat posed by the partnership (Bloomberg).
Infrastructure	DeepSeek plans 160,000 Ascend 950DT chip deployment in Inner Mongolia data center	DeepSeek planned large-scale deployment of at least 160,000 Huawei Ascend 950DT chips at a new Inner Mongolia data center, demonstrating infrastructure-scale commitment to the Ascend ecosystem beyond just software tooling.
Strategy	DeepSeek also developing its own AI inference chip beyond Huawei partnership	Reuters reported in July 2026 that DeepSeek is developing its own inference chip, extending its hardware independence strategy beyond the Huawei partnership and signaling a broader ambition to control its full compute stack.
Industry Update	New York Times coverage signals mainstream US business press attention	The New York Times published a September 30, 2026 report framing the DeepSeek-Huawei effort as directly targeting 'a key source of Nvidia's AI dominance,' marking the story's entrance into mainstream US business media and amplifying investor awareness.

Market Impact

Nvidia CEO Jensen Huang warns DeepSeek-Huawei shift threatens US AI leverage

Nvidia CEO Jensen Huang publicly stated that DeepSeek shifting to Huawei hardware and software represents a significant risk to U.S. technological leverage in AI — the first explicit acknowledgment from Nvidia's top executive of the competitive threat posed by the partnership (Bloomberg).

Infrastructure

DeepSeek plans 160,000 Ascend 950DT chip deployment in Inner Mongolia data center

DeepSeek planned large-scale deployment of at least 160,000 Huawei Ascend 950DT chips at a new Inner Mongolia data center, demonstrating infrastructure-scale commitment to the Ascend ecosystem beyond just software tooling.

Strategy

DeepSeek also developing its own AI inference chip beyond Huawei partnership

Reuters reported in July 2026 that DeepSeek is developing its own inference chip, extending its hardware independence strategy beyond the Huawei partnership and signaling a broader ambition to control its full compute stack.

Industry Update

New York Times coverage signals mainstream US business press attention

The New York Times published a September 30, 2026 report framing the DeepSeek-Huawei effort as directly targeting 'a key source of Nvidia's AI dominance,' marking the story's entrance into mainstream US business media and amplifying investor awareness.

Details

Type	Key Point	Context
Partnership	DeepSeek–Huawei joint open-source Ascend infrastructure	DeepSeek announced a partnership with Huawei to develop and open-source programming infrastructure for Ascend AI chips, including compute and communication libraries, announced via DeepSeek's WeChat account.
New Tech	TileLang: open-source CUDA alternative for AI chips	TileLang is a high-level open-source programming language for AI chips offering a simpler programming model than NVIDIA's CUDA while aiming to reach the hardware's full performance potential on Ascend.
Tech Info	128-chip Ascend 950 supernode jointly developed	DeepSeek and Huawei jointly advanced a supernode solution based on 128 Ascend 950 chips, optimising both computation and communication performance for AI training workloads at scale.
Context	Announced two weeks after Huawei's next-gen chip reveal	The partnership came two weeks after Huawei unveiled its next-generation AI processors and supernode computing systems, which Huawei expects to be widely used for model training in 2027.
Strategy	Goal: independent self-controlled GPU software ecosystem	DeepSeek explicitly stated the goal is to build a 'new generation of independent, self-controlled GPU software ecosystems,' with TileLang as the foundational universal high-level language.
Industry Update	Deepening Chinese AI firm ties as Nvidia alternative sought	Reuters described this as the latest sign of Chinese tech firms deepening ties as they seek alternatives to the Nvidia ecosystem, amid US export controls limiting Nvidia chip access in China.

Partnership

DeepSeek–Huawei joint open-source Ascend infrastructure

DeepSeek announced a partnership with Huawei to develop and open-source programming infrastructure for Ascend AI chips, including compute and communication libraries, announced via DeepSeek's WeChat account.

New Tech

TileLang: open-source CUDA alternative for AI chips

TileLang is a high-level open-source programming language for AI chips offering a simpler programming model than NVIDIA's CUDA while aiming to reach the hardware's full performance potential on Ascend.

Tech Info

128-chip Ascend 950 supernode jointly developed

DeepSeek and Huawei jointly advanced a supernode solution based on 128 Ascend 950 chips, optimising both computation and communication performance for AI training workloads at scale.

Context

Announced two weeks after Huawei's next-gen chip reveal

The partnership came two weeks after Huawei unveiled its next-generation AI processors and supernode computing systems, which Huawei expects to be widely used for model training in 2027.

Strategy

Goal: independent self-controlled GPU software ecosystem

DeepSeek explicitly stated the goal is to build a 'new generation of independent, self-controlled GPU software ecosystems,' with TileLang as the foundational universal high-level language.

Industry Update

Deepening Chinese AI firm ties as Nvidia alternative sought

Reuters described this as the latest sign of Chinese tech firms deepening ties as they seek alternatives to the Nvidia ecosystem, amid US export controls limiting Nvidia chip access in China.

Partnership details, technology specs, and strategic context for the DeepSeek–Huawei open-source AI chip programming stack.

What This Means

DeepSeek and Huawei's TileLang partnership is a concrete step in China's strategy to build a domestic AI chip software stack independent of NVIDIA's CUDA ecosystem. With US export controls blocking access to top Nvidia hardware, China's leading AI research lab and its foremost chip maker are jointly creating the programming infrastructure needed for Chinese developers to fully exploit Ascend hardware. If TileLang gains adoption, it could erode one of NVIDIA's most defensible competitive moats — the software ecosystem lock-in — and accelerate the viability of Ascend chips for serious AI training workloads at scale.

Sources

Update history (1)

Sep 30Nvidia's CEO has publicly warned the DeepSeek-Huawei software partnership represents a significant threat to US AI technological leverage.Added NYT/Bloomberg coverage; appended 4 new rows covering Jensen Huang's warning, 160,000-chip Inner Mongolia deployment, DeepSeek's own chip development, and mainstream US press attention.

Similar Events

DeepSeek Plans to Deploy 160,000+ Huawei Ascend 950DT Chips at Inner Mongolia Data Center

Sep 4

Huawei Report Claims DeepSeek V4 Post-Training on Ascend Chips at 34% MFU — Experts Remain Skeptical on Pre-Training

Jul 24