Google Advertisement
Beijing, Oktober 1, 2026 – Chinese artificial intelligence firm DeepSeek has joined forces with Huawei to develop open-source programming tools for Huawei’s Ascend processors, marking a significant step in Beijing’s effort to reduce reliance on Nvidia’s dominant CUDA ecosystem.
The partnership, announced on September 30 via DeepSeek’s WeChat account, highlights China’s determination to build a self sufficient AI infrastructure amid tightening U.S. export restrictions.
At the center of the collaboration is TileLang, a new programming language designed to simplify code logic while optimizing performance for Huawei’s Ascend 950 chips.
DeepSeek emphasized that TileLang offers developers a more accessible alternative to Nvidia’s CUDA, which has long been the industry standard for GPU programming.
Google Advertisement
By lowering barriers to entry, the company hopes to accelerate adoption of Huawei’s hardware across China’s AI sector.
The partnership also unveiled the Supernode project, a system that links 128 Ascend 950 chips to function as a single computational unit.
This architecture is intended to deliver high-throughput performance for large scale AI training, positioning Huawei’s chips as a credible substitute for Nvidia’s advanced GPUs, which remain restricted under U.S. export controls.
DeepSeek has further expanded its open-source portfolio with libraries such as DeepGEMM, optimized for matrix multiplication, and DeepEP, designed for large-scale communication tasks.
These tools are tailored to maximize efficiency on Ascend hardware, reinforcing Huawei’s ecosystem as a viable alternative to Nvidia’s tightly integrated software and hardware stack.
The timing of the announcement is critical. Washington’s restrictions have blocked Chinese firms from acquiring Nvidia’s most advanced GPUs, forcing domestic companies to accelerate innovation.
Huawei’s Ascend processors are now positioned as the centerpiece of China’s AI ambitions, with expectations of widespread adoption by 2027.
DeepSeek’s latest AI model, V4 Flash, was specifically optimized for Ascend chips, signaling a deliberate pivot away from Nvidia hardware.
Market watchers note that DeepSeek’s valuation ambitions reportedly targeting $74 billion in its latest funding round are buoyed by the Huawei collaboration.
Yet challenges remain. Ascend chips must prove their reliability and efficiency at scale to win developer trust, while supply chain constraints could hinder rapid deployment.
Still, the partnership underscores a broader geopolitical reality China is racing to build an independent AI ecosystem resilient to U.S. technology sanctions.
If successful, Huawei and DeepSeek could reshape the competitive landscape, eroding Nvidia’s dominance and accelerating the rise of a parallel AI infrastructure rooted in domestic innovation.
