Chinese AI lab DeepSeek has partnered with Huawei to develop chip programming tools, a move aimed at reducing reliance on Nvidia's hardware and software ecosystem, whbl.com reports.

The partnership addresses one of the most stubborn bottlenecks in China's push for AI self-sufficiency: not chip fabrication itself, but the software layer that lets developers write code for those chips. Nvidia's dominance in AI compute rests as much on CUDA, its mature programming platform, as on its GPUs. Any challenger silicon needs a comparable toolchain before developers will adopt it at scale.

Huawei has its own AI accelerator family — the Ascend series — which the company positions as a domestic alternative to Nvidia's data center GPUs. Ascend chips already ship inside Chinese cloud and enterprise systems, and Huawei has built its own compute platform, CANN, around them. What Ascend has historically lacked is the breadth of developer tooling, framework support and optimization maturity that CUDA has accumulated over nearly two decades.

DeepSeek brings a different kind of asset to the table. The lab has demonstrated, through its R1 and earlier models, an ability to train and run frontier-class AI systems under severe hardware constraints — a capability born of necessity, since US export controls restrict Chinese firms' access to Nvidia's top-tier accelerators. DeepSeek's engineering team has direct, hard-won experience in extracting performance from non-Nvidia silicon and in porting large-model training and inference workloads onto domestic accelerators.

That experience is precisely what a programming-tool effort needs. Building compilers, kernel libraries, debugging and profiling tools for AI accelerators requires deep collaboration with the model developers who will consume them, because the workloads define the performance targets. A joint effort between the company making the chips and the lab running some of the most closely watched training workloads in China shortens that feedback loop considerably.

The commercial logic tracks the geopolitical one. Washington's escalating export controls have progressively cut off Chinese AI developers from Nvidia's most advanced GPUs, and Beijing has responded by channeling state support toward domestic compute. US restrictions on Nvidia's China-specific products have already cost the company significant revenue in that market, and each tightening of the rules has pushed Chinese buyers further toward Huawei and other domestic suppliers. Software, rather than hardware, is the last major holdout of Nvidia's ecosystem advantage in China.

For Huawei, a credible, developer-friendly toolchain built with input from a top AI lab could expand Ascend's addressable market beyond state-backed and hyperscale customers to a broader base of Chinese AI developers. For DeepSeek, native tooling reduces the engineering overhead of adapting every new model generation to domestic hardware — an overhead that currently consumes scarce specialist time.

Nvidia has not commented on the partnership, and neither company has disclosed the financial terms, a timeline for tool availability, or which specific product lines the collaboration will cover. Whether the resulting tools reach general release or remain internal infrastructure is not yet clear.

If the partnership produces tools that meaningfully lower the cost of developing on Ascend silicon, it would tighten a competitive squeeze that Nvidia already faces in its largest single market — and strengthen the case that China's AI compute stack is becoming functionally independent, one software layer at a time.