DeepSeek and Huawei Are Building the Software Side of China’s Chip Push. Here Is Why That Matters
The partnership targets something less visible than a new processor: the tools developers use to program one.
Key Takeaways
- What happened: According to Reuters, DeepSeek said on Wednesday that it has partnered with Huawei to develop programming tools optimised for Huawei’s Ascend chips.
- What is being shared: DeepSeek said it is open-sourcing programming infrastructure for the Ascend platform, including compute and communication libraries.
- The hardware link: The two companies jointly advanced a “supernode” solution based on 128 Ascend 950 chips.
- The language: DeepSeek highlighted TileLang, an open-source programming language for AI chips, as a way to simplify development.
- The bigger picture: Reuters describes this as the latest sign of Chinese tech firms deepening ties while seeking alternatives to the Nvidia ecosystem.
What Was Announced?
Chinese AI firm DeepSeek said in a post on its official WeChat account that it is open-sourcing programming infrastructure for Huawei’s Ascend platform. The release includes compute and communication libraries.
DeepSeek added that Huawei provided full support in building the infrastructure. The two companies also worked together on a supernode solution built around 128 Ascend 950 chips, tuning both the computation and the communication between them.
The announcement arrived about two weeks after Huawei unveiled its next generation of AI processors and supernode computing systems. At that launch, Reuters reports, Huawei said it expects its AI systems to be widely used for model training next year.
Why Software Is the Real Story?
Most coverage of AI chips focuses on hardware: speed, memory and how many chips can be linked together. Developers, however, do not work with the hardware directly. They work through software layers, libraries and programming languages that turn their ideas into instructions the chip can run.
That software layer is where switching costs live. A team that has built its tools and habits around one company’s ecosystem cannot simply move to another chip overnight, even if the new chip is powerful. The code, the libraries and the developer know-how have to move too.
This is why DeepSeek’s choice of words matters. In its statement, DeepSeek said the first priority in building “independent, self-controlled” GPU software ecosystems is a high-level language that is universal, easy to program and still able to reach the hardware’s full performance. It said TileLang was created for that need and offers “a simpler programming model” than Nvidia’s CUDA.
Three Pieces of the Announcement
1. Libraries. Compute and communication libraries are the building blocks developers call on to run large workloads across many chips. Releasing them as open source lets others inspect, reuse and improve them.
2. The supernode. A supernode links many chips so they behave more like a single system. The 128-chip Ascend 950 design is described as jointly advanced by the two companies, with both computation and communication optimised.
3. TileLang. According to DeepSeek, this open-source language improves development efficiency and simplifies code logic. If developers find it easier than the alternatives, it could lower the barrier to writing software for Ascend hardware.
The Timing
Two details stand out from the Reuters report. First, the announcement came just two weeks after Huawei showed its next-generation chips and supernode systems. Second, Huawei has said it expects wide use of its AI systems for model training next year.
Put together, the sequence reads like hardware first, then the software tools to make it usable. Whether that plan succeeds depends on adoption, which is something only time will show.
Our Analysis: What This Could Mean
The following is editorial analysis, not reported fact.
1. Ecosystems are built, not bought. A capable chip is only the start. What decides whether developers move is how easy it is to write, test and scale code. Open-sourcing libraries and backing an open language is a way to invite outside developers in.
2. A high-profile partner helps. DeepSeek is one of the better-known names in Chinese AI. Its involvement gives the Ascend tools visibility and a real-world reference point that a hardware announcement alone would not.
3. Open source cuts both ways. Open code can spread quickly, but it also has to be maintained. The value of this release will depend on whether developers actually use and improve it.
4. Claims need proof. DeepSeek says TileLang offers a simpler model than CUDA, and that the tools help performance. Independent benchmarks and developer feedback will be the real test.
What to Watch Next?
- Whether other Chinese AI firms adopt the Ascend tools and TileLang.
- How the 128-chip Ascend 950 supernode performs in real training workloads.
- Whether Huawei’s expectation of wide model-training use next year holds up.
- Independent comparisons between TileLang and existing programming approaches.
Frequently Asked Questions
What did DeepSeek announce?
- According to Reuters, DeepSeek said it has partnered with Huawei to develop programming tools optimised for Huawei’s Ascend chips, and that it is open-sourcing programming infrastructure including compute and communication libraries.
What are Huawei Ascend chips?
- Ascend is Huawei’s line of AI processors. The announcement refers to the Ascend platform and to a supernode solution built on 128 Ascend 950 chips.
What is TileLang?
- DeepSeek describes it as a high-level, open-source programming language for AI chips that improves development efficiency and simplifies code logic.
How does this relate to Nvidia?
- Reuters describes the partnership as a sign of Chinese tech firms seeking alternatives to the Nvidia ecosystem. DeepSeek also said TileLang offers a simpler programming model than Nvidia’s CUDA.
When did Huawei unveil its new AI chips?
- Reuters says Huawei unveiled its next generation of AI processors and supernode computing systems about two weeks before this announcement.
What does Huawei expect next year?
- Huawei said it expects its AI systems to be widely used for model training next year.
What is a supernode?
- It is a computing system that links many chips so they work together more like a single machine. In this case it is based on 128 Ascend 950 chips.
The DeepSeek and Huawei partnership is less about a new chip and more about the tools that make a chip useful. By open-sourcing libraries and promoting a simpler programming language, the two companies are targeting the part of the AI stack where switching is hardest. Whether developers follow is the question to watch.
