Beijing, September 30, 2026: Chinese artificial intelligence company DeepSeek and technology giant Huawei are expanding their collaboration around AI computing, with DeepSeek announcing a new set of open-source software components designed for Huawei's Ascend AI chips.
The move focuses on one of the most important parts of modern artificial intelligence infrastructure: the software that allows AI models to efficiently use specialized processors.
DeepSeek said it is open-sourcing infrastructure for the Huawei Ascend platform, including the TileLang programming system as well as computing and distributed communication libraries.
The development could help developers build and optimize AI workloads for Huawei processors while giving China's growing AI ecosystem more software tools that are not dependent on Nvidia's CUDA platform.
DeepSeek Opens TileLang for Huawei Ascend
One of the central components in the announcement is TileLang, a high-level programming language and compiler technology designed for developing high-performance AI computing kernels.
DeepSeek describes TileLang as a simpler programming approach that can make it easier for developers to write code for advanced AI hardware while still allowing them to access the underlying processor's performance.
The Ascend version is designed to work with Huawei's Ascend C programming environment and related hardware capabilities.
The objective is to give AI developers a higher-level way to program specialized computing operations without having to manually handle every low-level hardware detail.
Why AI Software Is as Important as AI Chips
Building a powerful AI processor is only part of the challenge.
AI developers also need software frameworks, compilers, mathematical libraries and communication systems that allow models to efficiently use the available hardware.
This is one reason Nvidia's CUDA ecosystem has become so important to the global AI industry.
CUDA provides developers with tools that allow them to build and optimize applications for Nvidia GPUs.
DeepSeek's latest work with Huawei is aimed at developing a broader software environment around China's own AI accelerator hardware.
If successful, such an ecosystem could make it easier for developers to move AI applications between different computing platforms.
Several AI Libraries Are Being Open-Sourced
DeepSeek's announcement goes beyond TileLang.
The company is also releasing several computing and communication components designed for AI workloads on the Ascend platform.
These include DeepGEMM for accelerating general matrix operations and DeepEP for large-scale communication between computing devices.
Other components include TileKernels, FlashMLA and DeepSelect.
These tools address different parts of the AI computing pipeline, including vector calculations, memory operations, sparse attention and data selection.
Together, the components are intended to provide developers with reusable building blocks for creating and optimizing AI systems.
128 Ascend 950 Chips in a Supernode
DeepSeek and Huawei have also been working on a 128-chip supernode configuration based on Huawei's Ascend 950 processors.
The companies have collaborated on both computing and communication optimization for the system.
A supernode combines multiple AI processors into a tightly integrated computing system so that they can work together on demanding workloads.
This type of architecture is particularly important for training and running large AI models, which can require enormous amounts of computing power and communication between processors.
DeepSeek's Software Could Help Huawei Hardware
Huawei has been developing its Ascend processor family as part of its broader AI computing infrastructure strategy.
The company has also been investing heavily in software tools designed to help developers create AI applications for its hardware.
DeepSeek's involvement adds another important layer because the AI company has experience optimizing large language models and other advanced AI workloads.
The collaboration therefore connects AI model development with the underlying hardware and software infrastructure needed to run those models.
DeepSeek Has Already Adapted Its AI Models for Huawei Chips
The latest announcement follows earlier work by DeepSeek to adapt its AI models for Huawei's Ascend architecture.
DeepSeek's V4 model was introduced with support for Huawei's Ascend processors, demonstrating the company's continuing effort to make its AI systems work efficiently on domestic Chinese hardware.
The expansion of software support could make this process easier for future AI models and for other developers using the same hardware.
Instead of optimizing each application independently from scratch, developers can potentially use common open-source libraries and programming tools.
TileLang Is Designed to Simplify AI Kernel Development
AI models rely on thousands of highly optimized computational operations known as kernels.
These operations perform tasks such as matrix multiplication, vector calculations and attention processing.
Efficient kernels are critical because even small improvements in the performance of individual operations can significantly affect the speed and cost of training or running a large AI model.
TileLang is designed to simplify the development of these kernels by providing a higher-level programming model.
The project already supports multiple hardware environments, and its Ascend development expands that ecosystem to Huawei's AI processors.
Open Source Could Accelerate Adoption
Making the software available as open source could encourage researchers, developers and companies to test the technology and contribute improvements.
Open-source projects can grow quickly when a large developer community begins creating extensions, optimization techniques and integrations around them.
For AI hardware manufacturers, software availability is particularly important because developers are more likely to adopt a processor when they already have familiar tools and libraries available.
This means the competition between AI computing platforms is increasingly taking place at both the hardware and software levels.
The Growing Importance of Alternative AI Platforms
The global AI industry has become heavily dependent on specialized accelerators, particularly GPUs and other high-performance processors.
As demand for AI computing continues to rise, technology companies are developing alternative hardware platforms to address performance, supply and cost requirements.
Huawei's Ascend family is one of the major Chinese efforts in this area.
DeepSeek's decision to contribute software infrastructure to the platform could help create a larger ecosystem around those processors.
AI Infrastructure Is Becoming More Diverse
The AI industry is increasingly moving away from the idea that one type of processor or software environment will handle every workload.
Different processors can be optimized for different tasks, including model training, inference, data processing and specialized AI applications.
Software layers that make it easier to program different accelerators could therefore become increasingly valuable.
For developers, portability can reduce the amount of work required to adapt an AI application to another hardware platform.
What This Means for AI Developers
For AI researchers and software engineers, the new open-source components provide additional tools for experimenting with Huawei's Ascend architecture.
Developers can potentially use the libraries to optimize mathematical operations, communication between processors and AI attention mechanisms.
The open-source approach also gives researchers an opportunity to inspect and modify the software rather than relying entirely on proprietary development environments.
As the ecosystem matures, more AI frameworks and applications could potentially gain compatibility with Ascend processors.
Huawei's AI Hardware Roadmap Continues
Huawei has continued expanding its Ascend processor roadmap and AI infrastructure products.
The company has also been developing larger SuperPoD and SuperCluster systems designed to combine large numbers of AI processors.
Huawei says its newer Ascend generations are being developed alongside improvements in memory bandwidth, interconnect technology and software support.
These developments indicate that the company's strategy is not limited to producing individual AI chips. It is building a broader computing infrastructure around those processors.
Why the Software Ecosystem Matters
The history of computing shows that hardware performance alone does not guarantee widespread adoption.
Developers need programming languages, libraries, debugging tools, documentation and development environments that make hardware practical to use.
This is particularly important for AI because modern models are extremely complex and require extensive optimization.
DeepSeek's latest open-source release therefore represents more than a collection of software libraries. It is an attempt to strengthen the software layer surrounding an alternative AI computing platform.
Conclusion
DeepSeek and Huawei are expanding their cooperation with a new open-source software push for Huawei's Ascend AI processors.
The release includes TileLang along with computing and distributed communication libraries designed to improve the development and performance of AI workloads.
The two companies are also working on a 128-chip Ascend 950 supernode, showing that their collaboration extends from software optimization to large-scale AI computing infrastructure.
As artificial intelligence continues to require increasingly powerful computing systems, the competition is no longer only about who makes the fastest processor.
The software ecosystem surrounding those processors may be equally important.
DeepSeek's latest move could therefore become an important development in the evolution of China's AI computing ecosystem and in the wider global competition to build alternative platforms for next-generation artificial intelligence.
Journalist: Vijay Singh