Chinese artificial intelligence company DeepSeek has partnered with Huawei Technologies to develop programming tools designed specifically for Huawei’s Ascend AI chips, marking another step in China’s efforts to strengthen its domestic computing ecosystem and reduce dependence on Nvidia technology. The collaboration focuses on software infrastructure that can help developers make better use of Huawei’s processors as Chinese technology companies face increasing challenges in accessing advanced foreign-made AI hardware.
DeepSeek said on Wednesday that it is open-sourcing programming infrastructure for Huawei’s Ascend platform, including libraries designed to support both computing and communication functions. The move is significant because AI development depends not only on powerful chips but also on the software tools that allow those chips to run complex workloads efficiently. A strong programming ecosystem can make it easier for researchers and developers to build applications without relying heavily on technologies associated with Nvidia’s dominant AI computing platform.
The partnership comes as Chinese technology companies continue to invest in alternatives to Nvidia’s ecosystem. Nvidia has become a major force in the global AI chip market, with its graphics processing units and related software widely used for training and operating advanced AI models. For Chinese firms, however, access to some of the most advanced foreign AI chips has become increasingly complicated because of United States export restrictions.
Against that backdrop, improving the capabilities of Huawei’s Ascend processors has become an important part of China’s broader effort to develop a self-sufficient AI technology stack. Hardware alone is not enough to create a competitive alternative. Developers also need programming languages, compilers, computing libraries, communication systems and other software components that can allow AI models to operate efficiently on domestic processors.
DeepSeek’s contribution centers on programming infrastructure intended to address that challenge. The company has highlighted the role of TileLang, a programming language designed to make it easier to optimize computational workloads for different types of hardware. By improving software compatibility and performance, such tools can help developers adapt sophisticated AI workloads to processors outside Nvidia’s established ecosystem.
The collaboration also involves work on a supernode solution based on 128 Huawei Ascend 950 chips. A supernode connects multiple processors so that they can operate as a coordinated computing system, allowing large AI workloads to be distributed across a much larger pool of computing resources. Such systems are particularly important for training and deploying increasingly sophisticated AI models, which can require enormous amounts of processing power and memory.
The development of a 128-chip Ascend-based system illustrates the growing importance of scaling domestic hardware beyond individual processors. For AI companies, the performance of a single chip is only one part of the equation. The ability to connect large numbers of processors while maintaining efficient communication between them can have a major impact on the speed and cost of AI training.
DeepSeek’s involvement is particularly notable because the company has gained international attention for its approach to AI development. Its emergence has challenged assumptions about the amount of computing resources required to build competitive AI systems and has contributed to a broader discussion about efficiency in model development. Its latest work with Huawei shifts some attention from AI models themselves toward the underlying infrastructure needed to run them.
The software layer may prove just as important as the hardware in determining whether domestic AI chips can compete effectively. Developers accustomed to Nvidia’s CUDA ecosystem benefit from mature tools, extensive libraries and years of accumulated technical knowledge. Creating an alternative therefore requires more than producing a chip with competitive specifications. Developers must also have access to reliable programming frameworks and optimization tools that make it practical to move existing workloads onto another platform.
DeepSeek’s decision to open-source parts of its programming infrastructure could help address this issue. Open-source software allows researchers and developers to examine, modify and improve code rather than depending entirely on proprietary systems. If a broader developer community adopts these tools, improvements could potentially spread across China’s AI industry and contribute to greater compatibility between different domestic hardware and software projects.
Huawei has been steadily developing its Ascend product line as part of its strategy to compete in the AI computing market. The company has invested heavily in processors, data-center technologies and supporting software as Chinese technology firms search for alternatives to imported AI accelerators. DeepSeek’s collaboration adds expertise from an AI model developer to that hardware-focused effort.
The relationship between AI model companies and chip manufacturers is becoming increasingly important worldwide. AI systems are demanding more computing power, while developers are looking for ways to reduce costs and improve performance. Hardware makers, meanwhile, need software ecosystems that encourage developers to use their processors. Cooperation between the two sides can therefore help address problems that neither hardware nor software companies can solve independently.
For China, the issue has an additional strategic dimension. Restrictions on the export of advanced semiconductor technology have increased pressure on domestic companies to develop alternatives. Building a complete AI computing ecosystem inside the country could reduce exposure to external supply disruptions while giving Chinese developers greater control over the technology used to train and operate AI models.
That does not mean domestic alternatives can immediately replace Nvidia across every AI workload. Nvidia’s ecosystem has developed over many years and remains deeply integrated into research institutions, technology companies and data centers around the world. Matching that level of software maturity, developer familiarity and hardware performance is a substantial challenge.
The DeepSeek and Huawei partnership nevertheless demonstrates how quickly the Chinese AI sector is adapting to those pressures. Instead of focusing solely on producing new processors, Chinese companies are increasingly working on the programming infrastructure needed to make domestic chips more useful to developers. This broader approach could become an important part of the country’s long-term semiconductor and AI strategy.
