Huawei’s Ascend 910C: A Bold Challenge to NVIDIA in the AI Chip Market

The artificial intelligence (AI) chip market is growing rapidly, driven by increased demand for processors that can handle complex AI tasks. The need for specialized AI accelerators has increased as AI applications such as machine learning, deep learning and neural networks evolve.

NVIDIA has been the dominant player in this domain for years, with its powerful Graphics Processing Units (GPUs) become the standard for AI computing worldwide. However, Huawei has emerged as a powerful competitor with its Ascend series, challenging NVIDIA’s market dominance, especially in China. The Rise 910Cthe latest in the line, promises competitive performance, energy efficiency and strategic integration within Huawei’s ecosystem, potentially reshaping the dynamics of the AI ​​chip market.

Background information about Huawei’s Ascend series

Huawei’s entry into the AI ​​chip market is part of a broader strategy to establish a self-sustaining ecosystem for AI solutions. The Ascend series started with the Ascend 310, designed for edge computing, and the Ascend 910, aimed at high-performance data centers. Launched in 2019, the Ascend 910 was recognized as the world’s most powerful AI processor and delivers 256 teraflops (TFLOPS) of FP16 performance.

Built on Huawei’s proprietary technology Da Vinci architectureAscend 910 offers scalable and flexible computing capabilities suitable for various AI workloads. The chip’s emphasis on balancing power and energy efficiency laid the foundation for future developments, leading to the improved Ascend 910B and the latest Ascend 910C.

The Ascend series is also part of Huawei’s efforts to reduce dependence on foreign technology, especially in light of US trade restrictions. By developing its own AI chips, Huawei is working towards a self-sustaining AI ecosystem, offering solutions ranging from cloud computing to on-premise AI clusters. This strategy has gained traction among many Chinese companies, especially as local companies have been encouraged to limit dependence on foreign technology such as NVIDIA’s H20. This has created an opportunity for Huawei to position its Ascend chips as a viable alternative in the AI ​​space.

The Ascend 910C: features and specifications

The Ascend 910C is designed to provide high computing power, energy efficiency and versatility, positioning it as a strong competitor to NVIDIA’s A100 and H100 GPUs. It delivers up to 320 TFLOPS of FP16 performance and 64 TFLOPS of INT8 performance, making it suitable for a wide range of AI tasks, including training and inference.

The Ascend 910C delivers high computing power and consumes approximately 310 watts. The chip is designed for flexibility and scalability, allowing it to handle various AI workloads such as Natural Language Processing (NLP), computer vision and predictive analytics. Additionally, the Ascend 910C supports high-bandwidth memory (HBM2e), essential for managing large data sets and training complex AI models efficiently. The chip’s software compatibility, including support for Huawei’s MindSpore AI framework and other platforms such as TensorFlow and PyTorch, makes it easier for developers to integrate into existing ecosystems without significant reconfiguration.

Huawei vs NVIDIA: the battle for AI supremacy

NVIDIA has long been a leader in AI computing, with its GPUs serving as the standard for machine learning and deep learning tasks. The A100 and H100 GPUs, built on the Ampere and Hopper architectures respectively, are currently the benchmarks for AI processing. The A100 can deliver up to 312 TFLOPS of FP16 performance, while the H100 offers even more robust capabilities. NVIDIA’s CUDA platform has advanced significantly, creating a software ecosystem that simplifies the development, training, and deployment of AI models.

Despite NVIDIA’s dominance, Huawei’s Ascend 910C aims to provide a competitive alternative, especially in the Chinese market. The Ascend 910C performs similarly to the A100, with slightly better energy efficiency. Huawei’s aggressive pricing strategy makes the Ascend 910C a more affordable solution, delivering cost savings for companies looking to scale their AI infrastructure.

However, the software ecosystem remains a crucial area of ​​competition. NVIDIA’s CUDA is widely adopted and has a mature ecosystem, while Huawei’s MindSpore framework is still growing. Huawei’s efforts to promote MindSpore, especially within its ecosystem, are key to convincing developers to switch from NVIDIA’s tools. Despite this challenge, Huawei has made progress by working with Chinese companies to create a cohesive software environment that supports the Ascend chips.

Reports indicate that Huawei has started distributing prototypes of the Ascend 910C to major Chinese companies, including ByteDance, Baidu and China Mobile. This early involvement indicates strong market interest, especially among companies looking to reduce dependence on foreign technology. Since last year, Huawei’s Ascend solutions have been used to train almost half of China’s top executives 70 major language modelsdemonstrating the processor’s impact and widespread adoption.

The timing of the Ascend 910C’s launch is significant. With US export restrictions limiting access to advanced chips like NVIDIA’s H100 in China, domestic companies are looking for alternatives, and Huawei is stepping in to fill the void. Huawei’s Ascend 910B has already gained ground for AI model training in various industries, and the geopolitical environment is driving further adoption of the newer 910C.

While NVIDIA is expected to ship 1 million H20 GPUs to China, which is expected to generate approximately $12 billion in revenue, which Huawei’s Ascend 910C is expected to generate $2 billion in sale this year. Additionally, companies that adopt Huawei’s AI chips can become more integrated into Huawei’s broader ecosystem, increasing dependence on its hardware and software solutions. However, this strategy can also raise concerns among companies about becoming too dependent on one supplier.

Strategic partnerships and alliances

Huawei has entered into strategic partnerships to drive adoption of the Ascend 910C. Collaborations with major technology players such as Baidu, ByteDance and Tencent have facilitated the integration of Ascend chips into cloud services and data centers, making Huawei’s chips part of scalable AI solutions. Telecom operators, including China Mobile, have integrated Huawei’s AI chips into their networks, supporting edge computing applications and real-time AI processing.

These alliances ensure that Huawei’s chips are standalone products and integral parts of broader AI solutions, making them more attractive to businesses. Furthermore, this strategic approach allows Huawei to promote its MindSpore framework and build an ecosystem that can rival NVIDIA’s CUDA platform over time.

Geopolitical factors have significantly influenced Huawei’s strategy. With US restrictions limiting access to advanced semiconductor components, Huawei has increased its investments in R&D and collaborations with domestic chipmakers. This focus on building a self-sustaining supply chain is critical to Huawei’s long-term strategy, ensuring resilience against external disruptions and helping the company innovate without relying on foreign technologies.

Technical lead and future prospects

The Ascend 910C has gained fame thanks to its strong performance, energy efficiency and integration into Huawei’s ecosystem. It competes closely with NVIDIA’s A100 in several key performance areas. For tasks that require FP16 calculations, such as deep learning model training, the chip’s architecture is optimized for high efficiency, resulting in lower operating costs for large-scale use.

However, challenging NVIDIA’s dominance is no easy task. NVIDIA has built a loyal user base over the years because the CUDA ecosystem offers extensive development support. For Huawei to gain more market share, it must match NVIDIA’s performance and offer ease of use and reliable developer support.

The AI ​​chip industry is likely to continue to evolve, with technologies such as quantum computing and edge AI reshaping the domain. Huawei has ambitious plans for its Ascend series, with future models promising even better integration, performance and support for advanced AI applications. By continuing to invest in research and form strategic partnerships, Huawei aims to strengthen its foundations in the AI ​​chip market.

The bottom line

In conclusion, Huawei’s Ascend 910C poses a significant challenge to NVIDIA’s dominance in the AI ​​chip market, especially in China. The 910C’s competitive performance, energy efficiency, and integration within Huawei’s ecosystem make it a strong contender for companies looking to scale their AI infrastructure.

However, Huawei faces significant hurdles, especially in competing with NVIDIA’s established CUDA platform. The success of the Ascend 910C will depend heavily on Huawei’s ability to develop a robust software ecosystem and strengthen its strategic partnerships to solidify its position in the evolving AI chip industry.