
DeepSeek Emerges as a Strong Competitor to NVIDIA in the AI Field with Its New Model

On September 29, 2025, Chinese AI company DeepSeek announced a new AI model that surprised those closely following the technology. This new model, called DeepSeek-V3.2-Exp, is optimized with Huawei's Ascend accelerators and CANN software. This development is seen as a step that particularly threatens NVIDIA's market share.
DeepSeek, which has attracted attention in recent years with its innovations in the field of artificial intelligence, has begun to rapidly rise to prominence with the models it has developed. The DeepSeek-V3 and R1 models, in particular, attracted attention with their high performance and affordable prices. Now, with DeepSeek-V3.2-Exp, it is further strengthening its claim in this field.
DeepSeek-V3.2-Exp: Next-Generation Artificial Intelligence Model
DeepSeek's new model is equipped with a technology called “sparse attention.” This technology reduces processing power and memory requirements, enabling the model to operate more efficiently. In addition, the model has been optimized for Huawei's Ascend accelerators and CANN software, significantly reducing its dependence on NVIDIA's CUDA ecosystem.
The launch of the new model reinforces DeepSeek's strong position in the artificial intelligence market. Particularly following the release of DeepSeek's R1 model, NVIDIA's market value dropped by approximately $590 billion in a single day. This highlights DeepSeek's impact on the market.
The Advantages of DeepSeek: A Successful AI Player
There are several key advantages behind DeepSeek's rapid rise. Here are some of them:
1. High Performance, Low Cost
DeepSeek develops models that provide high processing power while keeping costs low. This allows the company to stand out from its competitors. Users can obtain this technology at much more affordable prices while still enjoying high performance.
2. NVIDIA CUDA Independence
DeepSeek significantly reduces dependence on NVIDIA's CUDA ecosystem by focusing on Huawei's Ascend accelerators and CANN software. This allows DeepSeek to be compatible not only with its own platform but also with other hardware. This gives DeepSeek great flexibility in the technology world.
3. Innovative Technologies: Sparse Attention
The DeepSeek-V3.2-Exp model uses “sparse attention” technology to reduce memory and processing power requirements. This innovative approach enables faster and more efficient processing. Such technologies allow DeepSeek to differentiate itself from other players in the industry.
4. Open Source Support and Developer Community
DeepSeek shares its models as open source to offer researchers and developers greater flexibility. This fosters faster innovation processes and creates a broad developer community.
5. Global Competitiveness
Although based in China, DeepSeek stands out as a highly competitive player in the global market. In terms of both price-performance ratio and efficiency, DeepSeek poses a major threat to large technology companies.
6. Unique Optimizations
DeepSeek not only performs software and hardware optimizations, but also integrates with advanced platforms such as Huawei Ascend, taking industry standards one step further.
Security and Geopolitical Concerns
DeepSeek's rapid rise has caused significant concerns not only in the tech world but also in the areas of security and geopolitics. In the US in particular, decisions have been made to ban the use of DeepSeek applications on government networks and devices. New York Governor Kathy Hochul explained that she imposed this ban due to concerns that DeepSeek could potentially collect user data and steal technology secrets.
DeepSeek Launches a New Era in the World of Artificial Intelligence
With its new model, DeepSeek-V3.2-Exp, DeepSeek appears to have taken a major step forward not only technologically but also strategically. This model could seriously shake NVIDIA's dominance in the market. As DeepSeek's influence in this field grows day by day, competition in the world of artificial intelligence is certain to heat up.



