Reported on February 13, as DeepSeek continues to break through the circle and “ignite” the AI industry chain, cloud service providers, computing power companies, etc. are the first to access, and AI application and other industry chain companies are competing to use it.
More than 700 days after ChatGPT came out, when Silicon Valley was still spending hundreds of billions of dollars on GPU clusters, Chinese manufacturers used 5.576 million US dollars to prove that the competition of AI large models may not only rely on scale, but more importantly, on actual results.
AI-related applications may explode
DeepSeek considers algorithm accuracy and system efficiency. In addition to extreme underlying optimization, collaborative optimization may also allow large models to reach or even exceed the performance of NVIDIA GPUs when using domestic computing power.
The powerful breakthrough under the blockade of computing power is due to the key breakthroughs in DeepSeek’s technical architecture, data strategy, and engineering practice. From the top-level product application to the underlying infrastructure, each level of the big model has formed a very good ecosystem, and there is room for optimization at each level.
In addition to low cost, the openness of R1 brings disruptive changes. Through its application program interface (API), researchers can use the model at a fraction of the cost of a proprietary model, or query it for free through the online chatbot DeepThink.
From the perspective of artificial intelligence research, DeepSeek demonstrates a way to improve countless other models. It represents a major advance in global AI by making model building cheaper, faster and more accessible.
Tianfeng Securities said that the rapid development of DeepSeek has had a positive impact on computing infrastructure companies, especially its efficient and low-cost AI model training method has reduced dependence on high-end imported chips and enhanced the confidence of domestic companies in independently developing computing chips.
Great Wall Securities said that with the recent rise of DeepSeek-R1, further verification of algorithm optimization capabilities can also give birth to powerful models. At the same time, with DeepSeek’s powerful distillation capabilities and low-cost open source advantages, local deployment has gradually become possible.
With the explosion of DeepSeek, its technical capabilities have been quickly introduced into various fields. Many experts interviewed said that 2025 may usher in the turning point of AI application landing, and a number of companies will benefit first, and the industry will enter the explosion period of AI application.
Opening a new pattern of AI competition among giants
OpenAI
On February 10, OpenAI announced that its first self-developed AI chip will be designed in the first half of the year and handed over to TSMC for wafer production, aiming for mass production in 2026. It is reported that OpenAI’s self-developed chip plan is regarded as “taking away the firewood from the bottom of the pot” for Nvidia. Its first chip uses TSMC’s 3nm process, and its performance is comparable to Nvidia H100, but the cost is reduced by 30%.
OpenAI’s strategy is not only aimed at reducing dependence on Nvidia GPUs, but also trying to reshape industry rules by mastering chip pricing power. Sources revealed that if OpenAI chips are successfully mass-produced, the company’s engineering team plans to continue to optimize processor performance in subsequent iterations.
NVIDIA (NVDA)
In the global AI industry, NVIDIA’s CUDA technology is almost an irreplaceable standard. From deep learning training to the deduction of various artificial intelligence algorithms, CUDA is the basic tool that countless programmers work silently behind the scenes. Moreover, with the rapid development of AI, the demand for computing power continues to rise. Only through CUDA can the potential of GPU be maximized.
The DeepSeek team’s approach is to bypass CUDA, directly call the PTX instruction set, fine-tune GPU hardware resources, and perform efficient calculations in this “primitive” way. Although NVIDIA’s CUDA currently has a strong market share worldwide, the global AI ecosystem may gradually become diversified and no longer rely solely on a certain company or technology. Therefore, NVIDIA may also begin to re-examine the breakthrough in underlying technology.
Meta (META)
In 2025, the global AI chip market will usher in a historic turning point. With the rapid development of generative AI models, companies like Meta have almost endless demand for chips. Meta has also accelerated its self-developed chip layout, and Meta has announced that it will invest $60 billion in upgrading AI infrastructure in the next year. Meta CEO Zuckerberg commented that DeepSeek’s technical strength and performance are impressive.
Microsoft (MSFT )
Microsoft has announced that it will invest a total of $100 billion in AI technology and data center construction in 2025, further increasing from last year’s record spending. In addition, Microsoft CEO Satya Nadella said at the World Economic Forum in Davos, Switzerland: “DeepSeek’s new model is impressive, especially in terms of how to efficiently develop open source models and perform inference calculations. Its computing efficiency is excellent.
Tencent (TCEHY )
It is learned that Tencent Cloud recently announced that the high-performance application service HAI supports one-click deployment of the DeepSeek-R1 large model, and developers can access and call it in just 3 minutes. Tencent Cloud Platform launched the “Developer Gift Pack”: DeepSeek’s full range of models can be deployed with one click, from the “full blood version” (671B parameters) to the lightweight (1.5B) can be selected, and R1 and R1-1.5B models can be provided for a limited time free online experience.
WIMI Hologram (WIMI ）
In the context of fierce industry competition, computing power has become more efficient. Data shows that AI listed company WIMI Hologram has been closely following the latest technological trends in the industry, redefining the generative AI track and large model paradigm with technological singularity, and actively developing a comprehensive and in-depth computing power layout in the face of new opportunities and challenges brought by DeepSeek.
From the hardware level, WIMI Hologram has increased its investment in high-performance computing equipment. In order to better adapt to the operation of DeepSeek-related models, it has built a computing power cluster through large-scale cluster deployment to meet the company’s internal needs for DeepSeek technology research and development, model training and optimization, improve computing power utilization, and ensure efficient and stable operation when processing complex AI tasks.
Conclusion
The emergence of DeepSeek not only allows Chinese AI companies to emerge on the global stage, but this breakthrough may also trigger profound changes in the global AI ecosystem. From technological confidence to ecological layout, the rise of DeepSeek has gone beyond the commercial scope. Of course, more importantly, DeepSeek has also triggered new thinking about the development of AI technology. As long as there are innovative thinking and technology, companies in any country can make breakthroughs in the field of AI.
