In recent times, the DeepSeek large model product has surged into the spotlight, captivating global attention and significantly lowering the barriers to artificial intelligence (AI) applicationsExperts believe that the success of DeepSeek reflects China's rapid progression in AI innovation, leveraging a well-rounded industrial system, optimized collaboration, and rich application scenarios.

Since its launch in December 2024, DeepSeek has released an open-source language model, DeepSeek-V3, and a reasoning model, DeepSeek-R1. The models gained rapid popularity, topping the charts for free app downloads in U.S. and Chinese app stores due to their low cost, high performance, and open-source natureMajor tech companies and sectors like government, education, and healthcare have jumped on board, integrating DeepSeek models into their operations.

An analysis reveals that training DeepSeek-V3 costs less than one-tenth of OpenAI's GPT-4, while the training cost for DeepSeek-R1 ranges between 3% to 5% of what it costs to train OpenAI's modelAdditionally, the usage costs of DeepSeek-R1 for inference are only one-thirtieth of the costs associated with OpenAI's modelThrough innovative engineering techniques such as algorithm optimization, DeepSeek has effectively enhanced performance at a lower cost, breaking away from the conventional approach of merely increasing computational power and data.

Han Jian, the director of the Information and Software Industry Research Institute at CCID Consulting, views the success of DeepSeek as a significant milestone for China in the large model arena, charting a path of "low cost, high performance" that could set a new global development paradigm for AI.

According to Zhao Gang, the head of the SengZhi Industrial Research Institute, the breakthroughs represented by DeepSeek not only showcase excellence in the competitive arena of large model performance but also solidify China's contributions to the global open-source community

Advertisements

The developments have significantly reduced application costs for large models, marking a new phase in the widespread application of AI technologies in China.

Han Jian further underscores DeepSeek’s contributions to cost reduction in large model applications that stand to empower various industriesThe integration of DeepSeek's open-source approach attracts companies up and down the supply chain, potentially leading to a cascade of innovations across chips, cloud computing, and modeling.

Yang Yuanqing, Chairman and CEO of Lenovo Group, noted that more efficient and lower-cost AI technologies accelerate personal and edge computing capabilitiesThese advances simplify and promote AI deployments within enterprisesLenovo's AI computers, mobile phones, and tablets have all been linked to the cloud-based DeepSeek-R1 model, which is set for local deployment on devices.

The overall capabilities of the entire industrial chain are deemed crucial for competitive advantagesZhao Gang elaborates that China's AI industry stands among the global frontrunners, trailing only the United StatesWith a complete industrial chain, robust innovation capacity, and an abundance of application scenarios, China's AI landscape encompasses a wide spectrum from AI chips to large model algorithms and high-quality datasets.

Han Jian highlighted that in terms of technological innovation, China boasts the largest number of AI patents in the world, accounting for a staggering 61%. Moreover, there are now 302 large models approved for generative AI use in China, with standout products such as DeepSeek emerging from this robust marketplace.

Reports indicate that by 2024, China's intelligent computing power will reach an immense 725.3 quintillion operations per second, reflecting a year-on-year growth of 74.1%—over three times the growth rate of general computing power during the same periodThe financial market for these technologies is projected to hit $19 billion, witnessing an impressive 86.9% boost.

As AI continues to advance new industrialization, institutions like Inspur Cloud rely on their knowledge-enhanced models and multimodal understanding capabilities for solutions in various industries

Advertisements

For instance, they have developed advanced safety monitoring and intelligent customer service applications for Shandong Electric and Electrical GroupSuch AI-driven models significantly increase reaction times to potential hazards, thereby bolstering safety protocols and enrich customer interactions.

According to data from the China Iron and Steel Association, by January this year, an impressive 80% of enterprises had established intelligent control centers, while 81% were exploring the applications of big data models.

The intersection of AI with emerging sectors accelerates the creation of new productive forces, exemplified in the automotive industry where companies like BYD, Geely, Dongfeng, and SAIC are rapidly integrating DeepSeek-R1 into their vehicles to enhance advanced driving technologiesHan Jian emphasizes the urgency for industries to explore and capitalize on the shrinking competitive gap between China and the U.S., particularly regarding reasoning models.

However, Zhao Gang cautions that overall, we remain in a catching-up phase, as notable disparities still exist in AI chips, next-generation model core architectures, and the scientific applications of large models.

DeepSeek's innovations are reinvigorating demands for computational power, driving data center and edge computing expansions as more users adopt its technologiesGroundbreaking algorithmic advancements provided by DeepSeek are poised to revolutionize both China's and the global AI landscapeAlgorithms are emerging as the core engine propelling AI growth, driving developments in computing power and fundamentally reshaping data centers.

The Ministry of Industry and Information Technology plans to spearhead initiatives aimed at enhancing the competitive landscape, aiming to bolster technology-driven studies around computing power, algorithms, and data requirementsThis will also facilitate the establishment of general-purpose and industry-specific large models while pushing for advancements in AI product iterations and the establishment of high-quality data sets.

Han Jian argues for increased support towards innovations in chips, models, architectures, and toolchains, advocating for an AI innovation ecosystem conducive to creativity and advancement

Advertisements

Advertisements

Advertisements