Article | Smart Relativity

AI's widespread adoption across industries has now become a tangible reality. According to incomplete statistics, over 200 enterprises from diverse sectors such as technology, internet, cloud services, telecommunications, finance, energy, automobiles, mobile phones, and government agencies nationwide have accessed DeepSeek.

Currently, everyone is enthused by DeepSeek's innovative prowess and powerful performance. However, from the perspective of "Smart Relativity," it is equally noteworthy how China's AI industry has matured – what industrial foundation and technological ecosystem have successfully borne the immense traffic generated by DeepSeek in such a short span and effectively supported various industries in swiftly adopting DeepSeek to accelerate their digital and intelligent transformation?

Compared to global AI development, China's AI industry is transitioning from "following" to "competing," a journey inseparable from the collaborative efforts and joint development of the industrial ecosystem.

Recently, CCTV's Xinwen Lianbo highlighted technological innovation, featuring key projects like Tianyi Cloud's Xirang Intelligent Computing Platform and Xirang All-in-One Machine, sending a strong signal – the national cloud centered around Xirang is a cornerstone for China's AI development. China's AI advancement necessitates not only breakthroughs in domestic models but also robust domestic cloud services and computing power to form an industrial closed loop of "domestic models + domestic cloud services + domestic computing power," thereby fostering a collective industry boom.

Behind DeepSeek's viral success, China's AI industry has undergone a profound evolution.

With DeepSeek's popularity, China's AI industry has attained a new level of maturity.

When over 200 enterprises adopt DeepSeek within a short period, it's no longer just a question of DeepSeek's strength but rather a testament to the advanced integration capabilities and collaborative consciousness of the current domestic AI industry.

Initially, as DeepSeek gained traction, a surge of users encountered "server busy" messages. Subsequently, domestic cloud service providers, led by Tianyi Cloud, integrated the DeepSeek series of models. Supported by powerful computing power, this issue was swiftly resolved, enabling industry-level full-scene deployment and application.

Take Tianyi Cloud's Xirang Intelligent Computing All-in-One Machine - DeepSeek Edition as an example; it integrates domestic computing power, models, and cloud services, deeply embedding the DeepSeek-R1/V3 series of large models. This achieves full-stack localization from chips to inference engines to model services, ensuring autonomous control of the technology chain while enabling out-of-the-box and flexible deployment, catering to a wide range of needs.

Perhaps, it is precisely due to such robust support from domestic computing power and cloud services that the number of enterprises accessing DeepSeek continues to grow, with access speeds constantly being refreshed. This phenomenon fundamentally reflects the overall maturity of China's AI industry.

If DeepSeek, as a domestic model, lacked the necessary support of domestic computing power and cloud services, it would be challenging to replicate the current scene of various industries adopting DeepSeek to harness AI. Here, the trinity of domestic models + domestic computing power + domestic cloud services, forming a complete industrial closed loop, is the fundamental driving force behind everything.

China's AI industry now boasts its own "three pillars"

Certainly, the maturity of China's AI industry isn't solely about having domestic models, computing power, and cloud services. More importantly, the positioning and value of these three in the industrial chain have become increasingly prominent, collaboratively propelling China's AI towards leapfrog development. Just as the core three components of traditional cars – the engine, chassis, and transmission – determine the vehicle's performance, these three pillars underpin China's AI industry.

Post DeepSeek's popularity, why is China's AI industry's performance so "explosive"? In the entire industrial chain, the establishment of these three core components is crucial, determining the "performance" and industrial competitiveness of the rapid industry outbreak.

1. Domestic models are akin to the "engine" of the local AI industry. With DeepSeek as an exemplar, they provide robust power through innovative breakthroughs.

Models are the cognitive hub of AI systems, and their advancements directly shape the boundaries and performance of industrial applications. This resembles a traditional car's "engine," which releases kinetic energy by burning fuel (akin to data in AI) to propel the car. DeepSeek's innovative breakthroughs clearly demonstrate the value of models as "engines."

Through technologies like algorithm optimization, DeepSeek reduces its reliance on high-end computing power, training high-performance models using relatively low-powered NVIDIA H800 GPU clusters, thereby breaking the computing power bottleneck. In tasks like mathematics, coding, and natural language reasoning, DeepSeek-R1 performs on par with OpenAI GPT-4's official version, achieving an internationally advanced level in key metrics such as reasoning ability.

DeepSeek signifies that domestic AI has achieved parity with foreign counterparts in models. This combination of equally powerful performance and low cost allows domestic AI to swiftly take off in the industrial sector, akin to equipping the AI industry with a durable and reliable domestic engine.

2. Domestic cloud services are akin to the "chassis" of the local AI industry. With Tianyi Cloud as an example, they ensure stable operation and support industrial collaboration.

Cloud services serve as the physical backbone for implementing AI technology, and their stability directly impacts the continuity and capacity of the industry. This resembles a traditional car's "chassis," which supports the entire car's stability and coordinates steering, braking, and driving stability to ensure control safety. As a national cloud representative, Tianyi Cloud plays a similar role in China's AI industry development.

Currently, Tianyi Cloud has established an integrated capability encompassing "cloud, network, edge, terminal, data, intelligence, and security," achieving full-stack technology localization from cloud operating systems to distributed databases, with over 2,000 domestic mutual certifications. Simultaneously, it has built a national "2+3+7+X" public intelligent computing cloud pool, deploying two liquid-cooled intelligent computing centers, each with ten thousand cards, in Beijing and Shanghai, and covering over 280 cities and 1,000 edge nodes through the "one city, one pool" approach to meet diverse computing power needs.

These constructions and capabilities have endowed China's AI industry with a robust and stable domestic chassis. With Tianyi Cloud's support, significant progress has been made in the compatibility and adaptation between various domestic chips and application software systems during the information security substitution process. Currently, Tianyi Cloud supports "one cloud and six chips" in the XC field, covering domestic mainstream CPUs including Kunpeng, Haiguang, Phytium, Loongson, Zhaoxin, and Sunway.

3. Domestic computing power is akin to the "transmission" of the local AI industry. With the "Xirang" intelligent computing platform as an example, it releases diverse computing power on demand.

Computing power serves as the energy system for AI development, and its scheduling flexibility governs the efficiency and upper limit of technology commercialization. This resembles a traditional car's "transmission," which alters the transmission ratio between the engine and wheels to achieve power transmission, speed regulation, and even enhance fuel efficiency. The Xirang intelligent computing platform shares a similar positioning and value, constructing a national integrated computing power system that meets the diverse computing power needs of the AI industry's development, ensuring computing power is truly utilized.

Currently, through self-developed AI frameworks, 3D parallel acceleration, and other technologies, the Xirang intelligent computing platform has elevated the comprehensive computing efficiency of domestic computing power to an industry-comparable level, thereby supporting the complete lifecycle from basic model training to industry application deployment. Recent market feedback indicates that Xirang has successfully achieved full-stack localization of the DeepSeek-R1/V3 series of large models, from chips to inference engines to model services, effectively supporting privatized deployment and out-of-the-box usage, significantly lowering the threshold for enterprise AI applications.

In essence, the Xirang intelligent computing platform plays a pivotal role in connecting, scheduling, and utilizing computing power. This is the core of the national cloud in the AI era and the foundation for constructing a domestic AI application ecosystem of "domestic cloud + domestic models + domestic computing power."

Under the standardized landing paradigm, the national cloud centered around "Xirang" shines brightly in the AI era

Today, China's AI industry's surge is the result of the combined action of the core "three pillars." For a long time, domestic cloud services, represented by Tianyi Cloud, have served as the "chassis" providing foundational support; domestic computing power, represented by the Xirang intelligent computing platform, realizes diverse computing power supply under the "transmission's" influence.

Before DeepSeek's emergence, what China's AI industry lacked was a sufficiently powerful and durable "engine." Now that this crucial component is in place, the entire AI industry has a relatively complete standard paradigm for implementation, much like the combination of a traditional car's three core components.

Specifically, on Tianyi Cloud, the Xirang intelligent computing platform focuses on hosting DeepSeek's full series of large models. Enterprise users can swiftly invoke DeepSeek models through standardized APIs, while also supporting model training and fine-tuning. The synergy of these three components has effectively promoted the full-scene implementation and development of domestic AI.

Furthermore, considering that not every scenario or enterprise is willing or needs to operate on the cloud, the "three pillars"-centered landing paradigm can be adjusted and simplified, supported by "domestic computing power + domestic models" to provide more localized deployment and application options for various industries. For instance, Tianyi Cloud's fully domestic Xirang intelligent computing all-in-one machine, equipped with the DeepSeek-R1 full-blooded large model and the Xirang integrated computing power platform, also enables enterprise customers to enjoy robust computing power and powerful AI performance, catering to a wide range of needs.

From this perspective, post DeepSeek's popularity, the fundamental reason behind the rapid outbreak of domestic AI is a testament to the maturity of the industrial ecosystem. On the stage of the new era, those who are well-prepared have already begun to shine.

*All images in this article are sourced from the internet