This Spring Festival, DeepSeek has become hugely popular, with the open-sourcing of its V3 and R1 versions and many others, making advanced AI models no longer a black technology that can only be accessed by overseas giants and a few talents. The high-end AI models that were once inaccessible to the general public have been made available to everyone by DeepSeek, becoming AI technology for the masses.
The industry-wide intelligence based on DeepSeek cannot be separated from the convenient supply of massive computing power, and advanced computing power service platforms such as the National Supercomputing Interconnect have enabled the combination of "advanced models + powerful computing power", and have promoted the emergence of more AI application innovations.
Recently, the National Supercomputing Interconnect platform has quickly launched the full-version DeepSeek model series, including R1, and provided fast deployment API calls and AI web application services, helping enterprises and developers flexibly build customized AI applications.
The integration of advanced algorithm models and pervasive super-intelligent computing power is turning AI into a basic infrastructure like water and electricity. A new technology movement that accelerates the implementation of pervasive AI applications is unfolding across the country.
In the era of pervasive DeepSeek AI, we need to understand, explore, and fully embrace it. The first step to understanding pervasive AI is to understand why powerful models and super-intelligent integrated computing power were once seen as inaccessible to the general public.
Without the intelligence of thousands of industries, there would be no intelligence for the whole of China. The so-called AI pervasiveness means that all industries can access advanced AI algorithms and integrated computing power at a high capacity and low cost.
This requires the descent of two basic elements of AI:
One is the descent of advanced models.
Previously, models like OpenAI's o1/o3 reasoning models were the high-end flowers in the algorithm field, with high subscription fees, making it difficult for ordinary people to use them, and they were not open to domestic services. Since the beginning of 2025, DeepSeek has broken through the monopoly of advanced reasoning models overseas through technological innovations such as Mixture of Experts (MoE) and Multi-Head Latent Attention (MLA), as well as innovations in reinforcement learning training methods, significantly reducing the cost of model training. DeepSeek R1 has fully open-sourced its algorithm architecture and weights, breaking the monopoly of advanced reasoning models overseas and supporting low-cost access to high-performance AI for small and medium-sized enterprises.
The other is the descent of super-intelligent integrated computing power resources.
While DeepSeek's performance is high, the computing power required for single-model training has decreased. However, the introduction of DeepSeek to upgrade the intelligence of thousands of industries will lead to a surge in the total number of AI models and applications. As Jevons Paradox says, when the efficiency of using a certain resource is improved, the total consumption of that resource may actually increase rather than decrease.
Productivity tools like DeepSeek, whose efficiency is enhanced by technological innovation, will greatly increase the demand for computing power when widely applied. Looking at the current domestic computing power landscape, the price of GPU-based AI-dedicated computing power is expensive and resources are scarce, unable to support the computing power demand for the intelligence of thousands of industries.
At this time, the integrated computing power cluster dominated by supercomputers can provide powerful computing power guarantees to meet the computing power demand for large-model training, reasoning, and large-scale industry applications.
Therefore, reducing the computing power shortage and cost burden for various industries to introduce AI, the descent and pervasiveness of super-intelligent integrated computing power is imperative. But new challenges have emerged.
For a long time, supercomputers have been seen as the high-end flowers in the computing field, and it is not easy for ordinary enterprises and developers to use them. The challenges mainly come from several aspects:
One is the capability, using multi-heterogeneous integrated computing power and matching it with AI tasks to fully unleash performance requires certain R&D and engineering capabilities, which are beyond the capabilities of many enterprises and developers.
Two is the efficiency, the large-scale, high-concurrency, and stable and reliable demands of AI tasks are highly dependent on the coordination of multi-cluster and multi-center. This involves comprehensive capabilities such as cross-center resource coordination, inter-computing communication, and inter-network interconnection, which fully test the efficient operation and management capabilities of the computing power platform.
Three is the service, from AI development and operation to commercial realization, the links are many and the process is long, supporting the development of agents, vertical models, applications, and other needs, requiring a complete tool chain and ecological support, which requires the platform to transform from a "seller of shovels" to a service provider, which is a transformation that many traditional data centers have not yet made.
The prosperity of industry intelligence cannot be achieved without both advanced algorithms and supercomputing power. Relying on the massive resources of the national supercomputing interconnect and its distributed supercomputing and intelligent computing centers has become a must-have. DeepSeek has realized the pervasiveness of models, but who can remove the high-end flower of supercomputers for the industry? The National Supercomputing Interconnect platform has stepped in.
At the critical moment when DeepSeek was on fire and servers were frequently crashing, the National Supercomputing Interconnect platform stepped in and released a series of models including DeepSeek-R1, V3, Coder, and Chatbot services on February 3. The number of clicks on the first day exceeded one million.
The combination of the national-level computing power platform and Deepseek has given the public hope to use the "full-version" R1. Sure enough, the National Supercomputing Interconnect platform has not let everyone down, and has currently launched a 671B-parameter DeepSeek model for enterprise development use.
What supports the National Supercomputing Interconnect to handle millions of visits?
It turns out that before the DeepSeek explosion, the National Supercomputing Interconnect had already cultivated a fertile soil of industry-usable supercomputing through meticulous efforts.
On the supply side, in terms of the basic computing power situation.
The Supercomputing Interconnect has abundant resources, connecting more than 20 supercomputing and intelligent computing centers in 14 provinces across the country, and this national-level comprehensive computing power service platform is also the fertile soil for computing power.
Supported by the abundant, secure, and efficient supercomputing interconnect computing power foundation, users can now quickly access the platform and interact smoothly with Deepseek through the web and mobile terminals.
On the application side, in terms of the usage threshold for enterprises.
The Supercomputing Interconnect has previously accumulated rich AI ecosystem and deployment experience, with more than 6,600 commodities on the platform, including nearly 240 AI model services, both domestic open-source models like DeepSeek and Qwen, and foreign AI open-source models like Llama, Stable Diffusion, and Gemma. The rich model matrix and lightweight tools have made the Supercomputing Interconnect a more efficient AI platform. The platform also opens an AI community for developers to conduct development, debugging, and knowledge sharing.
With the multi-dimensional empowerment of models, tools, and communities, ordinary people can also quickly experience the capabilities of various AI large models on the platform, and easily master the deployment, optimization, and calling of models, progressing from AI users to AI developers, embracing the infinite opportunities of the intelligent era.
At the ecosystem level, in terms of the subsequent series of service guarantees.
The professional service team and ecosystem support plan of the Supercomputing Interconnect platform further promote the public's participation in AI technology innovation and also help the growth and development of AI startups like DeepSeek.
On the one hand, the professional team of the Supercomputing Interconnect can quickly respond to the personalized container service needs of enterprises, allowing enterprise customers to focus more on their own business development without having to spend energy on trivial matters such as AI training environment construction.
In addition, through the "AI Ecosystem Partner Acceleration Program", enterprise users can also obtain 3 months of free use of the DeepSeek API interface, a pool of millions of card-hours of computing power resources, and other incentives, empowering more enterprises to achieve AI innovation, and are expected to further accelerate the popularization of AI in more industry scenarios.
It can be said that the Supercomputing Interconnect has completed the comprehensive penetration from computing power supply to enterprise applications and ecosystem support, providing full-chain, full-dimensional protection for AI to enter thousands of industries, and taking the pervasive AI to the end.
The TRON fire is raging, but without the continuous access and supply of universal computing power, the resource constraints and busy access will quickly consume the enthusiasm of the public and enterprises to use AI and explore AI. Without AI innovation in thousands of industries, the breakthrough of domestic AI can only be limited to the developer community and cannot be converted into commercial opportunities, and domestic AI technology cannot enter the virtuous cycle from R&D to commercialization.
From this perspective, the supply of computing power is a necessary condition for China to seize the innovation opportunity of DeepSeek.
The National Supercomputer TRON and DeepSeek are walking hand in hand, advancing side by side, promoting the descent of diverse and massive computing power, and filling the crucial computing power link for inclusive AI.
The flowers of inclusive AI on the algorithm and supercomputing foundation are blooming in this spring, which will open the prelude to the acceleration of industrial intelligence.
When inclusive AI flows into thousands of industries, intelligent China is full of vitality.
