The Computing Power Network is currently witnessing a significant surge in popularity.

In April this year, a meeting of the Political Bureau of the CPC Central Committee integrated the Computing Power Network into the 'Six Networks' construction framework. Since then, discussions about the Computing Power Network have intensified across both the industrial sector and mainstream media.

Recently, CCTV reported on the 'Smart IP Wide Area Network (AI WAN) Application Promotion Initiative and China Tour,' a collaborative effort by the China Academy of Information and Communications Technology and telecom operators. The report emphasized that the Smart Computing Power Network, akin to an electrical grid dispatch center, can uniformly manage decentralized computing resources. This allows enterprises to access computing power as conveniently, securely, and stably as they do electricity. For instance, in scenarios involving cross-regional calls spanning hundreds of kilometers, computational performance loss can be kept within 5%. Additionally, the network has evolved from a mere 'data transmission pipeline' into an intelligent infrastructure capable of perceiving AI services, implementing intelligent scheduling, and providing differentiated guarantees, all while possessing automatic risk identification and handling capabilities.

Given the rising popularity of the Computing Power Network, it's easy to understand why AI WAN, which underpins the interconnection and efficient scheduling of computing resources, has come into the limelight.

What truly deserves attention is why the current focus is not on a specific single-point technological innovation but rather on the 'AI WAN Application Promotion Initiative.' What industrial signals does this move convey?

From 'Transmitting Data' to 'Scheduling Computing Power': The Rise of AI WAN

In recent years, China has continuously advanced the construction of a national integrated computing power network. Policies such as the 'Action Plan for High-Quality Development of Computing Power Infrastructure' and the 'Opinions on Deeply Implementing the 'AI Plus' Action' have been introduced, explicitly advocating for cross-regional scheduling of computing resources, the construction of computing-network convergence infrastructure, and the enhancement of computing power network transmission capabilities. In April this year, a meeting of the Political Bureau of the CPC Central Committee proposed strengthening the construction of water networks, new-type power grids, computing power networks, next-generation communication networks, and underground urban pipe networks, thereby integrating the computing power network into the 'Six Networks' construction system. This further underscores its critical role as key infrastructure in the intelligent era and signals an accelerated push to build the computing power network.

These policies indicate a unified trend: computing power is becoming a new type of productive force in the intelligent era. To fully unleash its potential, it is crucial to expedite the construction of a network capable of connecting, scheduling, and collaborating computing power, enabling it to flow as freely as electricity.

Behind these policies lie the genuine needs of industrial development. What new scenarios and businesses are driving the evolution of the computing power network?

First, there is the emergence of intelligent computing scenarios. With the rapid development of artificial intelligence, an increasing number of enterprises are leveraging external intelligent computing centers through 'computing power access.' Compared to building their own intelligent computing centers, renting computing power involves lower investment, faster deployment, and greater flexibility in responding to technological iterations and business growth demands. Consequently, massive amounts of data need to flow efficiently between enterprises and intelligent computing centers, driving the rapid rise of new models such as 'data access to computing,' 'separated storage and computing,' 'remote training,' and 'distributed inference.'

At the same time, as the scale of large-scale model training continues to expand, a single intelligent computing center, constrained by factors such as data center space and power resources, can no longer independently support the ever-growing demand for computing power. More and more scenarios require connecting computing resources distributed across different regions and nodes to form 'multi-point collaborative computing,' enabling cross-intelligent computing center collaborative training and inference. These new scenarios necessitate a wide area network to facilitate cross-domain flow and collaboration of data and computing power.

Second, there are new AI-driven businesses. The rapid proliferation of new terminals such as AI phones, AI PCs, and companion robots, along with the continuous emergence of innovative applications like cloud gaming, cloud fitness, industrial large models, AI assistants, and intelligent manufacturing robots, is rapidly integrating AI into various production and living scenarios. These businesses often require the network to upload multimodal data to the cloud or edge nodes for real-time processing and swiftly return inference results, imposing unprecedented demands on the network's real-time interaction capabilities.

So, can traditional networks still meet the demands of these new scenarios and businesses?

Traditional WAN networks primarily cater to internet businesses such as video, web browsing, and social media, with the core objective of enabling data connectivity and transmission. However, the network demands of the AI era have fundamentally changed. Whether it's massive data access to computing, cross-intelligent computing center collaborative training, or new scenarios like cloud-edge collaborative distributed inference and separated storage and computing, they all essentially require efficient collaboration of data and computing power on a larger scale. This imposes stringent requirements on the network for wide-area losslessness, ultra-high throughput, ultra-low latency, high security, and reliability. Simultaneously, new applications such as AI phones, digital humans, and intelligent robots not only demand higher basic capabilities from the network in terms of bandwidth, latency, and stability but also require the network to possess intelligent service identification, real-time experience perception, and differentiated guarantee capabilities to ensure that inference results are delivered to users quickly and stably.

In short, traditional WANs address the issue of 'whether data can be delivered,' whereas AI WANs tackle the challenges of 'whether computing power can be efficiently collaborated, whether data can flow securely, and whether business experience can be guaranteed.' It is precisely this shift in demand that renders traditional WANs inadequate in terms of transmission efficiency, resource scheduling, and business guarantees to meet the needs of AI business development. This, in turn, drives the accelerated upgrade and evolution of network architecture and technology systems, giving rise to AI WANs.

From Technological Exploration to Application Implementation: AI WAN Enters a New Phase

Driven by both policy and industrial demand, the industry proposed the industrial direction of Smart IP Wide Area Networks (AI WANs) as early as 2025, aiming to address the new challenges faced by traditional wide area networks and construct a new-generation network foundation for the intelligent era.

Over the past two years, AI WAN has gradually transitioned from technological exploration to industrial practice. At the third 'IPv6 Technology Application Innovation Competition' held last year, the competition introduced, for the first time, a dedicated AI WAN track, which showcased a batch of benchmark practical cases from operators, enterprises, and industry partners across the country. These cases covered multiple core scenarios, including government and enterprise intelligent computing training and inference, industry data circulation, and public user experience guarantees. They not only validated the technical feasibility of AI WAN but also demonstrated its application value in real-world business environments.

For Government and Enterprises: Enabling Efficient Flow of Computing Power

Faced with new scenarios such as enterprise computing power access, distributed training, and separated storage and computing, companies like Beijing Telecom, Zhejiang Telecom, and Hebei Unicom have undertaken extensive innovative practices based on AI WAN. By integrating technologies such as IPv6+, wide-area RDMA lossless transmission, SRv6, and AI-powered intelligent scheduling, they have successfully overcome challenges faced by traditional wide-area networks in long-distance computing power collaboration, including link packet loss, uneven load distribution, latency fluctuations, and difficulties in meeting differentiated SLA guarantees. This has enabled efficient computing power interconnection across hundreds or even thousands of kilometers.

Practice has shown that AI WAN is transforming the way enterprises access and utilize computing power. On the one hand, by constructing a wide-area lossless, highly reliable, and schedulable computing power network, enterprises can access external intelligent computing resources on demand, achieving elastic expansion of computing power and cross-domain collaboration. On the other hand, through innovative models such as hierarchical model deployment and cloud-edge collaboration, raw data can remain within its domain, with only training parameters or intermediate results being transmitted. This approach fully leverages external computing power while also ensuring data security and privacy protection, meeting the stringent compliance requirements of industries such as healthcare, finance, and government affairs.

Currently, these innovative achievements are serving fields including government affairs, healthcare, education, transportation, and manufacturing. In healthcare scenarios, Beijing Telecom has extended hospital training and inference tasks to an intelligent computing center 240 kilometers away, achieving less than 5% loss in computing efficiency without data leaving its domain. In Zhejiang, the AI WAN lossless network has supported multiple industries in conducting distributed training and cloud-edge collaborative inference, with overall computing efficiency exceeding 95%. In the Beijing-Tianjin-Hebei region, the creation of a 'computing power corridor' has enabled unified scheduling and efficient collaboration of cross-regional computing resources, allowing enterprises to access computing power resources as conveniently as water and electricity.

For Industries: Enabling Secure Data Circulation

High-quality data circulation is a crucial foundation for driving continuous performance improvements in artificial intelligence applications. However, cross-regional and cross-entity data circulation often faces challenges such as low collaboration efficiency, difficulty in data sharing, and high security and compliance risks, making it difficult to meet the demand for real-time, trustworthy circulation of massive, high-quality data required for AI training.

To address this pain point, Tianjin Mobile has innovatively adopted an AI WAN data circulation network architecture featuring four-dimensional collaboration among 'network-computing-data-security,' establishing an elastic, deterministic, and secure data transmission channel. This effectively resolves the efficiency and security challenges in cross-domain data circulation, providing robust support for the efficient flow and value realization of data elements.

Currently, this solution has been implemented and validated in scenarios such as automotive consumption and financial services, creating significant value. In the automotive consumption sector, by securely integrating human and vehicle data and breaking down data barriers, the accuracy of federated learning models has been improved by 34%, helping partner automakers increase lead conversion rates by 34%. In the financial services sector, by enabling secure integration and trustworthy circulation of government and enterprise data, it effectively supports innovative inclusive financial services, improving enterprise compliance review efficiency by 70%.

For the Public: Bringing AI Services into Every Household

The era of AI WAN applications is not only evident on the production side but is also accelerating its reach to the consumer side. With the rise of new 'weak-terminal, strong-cloud' family AI services such as cloud computers, cloud phones, cloud gaming, and cloud fitness, a significant amount of computing tasks are shifting from terminals to the cloud or edge, and business experience increasingly relies on network quality. Issues like latency fluctuations, network jitter, or insufficient bandwidth can easily cause problems such as lag, disconnections, and unsmooth visuals, directly impacting user experience.

To address this issue, operators such as Guangdong Unicom and Shaanxi Mobile have innovatively constructed a 'terminal-network-cloud-computing' collaboration system based on AI WAN. By introducing technologies such as SRv6, AI service identification, and intelligent scheduling, they have achieved collaborative linkage between family terminals, networks, and cloud computing resources, creating dedicated cloud access channels and deterministic experience guarantee capabilities for services like cloud computers, cloud gaming, and cloud fitness. The network can not only accurately identify high-value AI services but also dynamically optimize transmission paths based on business demands, delivering low-latency, low-jitter, and highly reliable service experiences.

Practice has shown that with the support of the deterministic experience network built by AI WAN, the operational response speed and smoothness of applications like cloud computers and cloud gaming have significantly improved, delivering a nearly 'zero-lag' user experience. Currently, this solution has served countless family users and expanded to new scenarios such as AI somatosensory games, intelligent medical consultations, and cloud fitness, driving the popularization of AI-driven new services from a niche to the mainstream.

A Promising New Foundation for the Intelligent Era

In hindsight, whether it involves the integrated computing-network services that facilitate cross-domain training and inference for governments and enterprises, the data circulation networks that promote the efficient flow of data elements, or the family AI services that introduce cloud computing, cloud gaming, and cloud fitness into every home, these vivid examples all underscore a common reality: AI Wide Area Networks (AI WANs) are tackling several critical bottlenecks in the large-scale development of artificial intelligence.

On the one hand, AI WANs enable the efficient collaboration of computing resources scattered across different locations, significantly lowering the barriers for enterprises to access computing power and propelling the gradual transition of 'Computing Power as a Service' from a conceptual idea to a tangible reality. On the other hand, while unlocking the value of computing power, AI WANs also ensure data security and the efficient flow of data elements, empowering more government and enterprise clients to confidently and securely harness computing power. Concurrently, AI WANs continue to spur new application innovations. From industrial large-scale models, intelligent manufacturing, and medical diagnostics to cloud gaming, digital humans, AI fitness, and companion robots, an increasing number of innovative businesses that were previously hindered by computing power and network capabilities are now accelerating their deployment.

These cases not only confirm the technical feasibility and commercial viability of AI WANs but also provide the industry with a development path that is referential, replicable, and scalable for widespread adoption. This is precisely the significance of launching the 'AI WAN Application Promotion Initiative'—to expedite the transition of AI WANs from 'sporadic innovations' to 'large-scale applications' based on existing practical achievements.

Looking forward, as the application promotion initiative continues to gain momentum, collaboration and synergy among upstream and downstream players in the industrial chain will further strengthen, and a larger industrial ecosystem is accelerating its formation. AI WANs will extend to more industries and business scenarios, giving rise to richer and more valuable AI services and continuously unleashing the innovative potential of 'AI Plus.' By then, a smart IP wide area network that 'understands computing power, comprehends businesses, and masters scheduling' will truly become the new cornerstone of the intelligent era.

As AI WANs accelerate into the stage of large-scale application, their development has opened up a new perspective for observation. As an industry leader, the upcoming Mobile World Congress (MWC) will showcase the latest achievements and practical advancements in AI WAN-related fields. What new breakthroughs will it bring at this industry event, and how will it further drive industrial evolution? The answers are eagerly anticipated.