The global AI computing infrastructure is currently entering a critical period of scaled deployment. Industry resources and attention are widely skewed toward chip performance, computing cluster scale, and large model algorithm iteration. However, few notice that the information highway supporting cross-regional flow of computing power and enabling compute-network synergy—the wide area network (WAN) foundation—is becoming an implicit bottleneck constraining the release of computing efficiency.
Recently, the 12th Ultra-Broadband Forum (UBBF) was held in Vienna, Austria. As an annual bellwether for the carrier network sector, this summit specifically established the Net5.5G Intelligent IP Network Summit. You may also be curious: What new changes are occurring in the WAN era of AI, and what is the progress of its commercialization?
According to Cai Jun, Vice President of Huawei's Data Communication Product Line, more than 60 operators worldwide have deployed AI WAN solutions, covering multiple mainstream markets, which validates the solution's universality and commercial maturity. Cai Jun stated, "The scaled commercial deployment of Huawei's next-generation AI WAN solution marks the transition of the AI WAN industry from the technology verification stage to full-scale implementation."
Behind this announcement lie many questions worth examining: What bottlenecks do traditional WANs face? Is Huawei's AI WAN upgrade a routine technical iteration or a fundamental paradigm shift? And what changes will it bring to the industrial landscape of compute-network convergence?
1. AI Forces WAN Evolution, Traditional Architectures Cannot Support Intelligent Computing Demands
Currently, AI has become a mandatory question for operators. This year, over 45% of operators' AI strategies have shifted toward leveraging AI to achieve new growth. While AI brings immense business opportunities to operators, its technical characteristics are breaching the capability boundaries of traditional WANs, making WAN generational upgrades an inevitable trend for industry development.
First, inherent pain points of traditional WANs. Traditional WANs rely on best-effort transmission logic, depending on manual configuration, troubleshooting, and optimization. A single network change cycle typically takes days, and response speeds struggle to match business changes. Moreover, security relies primarily on perimeter defense, often leaving operators passive and lagging behind complex penetration attacks.
Second, new essential demands of AI intelligent computing. As AI businesses penetrate core production scenarios, their requirements for networks undergo fundamental changes. For instance, scenarios such as cross-regional computing scheduling, remote model inference, and distributed training require latency jitter and packet loss rates to improve by an order of magnitude, hitting the performance ceiling of traditional WANs.
Third, pressure on operator transformation. Operators are currently upgrading from traditional bandwidth-selling pipe businesses to value-based businesses selling computing services and security-enhanced services. However, traditional WANs only provide basic connectivity capabilities, lacking operational abilities for computing scheduling, efficiency assurance, and security enhancement, thus failing to support new business growth curves.
2. Huawei AI WAN's Three Major Updates: From Technology Verification to Scaled Commercial Use
How can these pain points be resolved? In his speech, Cai Jun stated: "Huawei continuously enhances product and solution capabilities around three dimensions—three-dimensional security, lossless efficiency, and autonomous networking—to build a next-generation IP bearer network, stimulating new growth for operators in the AI era." Furthermore, all capabilities of Huawei's next-generation AI WAN solution have completed commercial verification.
First, three-dimensional security with multi-layered defense embeds security into the network foundation. According to Cai Jun, Huawei AI WAN constructs a complete three-dimensional security system across the device layer, link layer, and service layer, breaking through the limitations of traditional perimeter defense.
At the device layer, Huawei pioneered the RDR solution. Based on independent Threat Protection Processing Units (TPU) and multimodal behavioral expert algorithms, the threat detection rate exceeds 95%, achieving minute-level tracing of Advanced Persistent Threats (APTs). On the data plane, Huawei's intelligent flow identification algorithms reduce overall DDoS risks by 90%. Notably, TPUs have been deployed at scale; at one operator, they successfully defended against324 attacks from 8 countries within just over three months of launch. At the link layer, Huawei launched the industry's first endogenous QKD solution, achieving three-fiber integration, saving 60% in network construction investment, and breaking the 80km distance limit. The author learned that Huawei recently collaborated with a Macau operator to release the world's first quantum-safe target network, significantly enhancing dedicated line security value-added services.
At the service layer, Huawei AI WAN provides "1+4" deep defense, reducing boundary DDoS response time from 1 minute to under 5 seconds, achieving a 100% ransomware detection rate, and shortening CT-SOC threat closed-loop disposal time to under 30 minutes.
Second, lossless computing efficiency enables compute-network synergy, achieving zero packet loss over thousand-kilometer distances. As AI accelerates integration into enterprise core businesses, intelligent computing services are highly sensitive to network packet loss. Huawei achieves lossless wide-area computing bearing through three-layer upgrades in architecture, scheduling, and transmission.
At the architecture level, Huawei's minimalist two-layer multi-plane architecture increases network scale by 4 times and reduces hardware usage by 30%, significantly lowering construction costs. At the scheduling level, NPLB (Network Path Load Balancing) achieves 98% link utilization, improving training and inference performance by over 7%. At the transmission level, the AI WAN Starlink lossless algorithm achieves over 96% computing efficiency for long-distance links, ensuring zero packet loss over thousand kilometers, enabling operators to offer enterprises remote computing experiences comparable to local ones, and accelerating ToB new service monetization.
According to Cai Jun, a Chinese operator deployed this solution to support hospital cloud medical large models, saving 30% on computing costs and increasing dedicated line ARPU by 5 times. The results were significant.
Finally, autonomous networking achieves L4-level autonomy in four core scenarios. Huawei builds a "one map, one brain" autonomous network architecture, achieving L4-level capabilities in fault management, network changes, quality optimization, and security management.
For example, network changes achieve a full-process closed loop from configuration verification and online simulation to configuration comparison, reducing human errors. Simultaneously, security management achieves a minute-level closed loop from intrusion detection and attack path tracing to one-click blocking, improving response speed.
What is the commercial progress of this capability? As far as the author knows, Huawei has already helped multiple operators achieve L4-level network autonomy, with one South African operator's IP network becoming the world's first case to reach this level.
3. Core Value of Scaled Commercial Use: Reconstructing the Tripartite Industrial Value System
In the author's view, the scaled deployment of AI WAN brings not just product capability upgrades but also a reconstruction of the entire industry's value logic, reshaping the value system from three perspectives: operators, government and enterprise clients, and the whole industry.
For operators, this is a key lever to transform from pipe providers to computing service providers, opening up a second growth curve. On the cost reduction side, the minimalist architecture combined with L4-level autonomous capabilities significantly lowers network construction and O&M costs. On the revenue increase side, value-added services such as lossless computing dedicated lines and quantum-safe dedicated lines allow operators to boost dedicated line ARPU and escape homogenized price wars.
For government and enterprise clients, they gain access to high-reliability, low-cost wide-area computing services. In terms of experience, remote computing calls approach local user experiences, and computing resources can be reused across regions on demand. Regarding cost and compliance, improved computing efficiency directly reduces overall computing costs, while full-link three-dimensional security capabilities meet the requirements of high-compliance scenarios such as finance, government affairs, and healthcare.
For the entire industry, this announcement establishes the commercial paradigm for AI-native WANs, validating the feasibility of AI WANs transitioning from technical concepts to scaled commercial use. It propels the WAN industry into the new era of AIWAN, providing a replicable network transformation path for the AI era for global operators and accelerating the overall pace of compute-network convergence.
Author's Observation: Beyond Solution Upgrades, A Fundamental Reconstruction of Industrial Logic
Looking beyond the announcement itself, from the perspective of industrial competition and long-term evolution, the author believes there are three deeper industry signals worth noting.
First, the core of AI WAN is AI-defined computing services, serving computing scheduling, efficiency assurance, and security enhancement. It is a completely new product paradigm for the intelligent computing era, not merely a patch-up of old solutions.
Second, security will shift from a cost item to a profit item. Previously, network security was a cost input for operators. Now, capabilities like three-dimensional security and quantum security can be packaged as value-added dedicated line products, generating direct premium revenue. By deeply embedding security capabilities into the network foundation in advance, Huawei is essentially positioning itself for the operators' next profit growth point.
Third, in the second half of compute-network competition, the main battlefield shifts to the WAN. Over the past two years, the industry focused intensely on computing density and chip performance within data centers, with the center of gravity for computing construction inside nodes. Going forward, cross-regional, cross-node wide-area computing scheduling capabilities will become the core factor determining the experience of intelligent computing services.
A vast journey relies on a solid foundation; with favorable winds and calm tides, one sails smoothly. In the long run, the scaled commercial use of AI WAN is a significant milestone in the global communications industry's evolution toward AI-native networks. In the future, a secure, lossless, and autonomous next-generation WAN foundation will become the key cornerstone supporting sustained growth in the intelligent computing industry, driving the computing sector to accelerate from "building fast" to "using well."