The AI WAN revolution
Published on
March 20, 2025
Interview
At a Mobile World Congress consumed with AI’s evolving role in mobile networks, Huawei’s new AI WAN solution shows that the AI transformation is extending to every facet of network infrastructure
We caught up with Huawei’s President of Router Domain, Data Communication Product Line, Mr Zuo Meng, to discuss this new solution and what incorporating AI into WAN technology will mean for carriers.
The growing IP network challenge
For many years now, it has become increasingly clear that carriers’ traditional business model, based primarily on selling bandwidth, is producing lacklustre results. Efforts to diversify these offerings have similarly failed to generate a significant boost in revenues, leaving the market.
“User traffic growth is relatively slow, and this means that traditional revenue growth has also been small,” said Meng. “At the same time, new business is not growing as fast as hoped. Carriers are trying to move away from selling bandwidth and instead sell services, but they have seen only limited success.”
To make matters worse, Meng explained, IP networks continue to grow in complexity, with more nodes, more devices, more diverse communication protocols, and greater encryption than ever. This makes O&M configuration a major challenge, driving up OPEX for operators and requiring greater levels of automation to ensure a smooth user experience.
“Carriers are not only looking for operational simplification, but also the potential to explore new business models and revenue streams. For that, they will need AI integrated in their WAN networks.”
A solution in three parts
To overcome these fundamental challenges and create future-proofed IP networks, Huawei has focussed its AI WAN efforts on three key elements: embedding AI within the router, enhancing AI connections, and introducing AI agents in the ‘AI Brain’.
The base layer: AI Router
The first aspect of Huawei’s AI WAN solution is the incorporation of AI into the WAN hardware itself, allowing for more intelligent traffic management. Most importantly, this will allow the router to identify encrypted traffic without needing to access user data, improving both efficiency and security.
“The primary function of AI routers is to identify the application responsible for sending a data package”,” explained Meng. “Effective analysis of an application’s behaviour allows this identification without the need to inspect the data packet itself”
The middle layer: AI new connections
By leveraging AI, the solution can also help more effectively adjust itself to meet the network requirements in various applications in the ToC, ToH, and ToB segments. This includes flow-level scheduling, allowing multiple value-added services to run efficiently simultaneously, without compromising network quality.
“At this layer, we can make use of AI to improve IP connection reliability and throughput, or to provide guaranteed low latency for premium services,” explained Meng.
The upper layer: The AI new brain
At the topmost layer, the AI WAN solution is equipped with its own ‘AI brain’, based on a Network Digital Map, a kind of SDN controller. This layer supports three specialised AI agents that Huawei calls ChangeSpirit, AssurSpirit, OptimSpirit each of which can quickly assist carriers in Online change simulation, Minute-level optimization, and Fault self-disposal.
“I think these three spirits adde d to the Digital Map will help customers to simplify their O&M very greatly,” said Meng, adding that this Network Digital Map and Network Foundation Model serves as a foundation to which new ‘spirits’ can be added.
“In future, the existing spirts alone will not be enough. We will continue to import new spirits based on big AI models to continue to simplify carriers’ O&M,” he added.
Helping carriers make the most of the AI opportunity
With AI more closely incorporated into WAN, network operators will be better positioned to offer differentiated home services based around guaranteed quality of service.
“Our solution can help identify the type of application – for example, gaming or streaming – and then speed up the network accordingly,” said Meng. “It will also help us to understand the relationship between the network’s KPI and the application’s KQI, so we can better optimise performance.”
It will also make these networks better at handling AI-generated traffic, such that generated when training large language models.
“A traditional router cannot identify an ‘elephant flow’ [the large amounts of remote direct memory access traffic created by GPUs during AI training], which can create challenges for the network,” explained Meng. “Our AI WAN solution can identify these flows automatically and adjust the network accordingly.”
What does the future hold?
Ultimately, Meng suggests that this is just the start of the telecoms industry’s journey with AI, with AI WAN set to continue to evolve in the coming years.
“I think our three-part foundation is very stable. We’ll continue to iterate. We’ll work with our customers to identify pain points, as well as new applications,” explained Meng. “We’ve talked to many customers about AI WAN and they are all very excited – they want to advance to tests and commercialisation very quickly.”
“In our industry, we really needed a big change, and I think AI can be that change for us. The opportunity is far greater than the challenge,” he concluded.