China Telecom on why merging network and AI infrastructures is worth the effort

At the FutureNet World conference in London in May speakers debated whether AI workloads will generate enough data to warrant investment in designing networks for AI. For China Telecom, however, the question is already settled.

“We really do think telco has a role in merging the two to provide both connectivity and [cloud] compute on a coherent platform, which we call a future digital infrastructure,” says Dr. Yue Wang, Chief Technologist, Network and AI, China Telecom.

“If we look at the volume growth of the data at the moment, we'd be like, ‘Yeah, our network is doing fine.’ But what we need to realize is that the network in the future will need to support connectivity and compute demands at the same time,” explains Wang. “I think the development of AI and the network needs to be interdependent … and for that we would want the ecosystems to evolve together.”

Equally, if telcos don’t merge networks and AI computing, then she believes other players will.

“If we don't do it, someone else will do it to for us, and then we lose the opportunity … and we get marginalized,” Wang says.

Even today, “for B to C cases, telco is still serving as a pipe and the financial value generated by AI is going through the network [and flowing] … to the hyperscalers and chip vendors,” according to Wang.

Striking out alone

China Telecom has designed its converged digital infrastructure and has built a total of 35 exaflops of computing power spread across a handful of locations. For an idea of what this means in practice, Indiana University states that “to match what an exaflop computer can do in just one second, you’d have to perform one calculation every second for 31,688,765,000 years.”

With more than 400 million mobile customers and nearly 200 million wireline broadband subscribers, China Telecom has the advantage of massive market scale. In addition, it can tap into a large skilled technology workforce to “quickly scale up technologies,” says Wang.

The company benefits from the ability to take a long-term strategic view. It began laying the groundwork for cloud-network convergence, integrating computing and connectivity to create a more agile, scalable infrastructure, several years ago.

“Today, that vision has taken shape and is being accelerated by the demands from AI, positioning us at the forefront of intelligent network transformation,” says Wang.

For smaller telcos, however, building a converged infrastructure can be a daunting endeavor. Especially as there is no guarantee of future revenues.

“For telecom there's a significant upfront investment, but at the moment the industry doesn't seem to be a clear monetization model and that [is stalling] investment,” acknowledges Wang.

A meeting of opposites

Money isn’t the only obstacle. Trying to unite two very different worlds poses considerable technical and operational challenges.

China Telecom’s goal is to integrate AI with connectivity and computing on the resource, operational and service levels, explains Wang. To this end, it is working to unify the management and orchestration of cloud and network resources so that their evolution shifts from independent development to comprehensive integration.

However, “AI and the way traditional telecoms were designed … don't match,” says Wang, adding that there is “a mismatch in the pace of evolution between the telco and the AI.”

Whereas telecoms networks have long development cycles, with a new ‘G’ coming around every 10 years, “AI moves so fast … we are always trying to keep up,” says Wang. “And … new technologies in AI, they are not just rapid, they are also quite disruptive.”

This means that “when we talk about enabling AI-native networks, the reality is the technical infrastructure is struggling to keep up,” she explains. “It's not just a matter of software updates. There are significant challenges around the scope, the deep deployment, the complexity of the system [and the] backward capability [with] the legacy infrastructure.”

Serving enterprises

While China Telecom strives to crack the technical challenges of merging cloud computing and network infrastructures it is also creating AI services for customers. Again, it is taking a foundational approach, which includes building multiple large language models to serve a variety of vertical B2B requirements.

Wang, who is a member of the board of directors of the O-RAN Alliance, emphasizes that a unified platform for AI services needs to be end-to-end: “From the terminal to the RAN to the core to the cloud, every part of the chain needs to be intelligent.”

In China, burgeoning AI use cases and services include new forms of robotics.

“There are already robotic dogs running in the street of China, lifting goods and buying groceries for people. So, although not a norm yet, it is emerging,” says Wang. “And for those kind of use cases edge plays a perfect role because it reduces latency and brings intelligence closer to the user.”