Sponsored Feature The telecoms and computing industries have embraced decoupling over recent years, as each sought to decompose or modularize technologies and products to deliver better performance and lower costs, while becoming ever more agile.

However, with AI changing the technology landscape, ZTE CEO Xu Ziyang has called for recoupling where appropriate to deliver the power and capacity that customers need, and more importantly to deliver it sustainably.

Xu’s call to rethink how connectivity and computing work together comes against the background of spiraling investment in both by operators, cloud service providers, and customers as they struggle to keep up with the demands of AI.

Gartner predicts that worldwide spending on AI products, software and services will hit almost $1.5 trillion this year. This will increase again to over $2 trillion in 2026, with the increased integration of AI into smartphones and PCs being a key driver, along with spending on infrastructure.

In his opening keynote at ZTE’s Global Summit and User Congress in Milan, Xu said every telecom technology had to meet a main purpose and a clear demand, such as voice, terminal applications, or live video.

Now, he said, “the demand is moving to AI and computing.” To meet this demand, the company had to define two curves: networking and computing.

AI demands massive amounts of data, which has to be moved both within datacenters and across networks, pushing bandwidth requirements from hundreds of gigabits to the terabits range, as platforms strain to process more tokens per second.

But latency also becomes a pressure point, he said, particularly as inference moves to the fore. Latency tolerances are moving from the microsecond range to nanoseconds, Xu said. This also reflects more activity at the edge, whether that is for consumer applications, such as home entertainment or gaming, or to underpin industrial applications such as R&D or smart factories.

At the cutting edge

“We have launched a variety of products to support these needs,” Xu said, “Including high-speed parallel interconnects between dies, XLink serial interconnects between chips, 100T switching interconnects between clusters, and 800G long-haul interconnects for data centers.”

The company used the Milan event to detail its AI Core strategy, which will see it integrate large and small AI models into its core architecture, helping operators move from a “traffic-driven model” to an “experience-driven” model. It also unveiled Agentic Ops, a framework for fault-finding, complaint resolution, and networking change, it showcased its range of general purpose and AI servers.

But if telecom operators and cloud services providers have to change their networks to accommodate the demands of AI hungry customers and support the compute underpinning AI, they also need to incorporate AI into their own operations, Xu explained.

AI can enhance networks’ “flexibility and robustness” he said. For example, the firm had added an AI compute card to its 5G baseband unit, which intelligently manages and saves power, while prioritizing key traffic such as live streams, during busy networking periods. In the core network, Xu said, it had deployed AI to detect text messaging scams, “with a 99 percent success rate”.

More broadly, he said, network systems produce millions of log data points every day. It is impossible to manually correlate and analyze these to spot issues and optimize the system.

However, at ZTE, “We are using large-scale communication models in layer three and layer four autonomous networks, and AI is making the networking more efficient.”

For example, ZTE is working with Thailand’s AIS to integrate its Nebula model into the operators’ network. This had helped cut both fault response time and fault management MTTR by 30 percent, and reduced networking configuration and upgrade timescales by 50 percent.

At the consumer end, Xu said, there is a need for more affordable AI hardware, in line with its AI for All strategy. ZTE used the Milan event to showcase its Nubia Neo series of gaming and photography focused smartphones, which are particularly aimed at youth customers and feature large batteries and high-quality screens. The company has boosted its international smartphone revenue by 30 percent in the first half of 2025.

Sustainability is key

But all of this requires an ongoing focus on cost and sustainability. This is particularly important as, said Xu, it is clear that the requirements of inference are almost twenty times those of training.

Training might increase IQ, he said, “While inference increases productivity.” So, for more cost sensitive consumers, it was reducing high bandwidth memory use in chipsets to reduce costs, as well as PCI plug in cards and high-density GPU arrays, all of which can “further reduce the cost per token, bringing token prices down to the bit level.”

When it comes to the operator level, he said, “Sustainable AI construction is very important.” This includes focusing on elastic cooling and power, using more green energy, and pooling different GPUs in parallel to support inference services.

Reuse of components was critical, particularly the amount of investment some of them represented. “We should reuse different generations of the GPU card, ensuring every generation of GPU card can recover their commercial value and construction cost.”

He cited the example of China’s National AI Training Center, where ZTE provided cooling and flexible power infrastructure for 3,000 racks running over 10,000 GPU training cards. This installation achieved a PUE of under 1.2.

Xu’s call for a rethink of telecoms and networking, and sustainability more broadly, was echoed by GSMA CEO John Hoffman, who told the audience that telecoms connectivity was critical to ensuring the success of AI.

Hoffman flagged how ZTE had worked with China mobile to deliver a Green Telcom Cloud which optimizes sustainability while keeping business continuity and user experience to the fore.

This is achieved by using neural network models to monitor and analyze parameters such as server load and network traffic and achieving an optimal balance of compute in response.

As well as reducing energy consumption, this approach also slows down the aging of electronic equipment, extending its service life and reducing maintenance and service costs.

Accelerating this reconvergence of compute and networking, and doing so sustainably, is critical to fully realizing the potential of AI and telecoms.

“The world is moving from bits to tokens,” said Xu. “And increasing AI will boost productivity and make the world a better place.”

But, he added, this will all require more collaboration and more recoupling. “The promise from ZTE is that we will use the latest technology, our best technology, for the service for all of our customers,” he told ZTE’s customers in Milan. “So, we need your kind guidance…[to] find an opportunity [where] we can win together.”

Sponsored by ZTE.