AI Drives ZTE’s New Hardware and Software Solutions at MWC Shanghai

Written by Ian Fogg

AI was one of the major topics at MWC Shanghai, which took place last week with CCS Insight in attendance. Among the eye-catching announcements in this area, we saw 中兴 introduce its new Nebula Telecom Large Model. This will form the basis for AI functionality across the company’s portfolio of network infrastructure and digital service solutions. Perhaps even more significantly, ZTE expanded its range of products to run self-developed and third-party AI models, including new high-performance green servers, storage, and data centre products, enabling more efficient training, inference and operation of AI technologies.

In a keynote speech at the event’s AI First session, ZTE CEO Xu Ziyang set out a new vision for AI-driven development, which he framed as an AI-driven industrial revolution. To overcome obstacles to wider use of AI, he highlighted how connections at different levels enhance AI capabilities, including die-to-die, chip-to-chip, server-to-server and DC-to-DC. He explained how ZTE advances its intelligent digital solutions by adapting underlying connectivity and computing technologies, which are essential for supporting the parallel processing needs common to large AI models.

The company also sees a need for wide collaboration between players. ZTE’s server products are designed to work alongside a range of GPUs — both developed inside China and outside — so, for example, a customer can buy AI training servers from ZTE and then slot in the most appropriate brand of GPUs.

ZTE offers the TECS platform to help customers manage resources effectively across multilayered set-ups that include AI model training centres, edge training and inference integration solutions. Its offerings include high-performance processors, and data processing units. It offers what it calls an AI studio that can run complicated workflows ranging from data collection to AI inference including both domain-specific and universal base models, which reduce barriers for working with large language models.

Lowering energy usage in data centres becomes even more important with the increased compute demands from training AI large language models. ZTE is addressing this requirement in multiple ways, including by assessing the carbon footprint of more than 100 of its products. In data centres, there are now solutions that ZTE claims have a power usage effectiveness that is down to just 1.1 — lower figures indicate greener operation. For example, its IceCube cabinet can hold up to 40 1U servers with liquid cooling.

The Carbon Disclosure Project, better known as CDP, rates ZTE as “A”, up from “B” in the 2021 report, which indicates ongoing sustainability progress.