Insights from the 2023 Global 6G Technology Conference: AI‑Driven IT Stack and Future Trends

From March 22‑24, 2023, the 2023 Global 6G Technology Conference was held in Nanjing under the theme “6G connects the world, creating the future together,” organized by the Future Mobile Communication Forum and Zijinshan Laboratory with guidance from the national 6G R&D workgroup and expert committee. C114 provided live photo and text coverage of the event.

During the “2030 Technology Development Trends” sub‑forum, Chen Shangyi, Chairman of Baidu’s Technical Committee, highlighted a fundamental shift in IT architecture in the era of artificial intelligence.

Historically, the IT stack consisted of a chip‑OS‑application hierarchy, forming distinct industry ecosystems. In the AI era, the stack has evolved to chip‑framework‑model‑application. The framework layer is dominated internationally by TensorFlow and PyTorch, while domestically Baidu’s Paddle (Fei Pang) and Huawei’s MindSpore are prominent. Large models such as ChatGPT and Baidu’s Wenxin are driving the current hype.

The new stack creates a soft‑hardware coupled ecosystem, with open‑source frameworks. For example, Baidu’s Paddle hosts hundreds of thousands of community‑contributed models, which power a variety of industry services and form diverse ecosystems.

Chen emphasized that each layer in the new stack feeds back to the others. Frameworks must adapt to chip capabilities and provide high‑performance optimization, including standardized algorithms and interfaces. Models rely on frameworks for training and inference infrastructure, especially for large models that demand distributed computing, parallel training, and resource scheduling. Applications, in turn, impose strict requirements on performance, cost, and development cycles, driving end‑to‑end optimization.

Many companies, universities, and research institutes work on individual layers, but few achieve full end‑to‑end optimization. Only large commercial entities can push optimization to the extreme; Baidu, for instance, faces challenges in development speed, user experience, and cost, necessitating comprehensive stack‑wide tuning.

Chen outlined Baidu’s full‑stack deployment: Kunlun chips at the hardware layer, Paddle at the framework layer, Wenxin large model at the model layer, and business applications such as search, autonomous driving, maps, and Xiaodu at the top. He identified three key trends for AI and communication convergence:

First, AI for communications – integrating AI into communication infrastructure to provide AI services as part of the network.

Second, AI‑driven optimization of storage and transmission, dramatically improving efficiency, quality, and security of communications.

Third, AI‑enabled real‑time network monitoring, facilitating automated operations, fault detection, and self‑healing.