Recently, four new computing power projects were collectively launched during Guangzhou's new computing infrastructure construction and computing power supply-demand matchmaking conference. The projects, developed by Guangzhou Mobile, Guangzhou Unicom, Guangdong Cloudbase Technology, and Lianyun Big Data, go beyond traditional server room construction and represent Guangzhou's forward-looking move to build a "super intelligent computing hub."

According to official information, all four projects are located in the Huangpu District and the Guangzhou Development District, with a total investment of 4.839 billion yuan. Upon completion, the projects will be able to add an intelligent computing capacity of over 40,000 Pflops, marking Guangzhou's breakthrough in building a single intelligent computing center with 10,000 cards and 10,000 Pflops.

To put over 40,000 Pflops into perspective: 1 Pflops equals one quadrillion calculations per second. That means 40,000 Pflops could simultaneously support real-time traffic analysis for hundreds of thousands of autonomous vehicles. Such computing power can significantly shorten the time required for AI model training, climate prediction, new drug development, and DNA sequence analysis.

Among the four projects, the China Mobile Guangdong-Hong Kong-Macao (Guangzhou) Data Center will become Guangzhou's first single intelligent computing center with a computing capacity of 10,000 Pflops, providing core support for future industries such as 6G R&D and the industrial metaverse. The China Unicom Knowledge City Data Center will focus on latency‑sensitive applications, including autonomous driving, industrial quality inspection, and unmanned smart factories.

Guangzhou Cloudbase Technology's Huiyun Data Center, a private‑sector project, will use Huawei Ascend 910B chips to serve strategic emerging industries such as embodied AI, low‑altitude economy, autonomous driving, intelligent manufacturing, and biomedicine. Meanwhile, the Lianyun Data Center Phase II Project will be powered entirely by green electricity, complying with national requirements for green and intensive computing infrastructure.

广州打造“超级智算工厂”，首批4个算力新基建落子黄埔

近日，在广州市算力新基建开工暨算力供需会上，包括广州移动、广州联通、云下科技、连云大数据在内的4个算力项目集体开工。这不仅是简单的机房建设，更是广州面向未来打造的“超级智算工厂”。

据了解，本次开工的4个项目全部落子广州开发区、黄埔区，总投资达48.39亿元，建成后将新增智算规模超4万P，也标志着广州实现单体“万卡万P”智算中心建设零的突破。

如何直观地理解“超4万P”算力？1P算力约相当于计算机每秒进行千万亿次的计算，4万P的算力几乎可以同时为数十万辆自动驾驶汽车提供实时路况分析。无论是训练AI模型、预测气候变化，还是加速新药研发、分析基因序列，在“超4万P”的算力面前，科研项目的时间将被大大缩短。

在4个项目中，中国移动粤港澳（广州）数据中心是广州首个单体“万P”级新型智算中心，将为6G研发、工业元宇宙等未来产业提供核心支撑。中国联通知识城机房数据中心将重点保障自动驾驶、工业质检、无人智慧工厂等时延敏感类应用场景。

云下科技汇云数据中心是民营企业项目，采用华为910B芯片，重点服务具身智能、低空经济、自动驾驶、智能制造、生物医药等战略性新兴产业客户。而连云大数据中心二期则100%采用绿电，符合国家对算力基础设施绿色化、集约化的要求。

文｜记者 卢佳圳 通讯员 孙旭东
图｜李剑锋
译｜魏薇
审校｜曾敏