Huawei Chairperson Sees Scarcity and Expensiveness of Computing Power Major ...

Credit: Visual China

BEIJING, September 20 (TiPost) – Chinese tech giant Huawei’s Rotating Chairperson Meng Wanzhou said on Wednesday that computing power is driving the development of artificial intelligence (AI) and the scale of computing power determines AI’s iteration and innovation speed.

"The development of artificial intelligence is primarily driven by computing power. Large models require significant computing power, and the size of computing power determines the speed of AI iteration and innovation, as well as that of economic development," Meng said at the Huawei Connect 2023 Conference in Shanghai Wednesday.

She noted that the scarcity and costliness of computing power have become key roadblocks to AI development. Huawei is committed to building China's computing power infrastructure and offering a second choice for the world.

Meng believes that neural network models in artificial intelligence with parameters exceeding billions or even trillions are accelerating their integration into various industries, and AI development is experiencing a turning point.

"From the era of small models to the era of large models, there has been a qualitative leap in the practicality of AI technology," she stated. In the past, different application scenarios required the development of different models. Now, a large model can adapt to multiple business scenarios, significantly lowering the barriers to AI development and application. This shortens the cycle from technology to application, making it possible to shift from workshop development and customized scenarios to industrialized development and scenario-specific optimization, and to solve industry problems based on large-scale models.

She proposed that Huawei supports every organization in training its own large models using its own data, which allows each industry to develop its industry-specific large models using its own expertise.

She said that Huawei would leverage its comprehensive advantages in computing, storage, networking, energy, and other fields, changing the traditional server stacking model. Huawei plans to create AI clusters with a system architecture approach to achieve the integration of computing power, operational capacity, and storage capacity, breaking through computing power bottlenecks.

In October 2018, Huawei introduced its self-developed AI chip, Ascend, which competes with Nvidia's GPU (graphics processing unit). With the recent surge in demand for AI chips driven by models like ChatGPT at the end of last year, Nvidia's GPUs faced supply shortages and had to expand their production capacity. As Chinese AI chips lagged behind Nvidia in terms of performance and software ecosystem, they were targeted at Chinese customers with independent and controllable needs.

Huawei on Wednesday unveiled the Ascend AI Computing Cluster Atlas 900 SuperCluster at the conference. It utilizes Huawei's self-developed CloudEngine XH16800 to connect a large number of self-developed chips into a cluster, capable of supporting the training of models with over a trillion parameters.

Computing power has been an area where Huawei executives have frequently spoken out in recent times. In a speech disclosed on Tuesday, Huawei's founder Ren Zhengfei said that the fourth industrial revolution was vast in scale, and its foundation was substantial computing power.

On September 15, while attending the 2023 World Computing Conference in Changsha, Hunan province, Huawei's then rotating chairman Xu Zhijun also mentioned that the computing industry was at a crucial point in the competition between China and the U.S. He stated that the U.S. was using two methods to hinder China's computing industry: export controls preventing China from obtaining advanced computing chips and systems and sanctions targeting China's semiconductor and artificial intelligence industries, among others.

"In our prediction, the U.S. containment of China will be long-term, and we cannot entertain any illusions," Xu said.

He also mentioned that due to U.S. export controls on China's semiconductor industry, Chinese semiconductor manufacturing processes would remain in a catch-up state for quite some time. However, if China did not use its domestically produced chips, servers, and PCs, the gap would remain.

"If we use domestically produced chips, servers, and PCs on a large scale, this may drive progress in the technology and products, and then we can gradually catch up," he stated, emphasizing that China should be determined to build an independent computing ecosystem.