Alibaba Cloud's New System Cuts Nvidia GPU Usage By 82%, Amid Trump's Flip Flop On AI Chip Ban On China

Benzinga
2025.10.19 08:18
portai
I'm PortAI, I can summarize articles.

Alibaba Cloud has launched Aegaeon, a new computing pooling system that reduces Nvidia GPU usage by 82% for AI models. Tested for over three months, it decreases the number of required Nvidia H20 GPUs from 1,192 to 213 for models with up to 72 billion parameters. This innovation aims to enhance efficiency by allowing a single GPU to support multiple models, addressing resource inefficiencies. The development occurs amid concerns over Nvidia's presence in China and the impact of U.S. policies on its market share.

Alibaba Group Holding (NYSE:BABA) has introduced a new computing pooling system called Aegaeon, which dramatically reduces the reliance on Nvidia (NASDAQ:NVDA) GPUs by 82% for AI models.

Three Months In Testing

This innovation was tested in Alibaba Cloud's model marketplace for over three months, according to a research paper presented this week at the 31st Symposium on Operating Systems Principles (SOSP) in Seoul, South Korea.

The Aegaeon system has successfully decreased the number of Nvidia H20 GPUs required from 1,192 to just 213 for serving models with up to 72 billion parameters.

"Aegaeon is the first work to reveal the excessive costs associated with serving concurrent LLM workloads on the market," the researchers stated in the paper.

Researchers from Peking University and Alibaba Cloud emphasized the high costs associated with serving concurrent large language model workloads.

One GPU For Multiple Models

Alibaba Cloud, the AI and cloud services division of the Hangzhou-based Alibaba, aims to boost efficiency by pooling GPU resources, allowing a single GPU to support multiple models.

The system addresses resource inefficiency, as previously, 17.7% of GPUs were allocated to serve only 1.35% of requests in Alibaba Cloud's marketplace.

Cloud service providers like Alibaba Cloud and ByteDance‘s Volcano Engine manage thousands of AI models simultaneously, often leading to inefficiencies. The Aegaeon system seeks to optimize this process by reducing the number of GPUs needed.

Growing Concerns Over Nvidia's China Presence

This development comes amid growing concerns over Nvidia’s presence in China. Recently, China raised security concerns about Nvidia’s H20 chips, especially regarding potential backdoor risks. As part of its deal with Nvidia, the Trump administration has struck an agreement for a 15% revenue share from the company's chip sales to China.

Nvidia CEO Jensen Huang stated that Nvidia’s market share in China has plummeted from 95% to zero. He expressed concerns over the impact of U.S. policies on Nvidia’s market presence in China.

Despite these challenges, Nvidia has financially insulated itself from potential escalations, as its guidance assumes zero revenue from China, according to Huang.

  • Elon Musk Says 100K People, 1 Million Tons Of Cargo Needed To Sustain Mars Colonies, Reiterates Starship’s Importance: ‘Will Land Humans On Mars’

Photo courtesy: Shutterstock

Source: Benzinga The copyright of this article belongs to the original author/organization.
The views expressed herein are solely those of the author and do not reflect the stance of the platform. The content is intended for investment reference purposes only and shall not be considered as investment advice. Please contact us if you have any questions or suggestions regarding the content services provided by the platform.