AI Start-Up Challenges Nvidia's Dominance with Beijing Support

A new artificial intelligence (AI) framework has been developed by a team connected to Tsinghua University in China. This innovative framework aims to lessen the reliance on Nvidia chips for AI model inference, representing a significant push for the nation towards technological self-sufficiency.

The new framework, named Chitu, provides high-performance capabilities for large language models (LLMs). It is designed to work with chips manufactured in China and directly competes with Nvidia's Hopper series graphics processing units (GPUs). This was detailed in a recent announcement made by the start-up company Qingcheng.AI and a team headed by Professor Zhai Jidong from Tsinghua University.

AI frameworks are essential for building complex and intelligent AI models. They offer various libraries and tools that help developers to design, train, and validate models in an efficient manner.

Since its open-source launch on Friday, the Chitu framework has garnered attention for its support of popular models, including those from DeepSeek and the Llama series by Meta Platforms. Qingcheng.AI reported that when Chitu was tested with the full-strength version of DeepSeek-R1 using Nvidia's A800 GPUs, it achieved remarkable results: a 315 percent increase in model inference speed while cutting GPU usage by half compared to foreign open-source frameworks.

This move is part of a larger initiative by Chinese AI firms to diminish their dependency on Nvidia. The firm has been significantly impacted by U.S. export controls, which restrict the sale of its advanced H100 and H800 chips from the Hopper series to clients based in China.

AI, Nvidia, China