Indian Startup Develops System for Running AI Models on CPUs

Indian startup Ziroh Labs has developed a system that enables the operation of advanced AI models without the need for powerful chips, according to Bloomberg. 

The Kompact AI platform was developed in collaboration with the Indian Institute of Technology in Madras. It allows neural networks to efficiently run on central processing units (CPUs) used in standard computing devices. 

The startup focuses on the inference process, which is the operation of an AI model after its training. It optimizes leading neural networks for use on PCs. As an example, it demonstrated the operation of Meta’s Llama 2 and Alibaba’s Qwen2.5 on a laptop with an Intel Xeon processor.

Ziroh Labs claims that its approach delivers high-quality results, and the technology has been tested by American chipmakers Intel and Advanced Micro Devices. 

“In the coming years, this will have a very strong impact on the market,” said the startup’s technical advisor, William Raduchel.

The director of the Indian Institute of Technology in Madras, V. Kamakoti, highlighted the issue of access to powerful graphics processors (GPUs) in countries like India. 

“The gap in AI is related to the fact that only those with expensive GPU resources can access, develop, and implement powerful artificial intelligence. We are demonstrating that you don’t need a revolver to kill a mosquito,” he noted. 

Back in January, Chinese startup DeepSeek released an efficient AI model, R1, which raised questions about the necessity of spending billions on graphic chips. 

Meanwhile, Nvidia announced new processors for creating and running AI models—Blackwell Ultra and Vera Rubin. It also introduced an engine for simulating robot movement and a range of other products.