Ziroh Labs, is a global startup with offices in Palo Alto and Bengaluru, Ziroh Labs is redefining AI infrastructure with Kompact AI, an AI platform that enables a large language model (LLM) inference on CPUs. This innovation eliminates the dependency on expensive GPUs, making AI workloads significantly more affordable and accessible. The platform is uniquely positioned to drive AI adoption in cost-sensitive and data-regulated industries, delivering CPU-powered AI that runs three times faster than current standards—without the need for GPUs.
At the forefront of solving complex compute and data challenges through frugal innovation, Ziroh Labs has now reached a major milestone with the launch of Kompact AI—a transformative platform that enables complex AI models to run efficiently on standard CPUs, achieving 3x the performance of current solutions without compromising on quality. This innovation was developed in collaboration with IIT Madras and the IITM Pravartak Technologies Foundation, underscoring Ziroh’s commitment to delivering cutting-edge and accessible technology for global impact.
To support open and scalable AI innovation across India and beyond, IIT Madras and IITM Pravartak have also jointly launched the Centre of AI Research (CoAIR), with Ziroh Labs leading the development of domain-specific, CPU-optimized AI platforms. Kompact AI is built for India—designed to run large models affordably on CPUs without internet dependency, democratizing AI access across rural, enterprise, and edge environments. This launch represents a significant step toward India’s “AI for All” vision—enabling real-time intelligence at the edge, in classrooms, on farms, and in other low-resource settings. Some of the features of Kompact AI are below:
Bringing AI to those with limited compute resources, addressing the accessibility challenges posed by the dominance of “Big AI.”
Supports the “AI for All” mission by enabling the development and deployment of affordable AI on CPUs
Enabling AI models to run on lightweight, resource-constrained environments, opening doors for broader adoption and innovation.
Focusing on models under 50 billion parameters and optimizing for CPUs, democratizing AI for a enterprises and applications.
Read the full story that first appeared in The Global Indian here:
Leave a Reply