Tsavorite’s OPU promises 10x faster AI processing with just 10% of the energy use

AI startup Tsavorite Scalable Intelligence is betting big on a new computing paradigm. Its Founder and CEO, Shalesh Thusoo, says the company’s newly unveiled Omni Processing Unit (OPU) can deliver up to 10 times faster performance while using only 10% of the energy compared to conventional processors — a potential game-changer for the AI industry’s power-hungry workloads.

Thusoo told CNBC-TV18 that the OPU represents a fundamental shift from today’s GPU-centric world. “We set out to define a new processing unit for the next generation of AI compute — what we call the Omni Processing Unit,” he said. “Our processor can run ten times faster with only 10% of the energy use.”

The OPU blends CPU, GPU, memory, and high-speed interconnect into a single, composable architecture. Unlike GPUs — which were originally designed for rendering graphics — Tsavorite’s OPU is purpose-built from the ground up to handle AI workloads efficiently. “This goes beyond the GPU,” Thusoo explained. “GPUs will continue to exist, but the OPU is fundamentally a new architecture designed for all AI workloads.”

To prove its capabilities, Tsavorite has created a prototype system that allows partners to test real-world workloads directly on its hardware without modifying their existing software. This flexibility has helped the company secure over $100 million in pre-orders, with partnerships spanning Sumitomo Corporation in Japan, a leading OEM in Europe, and several deep collaborations in India.

India, in particular, will play a pivotal role in Tsavorite’s global growth. “From day one, Tsavorite was started simultaneously in both Bangalore and California. Going forward, India will likely be larger for us than even the US,” Thusoo said. The company is also developing the Helix platform, a system-level AI appliance that will be fully designed and manufactured in India and deployed worldwide by 2026.

Looking ahead, Thusoo believes the AI race will not be won by those with the biggest chips, but by those building the smartest and most efficient systems. “It’s not about chip size; it’s about intelligence and optimisation,” he said. “AI needs to be scaled across devices — from desktops to hyperscale data centres. We’re just at the beginning of that journey, and our goal is to enable sustainable AI for the future.”

Watch accompanying video for full show.Read about our editorial guidelines and ethics policy