Building software layer that cuts dependence on GPUs for everything in AI, says Infosys

Infosys is not alone in rethinking AI efficiency. Wipro, which has committed $1 billion to AI, has also been working on AI-driven solutions

Software major Infosys is working on a software layer that minimises reliance on graphics processing unit (GPUs) for artificial intelligence (AI) workloads. The development comes at a time when demand for high-performance chips has outpaced supply, leading to rising costs and deployment delays.

"Not every workload needs a GPU. We are working on a software layer that ensures AI models can run efficiently on commodity hardware, reducing the need for expensive and scarce GPUs," Satish HC, EVP & Chief Delivery Officer, Infosys, said at the NASSCOM Technology and Leadership Forum 2025 (NTLF) in Mumbai on February 25.

The issue of GPU dependency has been a challenge for enterprises as demand for AI processing power surged.

“One of the things that we are very proud of sharing is that we have already built four models for our business within Infosys,” Satish added. On January 16, India’s second-largest IT firm said it has built 4 small language models (SLMs).

An SLM is an AI model designed to process and generate text, but is smaller in size and tailored for specific tasks or industries, using less data and computational resources compared to larger models.

Meanwhile, Infosys is not alone in rethinking AI efficiency. Wipro, which has committed $1 billion to AI, has also been working on AI-driven solutions.

"We started by ensuring AI is embedded into our own operations before deploying it for clients. Our Vega Studio has helped us build AI use cases across HR, legal, and manufacturing, optimizing performance," said Sanjeev Jain, COO, Wipro.

With the global AI boom driving an unprecedented demand for GPUs, India has been largely dependent on imports from NVIDIA, AMD, and other global chipmakers.

Nonetheless, to mitigate the issue, the government is now actively working to build domestic semiconductor capabilities to reduce reliance on foreign suppliers.

As part of the Rs 76,000 crore semiconductor mission, India is setting up chip fabrication plants in collaboration. While these plants will primarily focus on logic and memory chips, the country’s first indigenous GPU in three to five years.