AI computing costs could fall as code generation becomes a one-time expense, says Zoho’s Sridhar Vembu

Vembu also expressed optimism that the cost of deploying AI for specific applications would continue to decline, as verified AI-generated code can be executed without relying on expensive AI infrastructure.

As companies pour billions of dollars into AI infrastructure and computing, Zoho Corporation’s co-founder Sridhar Vembu believes that the scale of investment may not be needed for most applications in the future.Speaking at the launch of the World Bank Group’s World Development Report 2026: The Promise of Artificial Intelligence, organised with the Ministry of Electronics and Information Technology (MeitY) on Thursday, Vembu said that AI models do need substantial computing power to generate, test and correct code.

But once the code is verified, it can be run repeatedly without the same level of AI computation.“The economic value of AI lies in how often the generated code can be reused.

Instead of relying on a large AI model every time a task is performed, companies can generate the code once and run it repeatedly using ordinary computing resources,” said Vembu.

He said that the emergence of smaller AI models trained to perform specific tasks rather than handle a wide range of subjects.He said that such models do not need extensive knowledge across different domains or the ability to answer unrelated questions. “Instead, they only need to perform the particular tasks they are designed for efficiently and accurately,” he added.

Vembu also expressed optimism that the cost of deploying AI for specific applications would continue to decline, as verified AI-generated code can be executed without relying on expensive AI infrastructure. “It may be possible to run many of these applications on ordinary computers, laptops or even mobile phones.

We are already moving towards a future in which smaller models can operate directly on devices,” he said.On India's AI potential, Vembu said that the country has a strong pool of scientists and engineers, particularly in software development, verification and validation.He added that rather than focusing solely on building larger general-purpose AI models, India should explore the economic potential of smaller, specialised models designed to perform specific tasks efficiently and at a lower cost.