THE TECHNOLOGY INTERVIEW

is the estimated time frame for physical AI to become as large as consumer and gen AI

Balancing investment with return Boards of directors are increasingly demanding a clear return on investment( ROI) within an 18-month window, according to HCLTech’ s The AI Impact Imperatives, 2026 report. This creates pressure on technology teams to deliver results rapidly while managing the significant costs associated with inference – the process of running a trained AI model – and energy consumption.

Vijay believes that an 18-month timeline is realistic, provided that organisations approach deployment with strategic intent. While some projects may naturally extend beyond this window, the goal is to keep payback periods short.

“ There’ s a definitive trend in terms of energy costs and inference costs,” he explains.“ Token costs are dropping significantly year-on-year. Today they are a fraction of what they used to be a couple of years back.”

Despite falling token costs,“ token maximisation” remains a risk. This occurs when organisations use more expensive, frontier AI models for tasks that could be handled efficiently by smaller, cheaper and more specialised models.

“ You need to make your system smart to use the right model for the right kind of work,” Vijay suggests.“ Technically, we call it model routing.”

By routing simpler tasks to smaller models and reserving more powerful models for complex queries, businesses can dramatically reduce their operational expenditure. HCLTech advocates for the use of AI factories, where enterprises build intelligence locally rather than relying exclusively on cloud-based frontier models.

This approach addresses data privacy, regulatory compliance and cost control. By planning infrastructure capacity six

32 October 2026