Token costs are not confined to software engineering and are increasingly becoming a concern across enterprises
| Photo Credit:
Chris Ratcliffe
AI-enthusiastic Indian corporates are now hard-pressed to put their money where their mouth is as recent tokenisation costs indicate that the average annual cost of AI coding is nearly triple the salary of the average software developer in India.For some years now, Indian corporates, particularly IT services companies have been talking about the transformative potential of AI technology, particularly in driving up productivity manifold with the same or even a small workforce.Last week, Gartner released a global prediction report stating that AI coding cost would exceed labour cost for the same tasks by 2028. Without clear visibility into token usage across development tasks, organisations risk budget overruns and reduced ability to track cost-to-value outcomes. Conversations with experts reveal that this prediction has already become a reality in India.“Organisations say that AI coding costs are exceeding $3,500 [around ₹3 lakh] per developer per month due to the token-based consumption based pricing model. In one case, a developer ended up consuming $20,000 worth of AI tokens. From an Indian perspective, this means that the AI cost is already more than that of the human labour employed,” said Nitish Tyagi, Senior Principal Analyst at Gartner.Token CostsTaking forward Tyagi’s estimates, AI coding can cost up to ₹36 lakh per year. A software engineer’s annual package pales in comparison.“Annual compensation in IT services is often around 1.5–2 times an engineer’s years of experience, while GCCs and product companies typically pay 2–2.5 times, extending to 3 times for highly sought-after skills,” said Ganesh S Padmanabhan, VP - Recruitment Business at CIEL HR.This means that even professionals with 5–8 years of experience receive salaries between ₹12–24 LPA in IT services and a slightly higher range of ₹25–45 LPA in GCCs.Token costs are not confined to software engineering and are increasingly becoming a concern across enterprises as leaders embrace the narrative of AI eliminating redundant work. In IT services, Infosys Chairman Nandan Nilekanni has spoken about tapping into an AI-first services opportunity of $300 to 400 billion by 2030. TCS Chairman N Chandrasekaran has outlined about a 5-pillar strategy towards AI-transformation and aims to double or triple the company’s $2.3 billion AI revenue with as many AI agents as human employees going forward.Skill BubbleTCS BFSI Products & Platforms uses over 100 AI agents across contact centres, back-office operations, complaints management, for speech recognition, call summarisation, complaint prediction, resolution-letter drafting, intelligent email and secure-message handling, and personalised customer journeys, as per TCS’ annual report for FY26. Even mid-cap IT services companies like Hexaware have stated that they have launched 600 active agents on their Agentverse platform.However, the cost disparity makes it harder for companies to justify the returns on their AI investments. Worse still, tokenisation costs are expected to increase in coming years. Gartner’s client-analyst interactions indicate that GitHub Copilot pricing could rise to 3-8 times current spending levels for software engineering organisations.An industry analyst speaking on condition of anonymity said, companies are now trying to solve this issue by turning the earlier narrative on its head – using AI to resolve complex prompts while assigning humans to more general-purpose tasks.This comes as yet another challenge to aspirants, who until now had been led to believe that niche and AI-led skills would help them secure higher salary increments.“If you’re doing generalist or maintenance work, that slowdown is very real. But if you have AI, cloud, or security skills, you’re looking at 15–25 per cent increments. AI and ML talent demand is the single largest driver of above-inflation salary growth in 2025–2026. The gap between a “regular tech salary” and a “scarce-skills tech salary” has never been wider. It might be a bubble, but for now, it is not bursting,” said Sarbojit Mallick, co-founder, Instahyre.Future DecreaseDespite the complications involved, Nasscom Chairman Srikanth Velamakunni assured businessline that the Human+AI approach will reap its benefits in coming years.“We compare AI to one human being right now but if it can do the work of 10-20 humans, even a 3x token cost is justified. Already, many labs are saying that 100 per cent of their code is generated by AI. Humans oversee and manage the process rather than perform the actual work. So, the Human+AI train has left the station,” he said.Rather than questioning earlier moves, Velamakunni argued the real challenge lies in reducing costs while improving the quality of AI-generated work.With several AI companies preparing to file for IPOs, the entities have revised their offerings from a subscription-based regime costing up to $200 a month (maximum) to a token-based model in the last three months. Further, as people become more ambitious with their AI agents, the models become more token-hungry and the bill goes up again, said Velamakunni. However, as competition kicks in, prices and cost-to-deliver are expected to come down. Chipmakers are also working on improving power efficient, which will again drive down token costs.In the meantime, Tyagi advised companies to focus on their content engineering, model routing techniques and broader model-portfolio adoptions strategies to manage AI costs.Published on June 30, 2026
