For more than two decades, India’s information technology industry built something genuinely admirable: a world class services engine that became the back office of the global economy.
For more than two decades, India’s information technology industry built something genuinely admirable: a world class services engine that became the back office of the global economy. TCS, Infosys, Wipro, HCL Tech, and Tech Mahindra collectively generated hundreds of billions of dollars in revenue, employed millions of engineers, and powered an entire ecosystem of technical education across the country. Then came the AI revolution – and the industry was caught almost entirely unprepared.
The launch of ChatGPT in November 2022 was not a sudden shock. It was the visible peak of a transformation building since deep learning went mainstream after 2015. But Indian IT firms, locked into low-risk service contracts and investing less than three per cent of revenues in R&D compared to ten to twenty per cent at global technology companies, had neither the incentive nor the institutional culture to see it coming. By the time the alarm was impossible to ignore, the frontier had already been claimed by OpenAI, Google DeepMind, Anthropic, Meta, and a set of well-funded Chinese laboratories.
What followed was a period of genuine reckoning. The traditional pyramid model, in which armies of junior engineers churned out standardised code with revenues scaling directly with headcount, began to crack. AI could now perform many of those base layer tasks: routine code generation, testing, data processing, and documentation. At the same time, the growth of over 2,000 Global Capability Centres meant multinationals that once outsourced wholesale were now building substantial in-house teams in India. The existential question became unavoidable: if Indian IT cannot build the AI, and its traditional clients are increasingly building their own teams, what exactly is the industry’s role? Four structural failures explain why Indian IT missed the frontier AI race entirely. The first is the services trap.
Indian IT thrived on labour arbitrage and client-centric execution. Revenue was built on contracts, not on products or intellectual property. There was little incentive and considerable risk in allocating capital to speculative, long-horizon research. Shareholders were rewarded with consistent cash returns, not with bold technology bets. The second was negligible R&D investment. While global technology companies reinvested ten to twenty per cent of revenues into research, Indian IT averaged below three per cent. The gap was not primarily financial. These were profitable businesses with strong balance sheets. The failure was cultural and strategic, rooted in leadership cadres drawn entirely from delivery backgrounds. The third was a compute deficit. Training large-scale AI models requires enormous concentrations of GPU infrastructure.
The hyperscale data centres needed for frontier model development were concentrated in the United States and China. India’s data centre buildout, now accelerating, is doing so primarily to serve Western technology companies, not to support indigenous model development. The fourth, and perhaps most consequential, was brain drain. India produces world class AI researchers, but the compensation and research culture available at well-funded global laboratories consistently pulled the country’s best minds abroad. Without an AI research culture at home, there was no institutional foundation from which frontier models could emerge. This remains true today. Here is what the frontier AI companies have discovered, and what has opened the door for Indian IT’s current strategy: powerful models, when deployed in actual enterprise environments, frequently do not work as expected. This is not a criticism of the models. It is a structural reality. Real enterprises, whether banks, manufacturers, hospitals, or logistics companies, operate in complex environments shaped by decades of legacy systems, proprietary data formats, regulatory requirements, and organisational habits. A general-purpose, large language model cannot simply be pointed at a company’s data and asked to transform its operations. The integration, customisation, governance, and change management work required to make AI function in production is substantial. And frontier AI labs are neither equipped nor inclined to do it. This is the middle layer.
It is where Indian IT is now staking its future. A chief executive of a large non-technology enterprise does not want to build an AI team from scratch, navigate a rapidly shifting landscape of models and tools, or spend two years on a technology project while running a day-to-day business. He wants a trusted partner who will take the complexity off his hands and deliver measurable outcomes. For decades, that trusted partner was an Indian IT firm deploying enterprise software or migrating infrastructure to the cloud. Now, the same relationship is being offered for AI. Each of the five major Indian IT companies has articulated this strategy through a distinct branded platform. TCS has launched WisdomNext, a unified enterprise AI orchestration platform built around multi LLM integration, AI agent deployment, and governance. Its differentiator is an integration first approach that combines cloud and AI delivery under a single contract.
Infosys has built Topaz, the broadest enterprise AI stack among Indian IT firms. With access to over 12,000 pre-trained AI assets, knowledge graphs, and workflow automation tools, it is heavily platformised and oriented toward specific industry verticals. Wipro’s ai360 embeds generative AI across every service, platform, tool, and delivery process. It offers multi cloud AI architecture support and a strong emphasis on responsible AI, including ethical filters and compliance guardrails. HCL Tech’s AI Force and AI Foundry platform focuses on agentic AI and workforce automation, featuring a modular AI stack, software engineering automation, and IT operations AI. The strategic emphasis is on engineering led productivity. Tech Mahindra’s Project Indus takes a domain led approach, with depth in telecom, banking and financial services, and manufacturing. It reflects the company’s view that industry-specific AI deployment will command a premium over horizontal solutions.
There is reason for cautious optimism. Enterprise AI implementation is a genuine, high-value problem, and Indian IT’s combination of enterprise relationships, domain knowledge, and delivery capability gives it a credible foundation. But the risks are real. Frontier AI labs are not standing still. If the implementation layer becomes commoditised, the current platform investments may prove insufficient. The window is not unlimited. Enterprise clients are making long-term partnership decisions now.
The firms that establish credibility and deliver measurable results in the next two to three years will be well positioned for the decade ahead. Those merely rebranding existing services with AI labels will be exposed. India will not lead the next generation of AI. That moment has passed. But a nation of skilled implementers, with deep enterprise relationships and a demonstrated ability to deploy complex technology at scale, is not without a future in the AI age. Whether Indian IT makes the most of that future is the question that will define the sector for the next decade.
The writer is director-Mrikal (AI/Data Center) and a young alumni member, Government Liaison Task Force, IIT Kharagpur. He tweets as @ipravinkaushal. Courtesy: The Statesman
