By Dipak Kurmi
In the mid-to-late 1990s, Harvard Business School professor Clayton Christensen introduced the influential concept of “technology overshoot,” arguing that companies frequently innovate faster than customers’ real-world needs evolve. The result, he warned, is a proliferation of products whose sophistication exceeds what the average user can meaningfully absorb. Nearly three decades later, that framework has found renewed relevance in the age of artificial intelligence. At a recent industry gathering, Nandan Nilekani, speaking in his capacity as chairman of Infosys, revisited Christensen’s thesis and recast it in contemporary terms as a “deployment gap.” In Nilekani’s formulation, the real constraint on the AI revolution is no longer model capability but the limited ability of enterprises to integrate and operationalise these powerful tools. His intervention amounted to both a diagnosis of the moment and a strategic defence of India’s $300-billion software services industry, which now faces one of the most consequential disruptions in its history.
The timing of Nilekani’s remarks is significant. The global technology sector has been rattled by successive releases of advanced AI systems from OpenAI and Anthropic, tools that promise to automate many of the coding, testing and support functions long performed by IT services firms. Market sentiment reflected this anxiety earlier in the month when stocks of both Indian and US technology companies declined following new AI announcements. Yet Nilekani’s presentation struck a deliberately contrarian note. Rather than viewing AI purely as a disruptive force, he argued that the widening gap between AI capability and enterprise readiness creates a large, durable opportunity for firms skilled in implementation, integration and customisation. In this framing, the winners of the next decade may not be only the model builders but also the system integrators who help businesses actually use the technology.
Concrete data from Infosys supports the early contours of this transition. The company disclosed that AI contributed 5.5 percent of its revenue in the December quarter, marking the first time it publicly quantified the scale of its AI-linked business. The figure is modest but symbolically important, signalling that AI is already becoming a measurable revenue stream rather than merely a future bet. On the same day, Infosys announced a strategic partnership with Anthropic to build and deploy advanced enterprise AI solutions across sectors including telecom, financial services, manufacturing and software development. The collaboration reflects a broader pattern emerging across the industry: traditional IT services firms are increasingly aligning themselves with frontier model developers to remain embedded in the AI value chain.
Nilekani identified two structural bottlenecks that could sustain demand for companies like Infosys. The first is the deployment gap itself, the distance between what AI systems can theoretically do and what enterprises are organisationally prepared to implement. The second is a deep hardware and architecture problem rooted in decades of deferred modernisation. Over roughly seventy years, many large corporations chose to layer new technologies atop old ones rather than undertake costly system overhauls. As a result, enterprise environments today often contain an awkward coexistence of 1960s mainframes, 1980s computing stacks and 2000s networking infrastructure, all operating in parallel silos. According to Nilekani, AI-driven tools now provide the first realistic opportunity to rationalise and integrate these fragmented environments at scale. If that proves correct, the modernisation wave could generate significant demand for firms capable of orchestrating complex enterprise transitions.
A further shift Nilekani highlighted concerns the economics of enterprise technology procurement. As AI becomes a larger component of corporate tech spending, the balance of advantage may tilt toward “build” rather than “buy.” In practical terms, enterprises may increasingly seek customised, domain-specific AI systems rather than off-the-shelf software packages. This trend could favour IT services companies that specialise in tailoring solutions to client workflows. Nilekani also pointed to the rise of agentic AI, systems capable of pursuing defined goals with limited human supervision. In his view, the foundational enterprise stack will increasingly become a passive “system of record,” while intelligent agent layers will sit on top, orchestrating workflows across applications. Designing, deploying and maintaining these agentic layers could become a major new services category for firms with deep enterprise relationships.
Despite this optimism, the central strategic question remains unresolved: how much of the traditional IT services value proposition can survive as AI systems grow more autonomous? The concern has intensified with the unveiling of increasingly capable developer-focused models such as OpenAI’s GPT-5.3 Codex and Anthropic’s Opus 4.6, which aim to automate substantial portions of the software development lifecycle. Nilekani is not alone in pushing back against the most pessimistic interpretations. Several Indian IT leaders at the recent AI summit argued that the productivity gains currently visible in the tech sector still depend heavily on humans remaining in the loop, even if the degree of human involvement gradually declines. Analysts at JPMorgan Chase have echoed this nuance, noting that while investors worry AI could compress IT spending, it may be overly simplistic to assume that enterprise-grade software and transformation programmes can be fully automated in the near term.
Policy circles in New Delhi are watching these developments closely, particularly for their potential spillover effects on India’s fast-growing global capability centres. GCCs have become a cornerstone of the country’s services export engine, now accounting for close to 40 percent of services exports and ranking just behind traditional IT services. Multinational corporations across sectors have established large in-house technology and operations hubs in India, attracted by the country’s vast engineering talent pool, relatively lower labour costs, competitive real estate prices and comparatively flexible labour regulations. The scale of the phenomenon is striking: India hosts nearly 1,600 GCCs, and almost one-fifth of the world’s chip designers are based in the country. Major firms maintain significant footprints, with Amazon operating its largest global back office in Hyderabad and Goldman Sachs employing nearly 20 percent of its workforce in Bengaluru and Hyderabad.
Yet the very success of the GCC model has triggered new anxieties. Policymakers worry about a growing overlap between the work performed by in-house capability centres and that delivered by outsourced IT services providers. Both models largely revolve around relocating outsourceable work to India, raising concerns that the expansion of GCCs could come at the expense of domestic IT firms that helped build the ecosystem in the first place. There is also unease about the nature of the work being performed. In many cases, officials believe the activities carried out in GCCs do not yet represent a decisive move into non-outsourceable, high-end innovation that generates intellectual property within India. A commonly cited diagnostic question in policy discussions is whether the parent company’s chief technology officer or key deputies are based in India, or whether cutting-edge patents are being registered locally. On both counts, many GCCs still appear to fall short.
The deeper structural worry is that advances in AI could gradually erode the labour arbitrage advantage that underpins both the IT services and GCC models. If AI systems can automate larger portions of coding, analytics, customer support and back-office processing, the economic logic of large offshore workforces may weaken over time. While this transition is unlikely to be abrupt, improvements in AI capability could progressively compress margins in value-added services that depend heavily on human scale. That possibility is already prompting internal government debates about how to ensure that India captures a greater share of intellectual property creation rather than remaining primarily a destination for execution work.
Even so, there are reasons for measured confidence. Historically, major waves of automation in the technology sector have tended to shift rather than eliminate demand for skilled services. The migration from mainframes to client-server systems, the rise of cloud computing and the mobile revolution each triggered fears of displacement that ultimately gave way to new service categories. The AI transition may follow a similar pattern, particularly if enterprises continue to struggle with integration complexity, regulatory compliance, data governance and domain customisation. These are areas where human expertise and institutional trust remain difficult to fully automate.
What Nilekani’s intervention ultimately underscores is that the AI era will not be defined solely by model breakthroughs but by the messy, uneven process of real-world deployment. The “deployment gap” he describes may prove to be one of the most consequential economic buffers of the coming decade, buying time for services firms while forcing them to evolve rapidly. For India’s technology sector, the challenge is twofold: to harness AI as a productivity multiplier while simultaneously climbing the value chain toward higher-end, IP-rich work. The raised concerns in policy and market circles are therefore not signs of decline but indicators of a sector entering a more complex and demanding phase of its evolution.
(the writer can be reached at dipakkurmiglpltd@gmail.com)
