Indian IT services companies are training at an unprecedented scale to prepare their workforces for an AI-led future. Yet even as hundreds of thousands of employees are reskilled in data and generative AI, the sector continues to chase a much smaller pool of specialised talent—data architects, machine learning engineers, and AI platform builders—whose skills cannot be created quickly or cheaply.
IT leaders indicate this dual-track strategy is deliberate.
During Tata Consultancy Services’ Analyst Day 2025 in December, executive director and chief operating officer Aarthi Subramanian said TCS has “doubled down on advisory and consulting talent” across priority areas such as cybersecurity, cloud, and enterprise solutions, while positioning these hires closer to customers. More than half of TCS’ experienced hires today come with what she described as ‘next-generation skills’—part of a broader talent transformation to make the company “ready for our AI future.”
At Infosys, chief delivery officer Satish HC, during the Q2 earnings call, said roughly 10% of the company’s top technology talent is now engaged in highly innovative AI solution-building work. This group includes what Infosys calls “forward-deployed engineers”—specialists embedded within client organisations to tackle complex industry problems.
He cited a global logistics client where such teams co-created a real-time AI platform that processes about 400 million messages a day with sub-minute latency, delivering $1.5 million in immediate benefits, $8 million in annual savings, and a 12% reduction in customer service calls.
Hiring volumes, however, have not collapsed.
In the July-September quarter, Wipro reported a lower attrition rate and a rise in net headcount, increasing by 2,260 to 2.35 lakh due to campus placement and improved resource utilisation. During the earnings call, CHRO Saurabh Govil said that with strong bookings in the first half, he expected to continue hiring both laterally and from colleges.
Infosys also hired more than 12,000 freshers during H1 FY26. The total headcount rose to 332,000, and the IT company kept attrition at 14.3%, underscoring that workforce expansion has not disappeared despite automation.
But look closer, it’s more about how headcount links to growth.
HCLTech chief executive C Vijayakumar told analysts during the Q2 earnings call that revenue has grown 4–5% in recent years without a corresponding rise in employee numbers, pointing to early signs of non-linearity. He said revenue per employee rose 1.8% year-on-year, and that gap could widen as the company builds more platforms, intellectual property, and SaaS offerings using agentic tools.
Even a 1% quarterly differential between revenue and headcount growth, he argued, would compound into a meaningful structural change over time.
HCLTech has also trained staff at scale, creating “black belt” cohorts across technology streams to strengthen readiness for AI delivery and embedding senior data and AI specialists in over half its priority accounts.
Structural Employment Issue
Industry analysts say these narratives explain only part of the picture.
In a conversation with AIM, research and advisory firm Forrester’s principal analyst Biswajeet Mahapatra described the most in-demand AI roles as a niche, elite layer that sits outside traditional IT services career paths. This cohort—data scientists, machine learning engineers and AI researchers with advanced degrees from institutions such as the IITs or IISc, and experience in product companies or startups—accounts for just 1–2% of the workforce.
Despite their small numbers, they have an outsized impact as they design high-value AI systems that automate work at scale and shape client strategy.
Mahapatra argued that Indian IT faces both a skills problem and a structural employment problem. Most employees are trained in legacy technologies, while traditional time-and-material delivery models are poorly suited to iterative, outcome-driven AI work.
As AI projects break the delivery model first, pressures then emerge around compensation and capability, making it harder to attract and retain specialists without squeezing margins. Large-scale upskilling, he added, can even increase attrition by making employees more marketable, an unintended effect analysts now factor into workforce models.
He also observed, “Gig, fractional, and advisory models are emerging as viable compromises, enabling firms to access specialised expertise without bearing long-term cost burdens.”
According to a Mint report, Tata Consultancy Services is considering gig-style work arrangements for hard-to-retain specialists, including data architects and data scientists, allowing flexible, part-time engagement even as they work elsewhere, as the Indian IT sector faces talent shortages and AI-driven uncertainty.
Recruitment firm Xpheno sees a similar split. Co-founder Kamal Karanth said most AI reskilling programmes by Indian IT companies create “AI-aware” talent for deployment and services, not deep specialists.
“The split is very clear. While industry narratives often refer to hundreds of thousands of AI-skilled professionals, most of this workforce is AI-aware rather than AI-operational,” he said.
Roles such as AI architects or platform engineers require three to five years of sustained, hands-on work on production systems, which cannot be fast-tracked through internal training alone.
Xpheno estimates India has only 18,000–20,000 professionals with meaningful enterprise AI exposure, of whom just 4,000–5,000 qualify as true specialists. Attrition at senior AI levels, he said, runs at 18–20% over 12–24 months, reflecting an acute demand-supply imbalance.
From the product side, Bengaluru startup NimbleEdge’s co-founder and chief technology officer Neeraj Poddar drew a sharp distinction between current GenAI upskilling and the foundational expertise required to build state-of-the-art agentic systems. Much of today’s training, he said, focuses on using tools such as LLMs or AI coding assistants to improve productivity.
In order to create products which use AI agentic systems well, he said, one needs to understand the core principles behind them and build frameworks to evaluate the success of these AI-driven products.
“I believe it’s a misnomer calling many agentic products ‘AI wrappers’ (that sit between the user and the AI model) as you often need to understand and implement core AI techniques ranging from multi-agent orchestration to trajectory guidance, to evaluation frameworks, observability, and runtime optimisations depending on where your AI system is running,” Poddar noted.
This, he observed, requires a deep understanding of AI and machine learning principles to implement them in real-life scenarios.
The result is a persistent paradox. Indian IT firms are spending heavily to make their vast workforces AI-ready, while simultaneously competing for a tiny cadre of specialists who can design and scale AI systems.
The post Indian IT’s AI Conundrum: Broad Workforce Upskilling But Only 2% Shape Outcomes appeared first on Analytics India Magazine.
