AI-driven employment displacement in India Report Interpretation

Overview

This India strategy report examines whether AI is creating enough employment to counter job displacement. Bernstein concludes that the evidence so far points to weak net job creation: advertised AI roles are often relabelled legacy work, while formal payroll growth, IT-company headcount and layoff data do not show an AI-driven hiring boom.

Core views

Bernstein begins by challenging the common claim that AI will replace some jobs but create more and better ones. The report argues that this outcome depends on adoption speed, demand elasticity, how productivity gains are distributed, regulation and workers’ ability to adapt. Unlike prior technologies, AI is increasingly directed at cognitive, analytical, creative and managerial work across the income pyramid. The key issue is therefore not only aggregate job creation, but which jobs disappear, which roles emerge, who receives the productivity gains and whether the transition is fast enough. The report tests the job-creation narrative through three indicators: formal employment, AI job postings and aggregate IT-company headcount. EPFO data, available through July 2025, show no meaningful acceleration in formal hiring since generative AI emerged. Net new payroll additions in FY2025 were 6.3% below the FY2023 peak, while net services payroll additions have edged down over the past three years. The 22–28 age cohort saw a 14.3% decline in net additions. In contrast, lower-end employment categories recorded increases, including 5% in cleaning and sweeping services and 6.2% in building and construction, which Bernstein does not view as evidence of AI-linked job creation. The job-posting analysis is the report’s central evidence. A broad search yielded about 40,000 AI vacancies, and more than 100,000 when logged-in and near-match results were included, but the report identifies extensive duplication. Reviewing approximately 36,500 postings across Delhi, Gurugram, Mumbai, Pune, Chennai, Hyderabad and Bengaluru produced roughly 30,000 unique job IDs, a 17% reduction from the initial count. About 85% of postings were from traditional IT-services firms. Within software, which represented 64% of listings, 55% offered annual pay below INR 1 million, with the largest group in the INR 600,000–1 million range—compensation that the report considers consistent with legacy software employment rather than a distinct AI labour market. Bernstein then classified postings by job descriptions and skill requirements into regular IT work, IT work using AI tools, substantial AI work, regular hardware work, data work for AI projects and consulting. It deliberately used a lenient definition of substantial AI work: a traditional-looking role could qualify if it mentioned skills such as RAG, LangChain, agents, prompt engineering, fine-tuning, Docker or Kubernetes. Even under this generous approach, only 54% of postings qualified as genuinely AI-related. At least 13,000 positions, or 43% of the roughly 30,000 unique postings, appeared to be conventional IT work relabelled as AI; Bernstein estimates that fewer than one-third—about 32% using a midpoint between its lenient and stringent tests—involve actual AI work. It also notes that annotation and data-preparation roles can be temporary and may themselves later be automated. Posting dates reinforce the caution: 67% were listed within one month, but 20% were more than two months old, 10% more than six months old, and nearly 500 dated from more than four years before ChatGPT’s arrival, showing that an AI keyword does not establish an AI-created job. Aggregate employment data from traditional IT firms do not, in Bernstein’s view, validate a hidden hiring surge. The top 10 public IT employers averaged roughly 60,000 headcount additions per quarter in 2021 and 2022, but their average net addition over the latest four quarters was negative. The report contrasts this with more than 67,000 Indian tech layoffs over the past five years reported by layoffs.fyi. Bernstein’s estimates raise startup and undisclosed-firm layoffs above 71,600; including TCS and estimated layoffs at Amazon, Microsoft and Oracle takes the total to nearly 88,000, and the report suggests the actual total could be at least 100,000 in recent years. It highlights TCS’s announced 12,000 layoffs as an example, while noting that companies often do not explicitly attribute cuts to AI. The report’s broader conclusion is that AI currently looks less like a broad employment cycle than a redistribution of economic gains. Productivity gains can enable firms to operate with fewer employees; Bernstein questions the assumption that companies will reinvest those gains in headcount rather than retain them or distribute them to shareholders. If displacement reduces middle-income demand, lower demand may further reduce incentives to hire or expand output. The report expects gains to accrue disproportionately to capital owners and highly skilled workers, describing a barbell-like outcome in which the top benefits while the middle faces deflationary pressure. It argues that India’s policy and institutional challenge is to convert productivity gains into wider employment and prosperity rather than allowing opportunity to narrow.

Analysis framework

Bernstein evaluates the AI-jobs thesis using a three-part evidence test: EPFO formal-payroll trends, a bottom-up review of online AI vacancies, and quarterly headcount data from major public IT employers. It removes duplicate job postings, classifies descriptions by the degree of actual AI work, compares compensation and employer type, and places those findings against layoff estimates and broader demand-and-productivity logic.

Methodology notes

• Industry AnalysisVolume-price decomposition

  Job-posting composition analysis by employer type, pay level, job description and required skills.

  The report breaks apparent AI hiring into its underlying components to distinguish substantial AI roles from conventional IT, data, hardware and consulting work carrying AI-related labels.

• Other

  Three-test employment validation using EPFO payrolls, vacancy analysis and IT-company headcount trends.

  Bernstein uses independent labour-market and company-level indicators to test whether AI vacancies correspond to a genuine acceleration in net employment.

Key data

• India tech job lossesMore than 100,000 in recent yearsReported as announced or estimated losses across startups and IT companies.
• AI postings from traditional IT services firms~85%Based on the report’s employer-type analysis.
• Unique AI job postings reviewed~30,000Derived from ~36,500 postings across seven cities after removing duplicates.
• Conventional IT roles repackaged as AIAt least 13,000; 43%Of roughly 30,000 unique postings.
• AI-related roles under lenient classification54%The report’s generous definition; its more realistic assessment is fewer than one-third of postings.
• Estimated actual AI-work share~32%Midpoint implied by the report’s lenient and stringent classification approaches.
• FY2025 formal payroll additions-6.3% versus FY2023 peakEPFO net new payroll additions.
• Net additions for ages 22–28-14.3%Decline cited for the key younger workforce cohort.
• Software AI listings below INR 1 million annual pay55%Software represented 64% of all AI job listings.
• Top 10 public IT employers~60,000 average quarterly headcount additions in 2021–22; negative average in the last four quartersUsed to assess whether AI hiring is exceeding displacement.

Impact & implications

Bernstein says AI-related productivity gains may be captured mainly by shareholders and highly skilled workers rather than translated into broad hiring. It sees potential pressure on wages, middle-income consumption and headcount-based industries if adoption expands without stronger demand creation, worker adaptation and institutional support.

Risks

• AI adoption could displace white-collar roles across coding, research, design and middle management while limiting new hiring.
• Temporary annotation and data-preparation roles may face further disruption as AI capability advances.
• Layoffs and weaker wage growth could depress middle-income demand, creating a feedback loop that further reduces hiring incentives.
• The distribution of AI productivity gains may widen inequality if benefits flow primarily to capital owners and highly skilled workers.