Verdict: India's $315 billion IT services industry is not collapsing — it is being deflated by artificial intelligence. Profits at TCS, Infosys, and HCLTech rose even as headcount fell in FY26, because the time-and-materials business model that built the industry (bill clients for engineers × hours × rate) gets smaller every time an AI tool makes an engineer 30% faster. The CEOs have a name for it: "AI deflation." If you build for, work in, or sell to Indian IT, the safe place to be is the top of the pyramid — judgment, accountability, and relationships — not the bottom, which is being automated first.
Last verified: 2026-08-05
- Indian IT industry revenue: $315B in FY26 (NASSCOM Strategic Review 2026)
- Infosys: AI revenue 8.2% of total, FY27 guidance trimmed to 1.5–3% (Infosys Q1 FY27 press release, July 23, 2026)
- HCLTech CEO: AI causing 2–3% annual deal deflation, $100M deals shrinking to ~$80M (Moneycontrol, April 21, 2026)
- Entry-level tech openings in India: ~10,000 in June 2026, down 44% YoY (Xpheno via OwnYourCareer, June 2026)
- India AI talent gap: ~1 million unfilled roles by 2027 (NASSCOM–Deloitte report, August 2024)
What is AI deflation in India's IT industry?
AI deflation is when a technology vendor's own productivity gains from AI get passed through to clients as lower prices, shrinking revenue per unit of work even as total output stays the same or rises. It is the opposite of what happened during the cloud migration cycle, where newer technology let vendors charge more for transformation projects. With AI, the more efficient the delivery team becomes, the less billable work there is to sell.
HCLTech CEO C. Vijayakumar named the dynamic on April 21, 2026: AI is causing a 2–3% annual deflation in deal values across HCLTech's traditional services business, with $100 million deals shrinking to roughly $80 million as AI compresses the effort involved. For a sector sitting on a $315 billion base, analysts estimate that AI-led deflation could drag 3–3.5% off industry revenue every year between FY27 and FY29 — that is more than $10 billion a year effectively being returned to clients.
Why this matters for builders: it is the first time in the 25-year history of Indian IT that a productivity technology is structural revenue risk, not a one-quarter blip.
Why was India's IT industry built on the bottom of the pyramid?
The deflation story only makes sense once you see what the industry is actually selling. Indian IT was not built on products, IP, or patents — it was built on selling engineer-hours at cheap rates, in very high volume. The original business model traces back to the 1999 Y2K remediation boom, when US companies needed millions of lines of date-code rewritten manually and did not have enough engineers to do it. Infosys, TCS, Wipro, and Tech Mahindra stepped in, sent Indian engineers to client floors in New Jersey and London, and eventually flipped the equation: brought the work back to India and billed clients for offshore hours at a fraction of the US rate.
The commercial machine that came out of that is the pyramid: thousands of freshers at the bottom, a smaller layer of team leads in the middle, and a tiny apex of architects and client partners at the top. The contract math is engineers × hours × rate = billing, which means the size of the bottom is the revenue. In FY22, Infosys grew revenue 19.7% in constant currency — its fastest in a decade — and hired 85,000 college graduates in a single year. The pyramid was the product.
Which IT jobs is AI actually eating first?
AI is not eating "programming" as a whole. It is eating the specific tasks at the bottom of the pyramid, because those tasks are the ones LLMs are best at. From transcripts and job descriptions across the top five firms, the most-exposed roles share five characteristics:
| Task type | Why AI eats it | Where it lives in the pyramid |
|---|---|---|
| Manual QA / test execution | Pattern enumeration (negative numbers, emojis, empty inputs) is exhaustively definable | Bottom |
| Boilerplate / CRUD code | Structurally repetitive — change only the entity name and address | Bottom |
| Legacy maintenance | AI reads 40,000 lines of old code and surfaces the suspect one faster than a human | Bottom–middle |
| System translation / migration | Piece-by-piece conversion of old systems to new is now an LLM pattern-match | Bottom–middle |
| Code documentation | Explaining what code does for the next reader is the canonical summarization task | Bottom |
What those jobs share is the three properties LLMs are best at: high volume, clear instructions, and a knowable answer. The bottom of the Indian IT pyramid is the widest base of patterned, low-ambiguity software work in the world — which is exactly why it is the first to be automated.
How is AI deflation showing up in the financials?
The 2026 results are the clearest evidence that deflation is not a theory — it is already in the numbers, and it shows up as a scissors pattern: profits and revenue rising alongside shrinking headcount.
| Company (FY26 / latest) | Profit direction | Headcount direction | Source |
|---|---|---|---|
| TCS | Restructuring cost ₹1,268 crore, profits flat-to-up | Net −23,460 employees in FY26 (announced 12,261 layoffs in 2025) | Financial Express |
| Infosys | Operating margin 21.1% (up from 20.8%), net profit +12% YoY in Q1 FY27 | Headcount −8,440 in Q4 FY26; FY27 guidance cut to 1.5–3% | Infosys Q1 FY27 press release |
| HCLTech | CEO publicly flagged 2–3% annual deal-size deflation from AI | Added 5,196 freshers in Q2 FY26 (bucking trend, but flags pricing pressure) | Moneycontrol |
| Top 5 combined | Mixed, mostly stable margins | −7,389 net employees in FY26 (vs +12,718 in FY25) | Moneycontrol / NationPress |
The single most revealing pair of numbers: Infosys reported AI revenue at 8.2% of total on the same call that it trimmed its full-year revenue growth guidance from 1.5–3.5% down to 1.5–3%. Every AI project Infosys delivers makes the rest of its business a little smaller.
Why doesn't productivity become profit for IT services?
This is the part that confuses most engineers. If your team gets 30% more productive with Claude or Copilot, surely that is good for your employer? For a product company like Microsoft, yes — Microsoft sells a product at a price, so a 30% efficiency gain lands as 30% additional margin. For a time-and-materials services company, the math runs the other way:
- You bill the client per engineer-hour.
- The client procurement team knows you have AI now.
- At contract renewal they say: "you need fewer people for the same work, so charge us less."
- Your 30% efficiency gain gets passed to the client as a 30% price cut.
- Your revenue shrinks by 30% on the same scope.
You have made yourself more efficient and shrunk your own business in the process. This is why product firms and sovereign AI infrastructure plays can hold the productivity dividend, while time-and-materials IT services cannot. NASSCOM's own 2026 strategic review pushes the same recommendation — the industry must move toward outcome-based pricing to escape the time trap.
How bad is the entry-level hiring collapse?
The numbers are stark and they are lane-specific. Industry staffing firm Xpheno's June 2026 data, cited across Indian Express, BusinessToday, and Outlook Business, shows:
- Active entry-level tech openings (0–2 years): ~10,000, down 44% YoY
- Total active tech openings: ~93,000 — a 28-month low
- Senior tech openings: down ~67% YoY in some datasets
- IT services fresher intake fell from ~600,000 in FY22 peak to ~120,000 in FY25 — an 80% drop in three years
Top-five IT firms saw 7,389 net headcount decline in FY26. TCS plans to hire 25,000 freshers in FY27 (down from the 40,000–42,000 range of earlier years) and Infosys targets ~20,000 — but those numbers are still a fraction of the FY22 onboarding surge. India's 2,117 Global Capability Centres are the one bright pocket, adding fresher-eligible roles in cloud, data, and cyber.
Is there a shortage of AI talent in India at the same time?
Yes — and the gap is the size of the entry-level losses. A NASSCOM–Deloitte report (August 2024) projects India's AI talent pool will grow from 600,000–650,000 to more than 1.25 million between 2022 and 2027, while the AI market grows at a 25–35% CAGR. Communications Today's February 2026 analysis puts the same gap at ~1 million unfilled roles by 2027 — 2.3 million openings versus 1.2 million qualified professionals.
So in the same country, in the same year, the entry-level IT services fresher count is collapsing while AI-specific roles go unfilled. The bottleneck is not jobs; it is the specific skills mix — applied ML/MLOps, agentic systems, AI governance, and judgement-led architecture — that the existing pyramid was never built to produce.
What is genuinely AI-safe at the top of the IT pyramid?
The apex of the pyramid is safe because LLMs are bad at exactly the things senior architects and client partners do for a living. Specifically, AI cannot:
- Sit in a room with a nervous bank executive and figure out what they actually need vs what they say they need
- Decide which of six defensible architectures fit a client's internal politics, not just the technical spec
- Take responsibility when a system fails at 3am and face the client the next morning
- Build the multi-quarter trust that lets a CFO sign a ₹100 crore contract
That is the work that grows when the bottom shrinks. Sovereign AI strategy advisors and GCC architects are seeing their compensation bands inflate precisely because the bottom — repetitive, patterned, low-ambiguity work — is being absorbed by tools, leaving the senior judgement work more valuable, not less.
How should an engineer build a defensible career during AI deflation?
Four moves, in order:
- Build the judgement layer first. Learn to solve problems without AI tools, so that when you have the tools you can spot the wrong, lazy, or inefficient answer. This is the difference between an AI-native engineer and an AI-user.
- Replace certificates with proof of work. Recruiters report 89% of engineers self-identify as "AI-ready" but only 19% have ever built an AI or ML system in production (Scaler–CMR study, 2026). GitHub repos, Kaggle entries, and Tableau dashboards outperform a 9 CGPA on an AI-screened resume.
- Get fluent in client communication. The future belongs to senior engineers who can sit in front of a ₹100 crore client, present the nuances of a project, and earn the trust that closes the contract. Engineering colleges in India do not teach this.
- Move to outcome-led engagements, not hourly ones. If you sell hours, AI deflation is a tax on you. If you sell outcomes, AI productivity is pure margin. This applies to freelancers, agencies, and small businesses automating with AI as much as to TCS.
What this means for you
For engineers and freshers in India: the IT services entry lane is structurally smaller; that is not a hiring slowdown, it is a business model shift. The leverage is in moving up the pyramid — judgement, communication, and AI-native delivery — not in waiting for the 2022 hiring boom to return. For builders and small businesses: AI deflation is your tailwind. The same compression that shrinks Infosys's billing shrinks your cost to build software. For investors and analysts: watch the deflation drag on IT services margins separately from AI revenue contribution; they are independent forces.
FAQ
Q: Is AI deflation going to wipe out India's IT industry?
A: No. India's tech industry revenue is forecast at $315 billion in FY26 (NASSCOM), up 6.1% YoY, and operating margins at Infosys held at 21.1% in Q1 FY27. AI deflation is shrinking revenue growth and headcount, not the industry itself — profits are rising while people count falls. The drag is estimated at ~$10 billion a year (3–3.5% of base) between FY27 and FY29, which is meaningful but not existential.
Q: How much has fresher hiring fallen in Indian IT?
A: IT services fresher intake fell from roughly 600,000 in the FY22 peak to around 120,000 in FY25 — an 80% drop — and active entry-level tech openings hit ~10,000 in June 2026, down 44% year-on-year (Xpheno). TCS plans to hire ~25,000 freshers in FY27, versus the 40,000–42,000 range in earlier years.
Q: If AI makes engineers more productive, why doesn't that become profit for IT firms?
A: Because services companies bill clients per engineer-hour. When engineers get 30% faster with AI, clients see the productivity gain and renegotiate the contract down by ~30%. The efficiency gain lands in the client's pocket, not the vendor's. That is HCLTech's CEO publicly flagged 2–3% annual deal-size deflation.
Q: Is the AI talent gap in India real?
A: Yes. NASSCOM and Deloitte project AI talent demand growing from 600,000–650,000 to more than 1.25 million between 2022 and 2027, while supply grows more slowly — leaving roughly 1 million roles unfilled by 2027 on some estimates. Genuinely AI-native engineers (who can reason about the architecture underneath the tool) are scarce; casual ChatGPT users are not.
Q: Which IT jobs are most at risk from AI in India?
A: The most-exposed roles are at the bottom of the pyramid: manual QA execution, boilerplate CRUD code, legacy maintenance, system migration, and code documentation. They share three properties — high volume, clear instructions, knowable answers — which is exactly what LLMs do well. Senior architecture, client-facing, and judgement-led roles are least exposed.
Q: Should an engineering student in India still study IT?
A: Yes, but not the way the 2010s curriculum teaches it. The entry-level volume job is not coming back at FY22 levels; the apex — judgement, communication, AI-native architecture — is inflating. Build proof of work, learn to reason about AI system failures, and get fluent in client communication. Those are the four skills that compound when the bottom of the pyramid automates.
Discussion
0 comments