Manish makes the strategic case brilliantly—Indian IT will survive because enterprises need 2-3 years of integration work before AI actually functions at scale. The data validates it: TCS disclosed $1.5B annualized AI revenue, HCLTech is monetizing $100M quarterly in AI projects, mega-deals are landing (NHS $1.6B over 15 years).
But here's the uncomfortable second-order problem the industry isn't discussing openly: profitability depends on executing the integration faster than the competition can, because the margin is going to collapse to whoever cracks the productivity paradox first.
The real crisis hiding in plain sight:
The 95% Failure Rate is the Constraint
MIT's recent study found that 95% of deployed generative AI projects in enterprises failed to generate profits or reduce costs. UBS data shows 42% of companies abandoned most AI initiatives in 2025. But here's what this really means for Indian IT: the "2-3 year integration window" Singh identifies is actually a problem-solving sprint where the first-mover advantage goes to whoever can systematize the operationalization.
The firms solving data governance, compliance frameworks, and AI-native architectural redesign faster than competitors will establish a moat. The ones that don't will compete on price. In a commodities race with Accenture, Deloitte, IBM GBS, and each other, price always wins—and Indian IT's cost advantage depends on maintaining utilization and headcount efficiency.
The Productivity Paradox is Eroding Margins in Real-Time
Infosys claims 5-40% productivity gains from AI. HCLTech reports $100M in AI revenue but has explicitly shifted pricing to "productivity-linked fees" and "outcome-based models." Translation: the productivity gains aren't creating incremental revenue; they're reducing utilization costs on existing contracts.
Do the math: if a firm delivers 20% productivity gains on a $50M contract through AI implementation, and the client demands that productivity be "passed through," the firm is now delivering the same value with 20% fewer people, earning the same revenue on the contract it won 3 years ago. The only way to grow is to win more contracts. The risk? The client's engineering team learns the system, integrates it into their operations, and eventually internalizes what used to be a billable managed service.
The Vendor Consolidation Trap
Enterprise CIOs are consolidating vendors in 2025 (SAP's CIO Trends 2025 confirms this). They're moving away from project pricing and toward licensing, outcome-based pricing, and "vendor of record" arrangements. This is great for the vendor selected—and catastrophic for the 3-4 losing firms. It's consolidation at the supplier level, which means fewer but larger relationships, but also means the client negotiates from a position of power: "Integrate my AI faster and cheaper, or I'll find a firm that can."
What This Means for the "Indian IT Will Survive" Thesis
Singh is right that Indian IT won't die. But the industry's crisis isn't survival—it's profitability in a race-to-the-bottom integration market.
The firms that win the next 2-3 years will be those that:
Systematize the integration process (turn ad-hoc AI projects into repeatable methodologies)
Capture compliance + governance as a core capability (45% of enterprises blocked by regulatory concerns; this is a moat)
Transition from implementation services to orchestration platforms (stop being billable hours; start being a platform that clients depend on)
The firms that don't will compete on cost, margins will compress further, and the stock multiples will reflect "mature infrastructure services" rather than "AI transformation partners."
The Real Question
Singh identified the demand wave. He got the timing right. But he's written a "survival thesis" when the actual debate happening in board rooms is "how do we monetize the 2-3 year integration window before competition commoditizes it?"
The firms that answer that question—by building genuine intellectual property, replicable processes, and platform dependencies rather than project work—will trade at different multiples than those still selling time-and-materials to the SAP integration market.
TCS's $6-7B investment in AI data centres is a bet on that. Infosys's Topaz platform (12,000 assets, 300 agents, 150 pre-trained models) is a bet on that. HCLTech's explicit pivot to productivity-linked pricing is a bet on that.
The question isn't whether Indian IT survives. It's whether they can escape the margin compression cycle before the 2-3 year window closes and enterprises start building in-house.
P.S. on the Data: If Singh's thesis is that AI creates 2-3 years of billable integration work, then HCLTech disclosing only $100M in AI revenue (3% of total), TCS growing just 2.39% QoY despite "AI being a bigger tailwind than deflation," and Infosys staying silent on AI revenue should concern investors more than it comforts them. These aren't signs of a booming new revenue stream—they're signs of a margin-compressing productivity cycle. The firms solving that paradox faster will separate from the pack. Everyone else becomes a commodity.
the way claude code and now, claude cowork is performing - AI does seem to work outside the box - delivering benefits/efficiency gains absolutely unprecedented. so that makes one condition true pertaining the bear case. aggravating the case further is enterprises going direct, building GCCs and hiring native AI talent to understand, contextualise, and deploy AI internally
if the long-tail of enterprises also happen to understand AI and figure out the preparatory work themselves (hard possibility), Indian IT companies really would not have anything left to sell as AI will takeover pure cost-optimisation and vendor consolidation deals that's paying the bills for the industry
i still believe the mid-cap Indian IT layer, particularly Persistent and Coforge will be increasingly relevant, and TCS + Infosys + HCL Tech in the large-cap. HCL Tech has turned around the business incredibly well since GenAI came out
Manish makes the strategic case brilliantly—Indian IT will survive because enterprises need 2-3 years of integration work before AI actually functions at scale. The data validates it: TCS disclosed $1.5B annualized AI revenue, HCLTech is monetizing $100M quarterly in AI projects, mega-deals are landing (NHS $1.6B over 15 years).
But here's the uncomfortable second-order problem the industry isn't discussing openly: profitability depends on executing the integration faster than the competition can, because the margin is going to collapse to whoever cracks the productivity paradox first.
The real crisis hiding in plain sight:
The 95% Failure Rate is the Constraint
MIT's recent study found that 95% of deployed generative AI projects in enterprises failed to generate profits or reduce costs. UBS data shows 42% of companies abandoned most AI initiatives in 2025. But here's what this really means for Indian IT: the "2-3 year integration window" Singh identifies is actually a problem-solving sprint where the first-mover advantage goes to whoever can systematize the operationalization.
The firms solving data governance, compliance frameworks, and AI-native architectural redesign faster than competitors will establish a moat. The ones that don't will compete on price. In a commodities race with Accenture, Deloitte, IBM GBS, and each other, price always wins—and Indian IT's cost advantage depends on maintaining utilization and headcount efficiency.
The Productivity Paradox is Eroding Margins in Real-Time
Infosys claims 5-40% productivity gains from AI. HCLTech reports $100M in AI revenue but has explicitly shifted pricing to "productivity-linked fees" and "outcome-based models." Translation: the productivity gains aren't creating incremental revenue; they're reducing utilization costs on existing contracts.
Do the math: if a firm delivers 20% productivity gains on a $50M contract through AI implementation, and the client demands that productivity be "passed through," the firm is now delivering the same value with 20% fewer people, earning the same revenue on the contract it won 3 years ago. The only way to grow is to win more contracts. The risk? The client's engineering team learns the system, integrates it into their operations, and eventually internalizes what used to be a billable managed service.
The Vendor Consolidation Trap
Enterprise CIOs are consolidating vendors in 2025 (SAP's CIO Trends 2025 confirms this). They're moving away from project pricing and toward licensing, outcome-based pricing, and "vendor of record" arrangements. This is great for the vendor selected—and catastrophic for the 3-4 losing firms. It's consolidation at the supplier level, which means fewer but larger relationships, but also means the client negotiates from a position of power: "Integrate my AI faster and cheaper, or I'll find a firm that can."
What This Means for the "Indian IT Will Survive" Thesis
Singh is right that Indian IT won't die. But the industry's crisis isn't survival—it's profitability in a race-to-the-bottom integration market.
The firms that win the next 2-3 years will be those that:
Systematize the integration process (turn ad-hoc AI projects into repeatable methodologies)
Capture compliance + governance as a core capability (45% of enterprises blocked by regulatory concerns; this is a moat)
Transition from implementation services to orchestration platforms (stop being billable hours; start being a platform that clients depend on)
The firms that don't will compete on cost, margins will compress further, and the stock multiples will reflect "mature infrastructure services" rather than "AI transformation partners."
The Real Question
Singh identified the demand wave. He got the timing right. But he's written a "survival thesis" when the actual debate happening in board rooms is "how do we monetize the 2-3 year integration window before competition commoditizes it?"
The firms that answer that question—by building genuine intellectual property, replicable processes, and platform dependencies rather than project work—will trade at different multiples than those still selling time-and-materials to the SAP integration market.
TCS's $6-7B investment in AI data centres is a bet on that. Infosys's Topaz platform (12,000 assets, 300 agents, 150 pre-trained models) is a bet on that. HCLTech's explicit pivot to productivity-linked pricing is a bet on that.
The question isn't whether Indian IT survives. It's whether they can escape the margin compression cycle before the 2-3 year window closes and enterprises start building in-house.
P.S. on the Data: If Singh's thesis is that AI creates 2-3 years of billable integration work, then HCLTech disclosing only $100M in AI revenue (3% of total), TCS growing just 2.39% QoY despite "AI being a bigger tailwind than deflation," and Infosys staying silent on AI revenue should concern investors more than it comforts them. These aren't signs of a booming new revenue stream—they're signs of a margin-compressing productivity cycle. The firms solving that paradox faster will separate from the pack. Everyone else becomes a commodity.
the way claude code and now, claude cowork is performing - AI does seem to work outside the box - delivering benefits/efficiency gains absolutely unprecedented. so that makes one condition true pertaining the bear case. aggravating the case further is enterprises going direct, building GCCs and hiring native AI talent to understand, contextualise, and deploy AI internally
if the long-tail of enterprises also happen to understand AI and figure out the preparatory work themselves (hard possibility), Indian IT companies really would not have anything left to sell as AI will takeover pure cost-optimisation and vendor consolidation deals that's paying the bills for the industry
i still believe the mid-cap Indian IT layer, particularly Persistent and Coforge will be increasingly relevant, and TCS + Infosys + HCL Tech in the large-cap. HCL Tech has turned around the business incredibly well since GenAI came out
articulate writing as always, Manish!
Regarding the article, that line about 'systems needing to talk' hit hoome. Exactly!