The AI Revolution Needs Plumbers After All
How Indian IT learned to stop worrying and sell the AI shovel
For the last two years, generative AI was going to kill Indian IT. The argument seemed almost self-evident — if machines can write code, a $250 billion industry built on getting humans to write it cheaper has nowhere left to go. Investors acted accordingly, and the sector has since underperformed the broader market by 30% or more.
The industry has spent this year pushing back. It cut margins, restructured workforces, built platforms, and told clients that AI has not transformed their enterprises because their enterprises are a 30-year accumulation of SAP, Oracle, Workday and middleware that was never designed to talk to anything. And finally, Indian IT is who you call when systems need to talk to each other.
Despite all the hype, generative AI is moving slowly. Less than 15% of organizations are meaningfully deploying the new technology at their firms, according to investment bank UBS. And the narrative about the Indian IT dying is beginning to recede. Investment group CLSA titled a note this month, “Discussion moving beyond AI,” a sign that the existential panic has subsided enough for analysts to return to debating deal pipelines and vertical demand.
Enterprise AI has underwhelmed, though of course not from lack of enthusiasm or capital. Industry players say the tech remains inadequate for regulated industries where someone has to sign off on the output. They cite “workslop,” weak governance and high error rate as reasons the gap between AI as boardroom theatre and AI as functioning software remains so wide.
In the meantime, the Indian IT companies are reporting gains from the same force that was supposed to disrupt them.
Infosys now calls AI-led volume opportunities a bigger tailwind than the deflation threat, a reversal from 2024, and orderbooks held steady in the third quarter even as pricing pressure filtered through renewals. Infosys expects its own orderbook to grow more than 50% this quarter, anchored by an NHS deal worth $1.6 billion over 15 years.
The AI capex cycle has been concentrated among a handful of hyperscalers and labs, while the Fortune 500 is still figuring out what to do with what they have bought. Indian IT is betting that figuring out what to do is billable work. Channel checks suggest a two-to-three year window of preparatory work – data cleanup, cloud migration, system integration – before enterprise-wide AI becomes feasible, and that window is where Indian IT plans to earn its keep.
The IT industry has always been reactive to new technology, late to consulting and early-stage advisory but quick to capture implementation spend once the experiments end and the plumbing needs building. The firms believe AI will follow the same arc: a hype phase they mostly miss, followed by a deployment phase where scale, client relationships and tolerance for unglamorous work become valuable again.
TCS, which cut its headcount by 2%, is spending on the “less fashionable” layers – a 1GW data-centre network in India, an indigenous telecom stack, a sovereign cloud – alongside platforms called WisdomNext and MasterCraft. It acquired Coastal Cloud, a Salesforce advisory firm, for capability it did not want to build from scratch.
HCLTech cut margins by 100 basis points, redirected savings toward specialist hiring, and became one of first large systems integrators to partner with OpenAI. The firm announced this week that it had acquired Jaspersoft for $240 million and Belgium-based Wobby to boost agentic AI capabilities.
Infosys has taken a different route, building an asset library rather than data centres. It runs 2,500 genAI projects, has deployed 300 AI agents in its own operations and claims productivity gains of 5-40% depending on service line. Its AI-suite, Topaz, holds 12,000 assets, 150 pre-trained models and 200 agents for code generation, IT operations and billing; Cobalt holds 35,000 cloud assets and 300 industry blueprints.
Leadership now describes the systems integrator as an “orchestrator” – not building models but making them function inside client businesses, where function means plugging into SAP, Oracle and Salesforce without hallucinating key details. Forward deployed engineers sit inside key accounts to identify use cases and move pilots toward production.
Wipro has built vertical platforms including AutoCortex, WealthAI and PayerAI and signed a sovereign AI deal with Nvidia, though the company faces stiff competition in vendor consolidation where productivity baselines already run at 15% before AI shows up. Tech Mahindra has invested in sovereign LLMs and a one-trillion-parameter domain-specific model, hoping India’s national AI push provides differentiation.
Among smaller IT firms, Persistent is reporting what it claims is early evidence of AI-driven productivity, with revenue growing at double digits while headcount stays flat. LTIMindtree has assembled an AI team of over 1,000 to build what it calls a “learning transfer” model to carry lessons from one deal to the next.
IT budgets have grown about 8% annually for the past five to six years for the industry, and the expectation is that the trend continues, with AI taking a larger share alongside cybersecurity and cloud migration.
Large customers want productivity passed through when they adopt genAI, and vendors concede the hit arrives at renewal rather than all at once, which means revenue growth may not return to mid-to-high single digits until FY28 or FY29. IT services are forecast to shrink from 38% of enterprise tech spending in 2018 to 25% by 2029, even as the absolute market grows to $1.3 trillion.
Valuations have not collapsed: the Nifty IT still trades at over 6% premium to the Nifty versus a 10-year average of 10%, and at an over 15% discount to the Nasdaq, close to historical norms. One risk for 2026 is that if the AI-led global tech rally fades, Indian IT would likely suffer rather than benefit, because relative performance has tracked broader tech sentiment even as the underlying business case has diverged.
The IT companies are not claiming victory. The argument is narrower: that the preparatory work AI requires – data cleanup, integration, compliance, tuning – creates enough billable hours to offset what automation takes away, that the middleman remains necessary for different reasons than before.
The bear case assumed AI would work out of the box, that enterprises would deploy it themselves, that Indian IT would have nothing left to sell. Two years in, AI does not work out of the box, enterprises have found it difficult to deploy it themselves, and the firms that were supposed to be dead are still hiring specialists and winning deals.