From reselling AI to owning it — the pivot inside India's IT services majors

Thematic essay — week of July 12–19, 2026

In the same nine days, TCS and HCLTech each did two things that used to be separable: they told investors, in dollar figures, how much AI is now worth to their business, and they committed capital to own the compute underneath it rather than lease it from a hyperscaler. TCS put a $2.6 billion annualized AI run rate on the board and walked into a full-stack technology mandate at JFK's New Terminal One in the same week. HCLTech disclosed $171 million in advanced-AI revenue, up 62% year-on-year, and approved up to ₹3,500 crore for its own AI data centres — while already sitting inside Sarvam's cap table as its lead Series B investor. Two of India's largest services firms, in one earnings season, converged on the same bet: that the money in enterprise AI won't stay in reselling other people's models, and that owning a slice of the stack — model equity, captive compute, or both — is now worth the capital outlay. Whether that bet pays off is not yet knowable. What is knowable, and worth tracing precisely, is the shape of the wager itself.

The events

TCS's Q1 FY27 print, July 9. TCS opened the fiscal year with revenue of ₹72,275 crore, up 13.9% year-on-year, and — the number the company chose to lead with — an annualized AI revenue run rate of $2.6 billion, AI-deal revenue up more than 13% year-on-year, and $9.5 billion in total contract value including an $800 million AI-led transformation deal with SKF. As the July 12 digest noted, profit grew at roughly a third the pace of revenue that quarter — AI-led growth is arriving with margin pressure attached, not behind it. TCS also named an expanding partnership bench: Anthropic, Mistral, Google Cloud, ServiceNow. Multi-lab provisioning, not exclusivity with any one frontier vendor.

TCS becomes the technology partner for JFK's New Terminal One, July 14. Five days later, TCS announced it would run the full technology stack for the new terminal at JFK Airport — passenger processing, AI-driven IT operations, infrastructure management, application support, and cybersecurity — as part of the Port Authority's $19 billion terminal transformation. No contract value was disclosed. What matters about this deal is what it is not: it is not a disclosed AI-revenue line item, it is a marquee physical-infrastructure client naming TCS across an entire technology estate, with AI operations built in rather than retrofitted. As the July 14 digest put it, the breadth of the mandate is the more durable commercial signal than any individual AI capability inside it.

HCLTech's Q1 FY27 print, July 13. HCLTech reported $171 million in advanced-AI revenue, up 62.1% year-on-year and 10.6% quarter-on-quarter, against record Q1 net-new bookings of $2,407 million. This was the second Indian major to put an AI number on the board in the same earnings season TCS opened. As the July 15 digest put it, "Two of the largest Indian services firms have now each put an AI number on the board in the same earnings season, which begins to standardize a disclosure that did not exist a year ago" — even though "advanced-AI revenue" remains a company-defined metric with no standard the way, say, cloud revenue eventually acquired one.

HCLTech commits up to ₹3,500 crore to owned AI data centres, July 13. Announced the same day as the Q1 results, HCLTech's board approved capex for AI data centres scaling to 50MW through a new subsidiary, underpinning a new full-stack AI offering. CEO C Vijayakumar's framing, as the July 16 digest recorded: "The biggest opportunity is not to rent AI, but to own the full stack." ₹3,500 crore and 50MW is mid-sized against the multi-gigawatt commitments elsewhere in India's data-centre buildout that same month — but it is a genuine departure from how Indian SIs have historically run, which is asset-light.

HCLTech's prior stake in Sarvam, June 15 (context). The data-centre capex does not stand alone. HCLTech led Sarvam's $234 million Series B at a roughly $1.5 billion valuation with a $150 million commitment, taking, as the June 15 digest described it, "an equity position in the model layer rather than a reselling relationship on top of it." Read forward from June to July, the same firm now holds equity in a foundation-model company, is building its own compute, and is disclosing services revenue that runs through both. That is three layers of the same stack under one balance sheet.

Emergent's $130M Series C, July 15 — the application-layer counterpoint. Not a services-firm story, but the same week's evidence that the alternative bet — building products on top of the frontier models rather than owning infrastructure beneath them — is also getting funded at scale. Emergent, a Bengaluru agentic-coding startup, raised at a $1.5 billion valuation on a reported ~$120 million annualized revenue run rate. As the July 16 digest framed it, it is "the application-layer counterpart to the foundation-model unicorn milestone the ecosystem hit weeks earlier — a different theory of where Indian AI value accrues, tested in the same funding cycle." Elevation Capital's $500 million ninth fund, closed July 13 and explicitly earmarked for seed and Series A application-layer AI, is fresh capital on the same side of that bet.

The mechanism: why own the stack when you can rent it

Indian IT services has run asset-light for three decades. The model: bill for people and process, buy compute and software from someone else, keep the balance sheet light and the margins predictable. AI-led delivery is testing that model on two fronts at once, and the two data centre and equity moves this week are direct responses to each.

Margin capture. When a services firm delivers an AI-led engagement running entirely on a hyperscaler's GPUs, a share of the deal's economics flows straight to the cloud provider before the SI's own margin is calculated. Owning captive AI compute — HCLTech's stated logic — keeps more of that spread inside the firm's own books, provided utilization materializes. This is the same economic logic that has pushed hyperscalers themselves toward custom silicon (Google's TPUs, Amazon's Trainium): whoever owns the compute layer captures the fattest margin in the AI stack, and everyone downstream of Nvidia has been trying to move up a layer for two years.

Residency. A second, India-specific driver sits underneath the margin argument. Regulated-sector workloads — banking, insurance, government — increasingly cannot leave the country, and a services firm with domestic AI-DC capacity can bid for that work directly rather than routing it through a foreign hyperscaler's India region. HCLTech's own framing ties the data-centre investment to exactly this: capacity that can serve residency-bound clients "hyperscaler-dependent delivery cannot reach," in the words of the July 16 coverage. This is the same residency logic that shows up elsewhere in the archive that week — MeitY's reported preference for sovereign cyber-defence models over OpenAI and Anthropic for critical infrastructure, and Amazon's own ₹60,000 crore Telangana build, sited specifically to hold data in-country. Residency is turning into a competitive variable that both foreign hyperscalers and domestic SIs are now capitalizing against, from opposite directions.

The equity route versus the infrastructure route. HCLTech is running both plays simultaneously, and they are not the same bet. Leading Sarvam's Series B buys exposure to the model layer — if Sarvam's foundation models succeed, HCLTech holds equity upside regardless of whether HCLTech itself delivers the services around them. Building owned data centres buys exposure to the infrastructure layer — a bet that pays off through utilization and services margin, independent of whose models run on the racks. TCS, by contrast, has taken neither route as visibly; its posture is the multi-lab provisioning bench (Anthropic, Mistral, Google Cloud, ServiceNow) plus large managed-services wins like JFK. Two of India's largest SIs, same quarter, two different theories of where to place a bet inside the same stack.

The honest caveat, present in every disclosure above: "advanced-AI revenue" and "AI run rate" are company-defined metrics with no accounting standard behind them yet, the way "cloud revenue" took several years to acquire a common definition across vendors. Two companies choosing to disclose in the same quarter is a real signal about where the sector believes the market's attention is going — but it is not yet a like-for-like comparison, and neither company has published what it excludes from the number.

The comparable: this is the vertical-integration pattern every AI-adjacent layer is running

The TCS/HCLTech pivot is not sui generis to Indian services. It is the same structural move that has played out one layer up the stack all year. Frontier labs that started as pure model shops have been buying or building consumer surfaces (OpenAI's ChatGPT Work, Anthropic's Claude Cowork) so the unit of value shifts from API calls to finished deliverables. Hyperscalers that started as pure infrastructure have been building models (Gemini, Nova) so they capture application-layer margin too. And now services firms that started as pure delivery shops are reaching down into infrastructure and equity so they are not permanently the layer that gets squeezed from both directions.

The memory-market comparable sharpens the stakes. SK Hynix's Nasdaq debut the same week — a $26.5 billion offering, a 13% pop, a $1.27 trillion market cap on HBM memory demand — is capital markets pricing a chokepoint layer of the AI stack at a scarcity premium. Every Indian data-centre build, including HCLTech's ₹3,500 crore commitment, inherits its input costs from that market without negotiating leverage over it. Owning compute in India does not exempt a services firm from the pricing power concentrated three layers upstream in HBM supply; it only changes which margin, further downstream, the firm gets to keep.

Globally, the vertical-integration trade has an uneven record. Cloud providers that built their own chips took years to reach cost parity with merchant silicon, and several AI labs that expanded into consumer hardware or enterprise-product surfaces are still proving out the unit economics of the deliverable-not-conversation model OpenAI and Anthropic are both now selling. HCLTech's ₹3,500 crore is small enough, and early enough, that the read has to stay conditional: this is a bet placed, not a bet that has paid off.

HCLTech's captive data centres are also entering a market where the country's compute-siting incentives are moving in the same direction from the state-policy side. The Uttar Pradesh Cabinet approved a Data Centre Policy 2026 the week before, targeting more than 2 GW of AI-ready, GPU-based capacity and over ₹2 lakh crore in investment — targets, not commitments, but a signal that states are now competing on data-centre siting the way they once competed for factories. And Amazon's ₹60,000 crore commitment to an AWS complex in Telangana, foundation-stoned on July 15, is the hyperscaler leg of the identical wave — a US cloud provider building India-region capacity for exactly the residency-bound and pay-as-you-go workloads HCLTech's captive build competes for on the SI side. Three different capital sources — a state government's incentive framework, a US hyperscaler's balance sheet, and an Indian SI's board-approved capex — converging on the same conclusion in the same ten days: that domestic AI compute capacity is worth building now, not renting from wherever it happens to already exist. HCLTech is not choosing to own compute in a vacuum; it is one bidder among several who have all reached the same conclusion in the same month.

Where it lands

The near-term test is legibility, not size. If Infosys and Wipro publish comparable AI-revenue figures in their own Q1 FY27 results later in July, TCS and HCLTech's unilateral disclosures become a sector disclosure norm rather than two companies' choices — the kind of standardization that eventually forces a common definition, the way cloud revenue reporting did. If neither larger peer follows, the disclosure stays a TCS/HCLTech-specific signal, and the comparison across the sector remains apples to a mix of other fruit.

The medium-term test is utilization, and it has a specific address: HCLTech's Q2 FY27 results in October. The question is not whether the new AI-DC subsidiary is operational — it is whether the advanced-AI revenue line, by then, reflects deals actually delivered on the owned infrastructure, or whether the capacity sits under-booked while services revenue continues flowing through hyperscaler compute regardless. Capital committed and capacity built are the easy parts; a services firm's asset-light history is precisely why the harder part — filling owned capacity at a margin better than renting would have offered — is unproven until a quarter's numbers say otherwise.

The larger question the window raises but does not answer: does AI-led delivery ultimately concentrate value at the infrastructure layer (favoring HCLTech's capex route), the model-equity layer (favoring HCLTech's Sarvam stake), the managed-services-breadth layer (favoring TCS's JFK-style mandates), or the application layer entirely outside the SI channel (favoring Emergent and the capital Elevation Capital is deploying toward it)? The honest answer this week is that India's two largest SIs are each hedging across more than one of those layers at once, which is itself informative — neither company is confident enough in a single theory of where AI value in services will land to bet on just one.

Sources

• 2026-07-09. TCS Q1 FY27 results press release →.
• 2026-07-12 (digest). India AI Digest 2026-07-12 — TCS Q1 FY27 AI run rate, SK Hynix Nasdaq debut →.
• 2026-07-14. TCS press release, "TCS Becomes Technology and Innovation Partner to New York's New Terminal One at JFK Airport" (via PR Newswire) →.
• 2026-07-14 (digest). India AI Digest 2026-07-14 — TCS JFK partnership →.
• 2026-07-13. HCLTech Q1 FY27 results release (via PR Newswire) →.
• 2026-07-13. HCLTech press release on its full-stack AI offering and AI data-centre investment →.
• 2026-07-15 (digest). India AI Digest 2026-07-15 — HCLTech Q1 FY27 advanced-AI revenue →.
• 2026-07-16 (digest). India AI Digest 2026-07-16 — HCLTech AI data-centre capex, Emergent Series C →.
• 2026-07-03. HCLTech's $1.14 billion AI-led operating-model deal, as covered in the July 7 digest →.
• 2026-06-15. Sarvam's $234 million Series B led by HCLTech, as covered in the June 15 digest →.
• 2026-07-15. TechCrunch, "Indian AI coding startup Emergent becomes a unicorn just over a year after launch" →.
• 2026-07-13. Entrackr, "Elevation Capital closes $500 mn India fund to back AI startups" →.
• 2026-07-10. CNBC, "SK Hynix rises 13% in Nasdaq debut" →.
• 2026-07-06. Uttar Pradesh Cabinet's Data Centre Policy 2026 approval, as covered in the July 12 digest →.
• 2026-07-15. Amazon's ₹60,000 crore AWS Telangana data-centre commitment, as covered in the July 18 digest →.