India's three largest IT services companies have looked at the same AI infrastructure opportunity and walked away in three different directions. Infosys has ruled out building data centers entirely, calling the capital requirements "quite massive." TCS is building over a gigawatt of AI-ready capacity with a $1 billion backing from private equity firm TPG. HCLTech is investing ₹3,500 crore ($365 million) in its own AI data centers while taking a $150 million stake in Indian AI startup Sarvam. The split reveals a deeper question every technology-dependent business now faces: when AI compute becomes a strategic input, do you own the infrastructure or rent it?
Verdict: For India's IT majors, the asset-light approach (Infosys) protects margins and return on equity in the short term, but the infrastructure-first approach (TCS, HCLTech) positions them to own the full AI value chain — from GPUs to applications — as sovereign AI demand grows. For your business, the same fork exists: own compute when control and data sovereignty matter, rent it when speed and flexibility matter more.
TL;DR
- Infosys rejected AI data centers after a board-level review — capital intensity doesn't fit its asset-light model (CNBC-TV18).
- TCS launched HyperVault with $1 billion from TPG, targeting 1+ GW of AI-ready capacity (Tata newsroom).
- HCLTech committed ₹3,500 crore for up to 50 MW of capacity, plus a $150 million stake in Sarvam AI (HCLTech press release).
- All three are growing AI revenue fast — TCS at $2.6B annualized, Infosys at 8.2% of revenue, HCLTech at ~$600M run rate.
- Last verified: 2026-08-04 · Pricing/figures are volatile — re-check quarterly.
Why Did Infosys Say No to AI Data Centers?
Infosys will not build or operate data centers for AI workloads. Outgoing CEO Salil Parekh confirmed the decision on the company's June 2026 earnings call, saying the proposal was reviewed by both management and the board before being rejected.
"We've looked at that, and we are not planning to go into the data center business. We've looked at, from our perspective, our balance sheet, the potential cash required to build a data center at scale, and then other requirements, which are essentially around energy projects. It's something that we have decided, at this stage, will not benefit us." — Salil Parekh, CEO, Infosys (CNBC-TV18 interview)
The rationale is capital discipline. AI data centers require enormous upfront investment in land, power infrastructure, cooling systems, and GPU clusters — plus ongoing energy costs that can run into hundreds of millions annually at scale. Infosys reported $5.08 billion in Q1 FY27 revenue with a 21.1% operating margin and free cash flow of $955 million (Infosys Q1 press release). Even at that cash generation, a multi-billion-dollar data center build would drag on return on equity for years before reaching meaningful utilization.
Instead, Infosys is concentrating on the software and services layer above the infrastructure: helping enterprise clients select and deploy AI models, build AI agents, and modernize existing systems through its Infosys Topaz platform. AI accounted for 8.2% of total revenue in Q1 FY27, with double-digit sequential growth (Infosys Q1 press release).
What Is TCS Building with HyperVault and TPG?
TCS is taking the opposite bet. In November 2025, the company announced HyperVault — an AI-ready data center business targeting more than one gigawatt of capacity in India. Private equity firm TPG committed up to ₹8,820 crore ($1 billion) to the venture, with a combined commitment of up to ₹18,000 crore ($2.15 billion) from both partners over several years (TPG press release; TechCrunch).
TPG's investment is facilitated through TPG Rise Climate and its Global South Initiative (a private equity strategy launched in partnership with ALTÉRRA), plus its Asia Real Estate business. TPG's final shareholding in HyperVault is expected to be between 27.5% and 49%, meaning TCS retains majority control while reducing its capital outlay (Tata newsroom).
TCS is also working with AMD on an AI-ready data center blueprint supporting up to 200 megawatts of capacity. The company reported annualized AI revenue of $2.6 billion for the June 2026 quarter, up 13.6% sequentially from $2.3 billion, and said it has completed more than 5,500 client engagements (CNBC-TV18; Economic Times).
The HyperVault bet is partly a sovereign-AI play. Much of the new capacity serves clients — including hyperscalers, AI-native firms, and enterprises — that want AI compute located in India for data residency and regulatory reasons.
How Is HCLTech's Approach Different from TCS?
HCLTech is building data centers too, but with a different architecture. Where TCS's HyperVault is infrastructure-first (land, power, cooling, and compute for hyperscalers and large AI firms), HCLTech is pursuing a "full-stack" model — owning the data center, the GPUs, the models, and the applications layered on top.
In July 2026, HCLTech announced it would invest up to ₹3,500 crore ($365 million) to establish AI data centers in India with the potential to scale to 50 MW of capacity (HCLTech press release). Days later, Reuters reported that HCLTech would invest 142.57 billion rupees ($1.48 billion) to set up its first AI data center in Bhubaneswar, Odisha, in partnership with Sarvam AI and the Odisha state government — including a 5,000-seat technology center starting operations by 2028 (Reuters).
The key differentiator is HCLTech's $150 million acquisition of a 10.5% stake in Sarvam AI, a Bengaluru-based startup building sovereign AI models for Indian languages (Financial Express; TechCrunch). Sarvam's foundation models — including open-weight LLMs in 30-billion and 105-billion parameters — give HCLTech an indigenous model layer to run on its own infrastructure, targeting government and enterprise clients that want AI computation kept within India.
HCLTech CEO C Vijayakumar framed the full-stack approach explicitly:
"Our whole value is in delivering full-stack AI services, which means it's the data centre, it's the GPUs, it's the models, it's the applications that we will deliver on top of it." — C Vijayakumar, CEO, HCLTech (Livemint)
How Do the Three AI Strategies Compare?
| Dimension | Infosys | TCS | HCLTech |
|---|---|---|---|
| Data center ownership | No — explicitly ruled out | Yes — HyperVault, 1+ GW target | Yes — up to 50 MW initially, 50 MW scaling |
| External capital | N/A | $1B from TPG (27.5–49% stake in HyperVault) | Exploring silicon OEM, debt, equity financing |
| Model layer | Partner-led (uses client-selected models) | Partnerships with Anthropic, Mistral, Google Cloud | Owns 10.5% of Sarvam AI (indigenous models) |
| Primary client | Enterprise AI services (Topaz platform) | Hyperscalers + AI-native firms + enterprises | Government + sovereign AI + enterprise full-stack |
| AI revenue | 8.2% of total revenue (~$417M/quarter) | $2.6B annualized run rate | ~$600M annualized advanced AI run rate |
| AI revenue growth | Double-digit sequential | 13.6% QoQ | ~20% QoQ |
| Flagged risk | Opportunity cost of missing infrastructure wave | Capex could weigh on ROE if AI spending cools | 2–3% AI-led revenue deflation on legacy services |
| Strategic logic | Asset-light: protect margins, serve the layer above | Infrastructure play: own the compute, sell to all | Full-stack: own compute + models + applications |
Sources: Infosys Q1 FY27, TCS Q1 FY27, HCLTech advanced AI, CNBC-TV18, HCLTech CEO.
Is Infosys's Caution Smart Capital Discipline or a Missed Opportunity?
Both, depending on the time horizon. The case for Infosys's approach rests on three pillars:
Margin preservation. Infosys holds a 21.1% operating margin and 116.5% free cash flow conversion — both among the best in Indian IT (Infosys Q1). Heavy data center depreciation and energy costs would compress those margins for years.
Execution focus. Infosys is already growing AI revenue at 8.2% of total revenue without owning infrastructure. Its Topaz platform, agent deployment business, and model-selection services generate revenue from the layer above — where margins are higher and capital intensity is lower.
Downside protection. If AI infrastructure spending cools off — or if hyperscalers build their own capacity in India (Google announced a 1 GW data center in Visakhapatnam as part of a $15 billion investment) — companies that poured billions into physical assets could be left with stranded capacity.
The case against Infosys's approach:
Sovereign AI demand is real and growing. India's data center capacity is expected to grow from 1.8 GW to 5–7 GW by 2030, with a significant portion for AI workloads (HCLTech press release). Clients — especially government entities and regulated industries — increasingly want compute located in-country.
The value chain is shifting down. As AI services become commoditized, owning the infrastructure underneath becomes a differentiator. TCS's partnerships with Anthropic and Mistral and its 1,996 AI-related patents (CNBC-TV18) give it leverage across the stack that a services-only competitor cannot match.
First-mover advantage in client lock-in. Enterprises choosing an AI infrastructure provider make multi-year commitments. By the time Infosys reverses course — if it ever does — the relationships may already be locked.
What About Wipro and Tech Mahindra?
The split isn't just three-way. Wipro is focusing on turning services into reusable AI platforms — it launched a dedicated AI-Native Business and Platforms unit in April 2026, building on an earlier $1 billion AI investment commitment (CNBC-TV18). Tech Mahindra is leaning on its telecom expertise and partnering with hyperscalers rather than owning large-scale compute.
Five companies, five strategies — but they cluster into two camps: those betting that physical AI infrastructure is the next growth wave (TCS, HCLTech), and those betting that staying asset-light protects them from it (Infosys, Wipro, Tech Mahindra).
What Does This Mean for Your Business?
The same strategic fork applies to any company deciding how to use AI:
Own compute when data sovereignty and control matter. If you operate in regulated industries, government, or handle sensitive customer data, keeping AI workloads on infrastructure you control — or within a sovereign jurisdiction — may be a requirement, not a preference. India's data localization push is one example; the EU AI Act's transparency requirements are another (see our guide to EU AI Act Article 50 compliance).
Rent compute when speed and flexibility matter more. Cloud-based AI APIs let you test models, pivot architectures, and scale elastically without multi-year capital commitments. For most small businesses, this is the right default — see our comparison of free AI API providers in 2026.
The "services above infrastructure" play scales. Infosys's bet is that the layer above — model selection, agent deployment, system modernization — grows faster and more profitably than the infrastructure beneath. If your business is AI-adjacent (consulting, integration, custom tooling), the same logic applies.
AI-driven productivity cuts both ways. HCLTech flagged 2–3% annual revenue deflation from AI on its legacy services (Economic Times). If AI is compressing your existing revenue, building new AI-native revenue streams isn't optional — it's survival. Our analysis of AI and India's IT jobs crisis covers the structural dimension of this shift.
India is becoming an AI infrastructure battleground. With Google's 1 GW Visakhapatnam facility, Reliance's green-powered Jamnagar facility, TCS's HyperVault, and HCLTech's Bhubaneswa center, the country's AI infrastructure landscape is shifting fast — creating opportunities for businesses that want local compute. This connects to broader India tech hub dynamics we track in our Chennai GCC and India GCC strategy coverage.
FAQ
Q: Why is Infosys not building AI data centers? A: Infosys reviewed the opportunity at the board and management level and concluded that the capital requirements — including land, power infrastructure, cooling, and energy costs — are too large to fit its asset-light financial model. CEO Salil Parekh said the cash needed to build at scale would be "quite massive" and would not benefit the company given its balance sheet (CNBC-TV18).
Q: How much is TCS investing in AI data centers? A: TCS and TPG have committed up to ₹18,000 crore ($2.15 billion) combined, with TPG investing up to ₹8,820 crore ($1 billion) for a 27.5–49% stake in the HyperVault subsidiary. TCS targets 1+ GW of AI-ready data center capacity in India (TPG).
Q: What is HCLTech's full-stack AI strategy? A: HCLTech is building data centers (up to 50 MW initially, ₹3,500 crore), acquiring a 10.5% stake in Sarvam AI ($150 million) for indigenous model capabilities, and combining these with its existing AI orchestration and services platform to deliver infrastructure, models, and applications as an integrated offering — primarily targeting government and sovereign-AI clients (HCLTech; Livemint).
Q: Which Indian IT company has the most AI revenue? A: TCS reported $2.6 billion in annualized AI revenue for Q1 FY27 (June 2026 quarter), the highest among Indian IT firms. Infosys reported AI at 8.2% of revenue (~$417 million annualized), and HCLTech reported an advanced AI run rate of approximately $600 million (CNBC-TV18; Infosys).
Q: Is sovereign AI driving data center demand in India? A: Yes. India's data center capacity is projected to grow from 1.8 GW to 5–7 GW by 2030, driven by AI workloads, data localization requirements, and government demand for compute kept in-country. Google, Reliance, TCS, and HCLTech are all building GW-scale facilities (HCLTech press release).
Q: Will Infosys change its mind about data centers later? A: Parekh framed the decision as "at this stage," leaving the door open. The company named Ashiss Kumar Dash as CEO Designate effective July 2026, succeeding Parekh on April 1, 2027 — so the strategic stance could evolve under new leadership (TopNews).
Discussion
0 comments