India’s big AI bet: tax holidays to lure global AI compute — and the infrastructure test that follows

• TL;DR
• India’s 2026 budget offers foreign cloud providers a tax holiday through 2047 for revenues from services sold outside India when those services run from Indian data centers.
• The move pairs fiscal incentives with expanded electronics, semiconductor and rare‑earth support, and it has already crystallized multibillion‑dollar commitments from hyperscalers and domestic partners.
• Power, water, land and permitting remain the real determinants of whether India becomes a cost‑effective hub for AI infrastructure — not the tax code alone.

What the cloud tax holiday actually means for AI compute in India

The headline: foreign cloud providers can earn tax‑free revenues through 2047 on services sold to customers outside India if those services are hosted in Indian data centers. Sales to Indian customers are treated differently — they must be routed through locally incorporated resellers and will be taxed domestically. The budget also proposes a 15% cost‑plus safe‑harbour — a predictable 15% markup floor for related‑party data‑center services that helps multinationals price internal transfers and reduces transfer‑pricing uncertainty.

Finance Minister Nirmala Sitharaman (paraphrase): India will exempt certain revenues earned by foreign cloud firms from taxation through 2047 if those services are delivered from Indian data centers, while domestic sales must be routed locally and taxed.

Put simply: the government is trying to attract export‑oriented AI compute hosted onshore while steering domestic commercial consumption through Indian entities. That combination favors hyperscalers as anchor tenants and seeks to thread a policy needle between foreign investment and domestic capture.

Who’s already signing up — and why it matters

Big commitments are already public and sizable enough to change the market dynamic:

• Google: announced roughly $15 billion to build an AI hub and expand data‑center infrastructure.
• Microsoft: planning about $17.5 billion of investment through 2029.
• Amazon: committed an additional $35 billion by 2030 (bringing its broader India investments toward ~$75 billion across businesses).
• Digital Connexion (Reliance + Brookfield + Digital Realty): targeting ~ $11 billion to build a 1 GW (gigawatt) AI‑focused campus in Andhra Pradesh by 2030.
• Adani Group: signalled up to $5 billion of potential investment alongside Google.

Rohit Kumar (paraphrase): Data centers are being elevated from backend infrastructure into a strategic business sector that can pull private capital — but execution hurdles like power and land access will determine how fast this translates into usable capacity.

The government pairs these incentives with expanded industrial policy: the Electronics Components Manufacturing Scheme was increased to ₹400 billion (~$4.36B), there’s a five‑year tax exemption for certain tooling and equipment suppliers in bonded zones, and the India Semiconductor Mission enters a new phase focused on equipment, materials, IP and supply‑chain localization. The aim is to stack upstream manufacturing and downstream compute in the same geography.

Why the tax holiday doesn’t automatically mean cheaper AI

Tax holidays are a powerful fiscal carrot, but energy and water are the real fuel. Data‑center operating costs are dominated by electricity, cooling and real estate. Shortfalls or unpredictability in any of these inputs — or long permitting cycles — can wipe out the value of a long tax holiday.

• Power: India’s data‑center power capacity is small today (~1 GW) but industry estimates project it could exceed 2 GW by 2026 and 8 GW by 2030 backed by >$30B in capital. Still, power reliability, grid constraints and commercial tariff levels vary sharply by state.
• Water: Many cooling designs rely on water; scarcity and local permits can constrain load factors and add costs or force alternative cooling strategies.
• Land & permits: Large contiguous sites with ready utilities are limited. State‑level clearances and local community engagement can slow projects for years.
• Supply chains & geopolitics: Hardware and critical minerals supply remains geopolitically sensitive; establishing local tooling and rare‑earth corridors helps, but dependencies remain.

Sagar Vishnoi (paraphrase): Data‑center capacity could jump dramatically by 2030 thanks to private investment; the tax holiday essentially bets on global Big Tech while India builds its own tech champions.

Strategic tradeoffs for business leaders

This policy creates several predictable tradeoffs:

• Anchor tenants vs. local champions: Hyperscalers anchor capacity quickly but may dominate export‑oriented compute, potentially sidelining smaller Indian cloud providers unless those firms specialize or partner.
• Short‑term fiscal gains vs. long‑term operating costs: Tax relief lowers headline costs for capital owners, but energy/water/land premiums change the total cost of ownership (TCO) calculus.
• Export orientation vs. domestic market protection: Routing domestic sales through local resellers aims to keep onshore revenue and jobs, but it may compress reseller margins and complicate sales models.

Scenarios to model (conservative, base, aggressive)

• Conservative: Permitting and utilities lag. Capacity grows slowly to ~2 GW by 2026, energy tariffs remain high, and firms rely on niche deployments and hybrid architectures. India becomes an export niche rather than a global scale hub.
• Base: Central incentives plus state cooperation speed projects; capacity reaches ~4–6 GW by 2030. Hyperscalers host major training runs while some domestic providers carve out enterprise edge and specialized services.
• Aggressive: Coordinated policy, utility upgrades, and local supply‑chain scale deliver >8 GW by 2030. India competes cost‑effectively with regional hubs for large model training and a vibrant hardware ecosystem emerges.

Practical checklist for executives planning AI infrastructure or investments in India

1. Build a TCO model that includes energy price per kWh, expected PUE (power usage effectiveness), water costs or alternative cooling OPEX, land acquisition timelines, and a permitting risk premium.
2. Evaluate partnership structures (JV, reseller, captive) to comply with the reseller routing rule and to capture local market share without sacrificing export economics.
3. Negotiate long‑term power and water contracts or secure renewable PPAs to stabilize costs and meet ESG goals.
4. Map workloads by sensitivity — place latency‑sensitive or India‑domestic workloads with local resellers; schedule large training jobs for export‑oriented, tax‑favored clusters.
5. Ask about state‑level incentives and timelines early. Central policy unlocks capital, but states deliver land, clearances and utilities.
6. Factor supply‑chain timelines for GPUs, custom racks, and power gear; local semiconductor/tooling incentives help but won’t eliminate lead times immediately.

KPIs and “questions to ask your cloud partner”

• Effective $/GPU‑hour including energy and cooling costs.
• PUE and average data‑center utilization.
• Average permitting time from LOI to commercial run‑rate.
• Water withdrawal permits and constraints.

Questions to ask:

How do you model energy and water price risk over 5–10 years?
Request guarantees or long‑term contracts where possible and quantify the escalation clauses.

What reseller and routing mechanics will affect our commercial billing and margins?
Clarify who books revenue, who invoices, and how transfer pricing will be managed under the 15% safe‑harbour.

What is the expected timeline from permit to production for your site?
Ask for historical data on similar projects in the target state and contingency plans for delay.

How this fits into the global AI infrastructure map

India’s offer lands in a global race where the U.S., EU, Singapore and the Middle East are also beefing up capacity and incentives. India’s long tax horizon (through 2047) is unusually generous and signals a commitment to host export‑oriented AI workloads for decades. But cost competitiveness will hinge on how quickly the country can solve the hard engineering problems — reliable, low‑cost power; sustainable cooling; and predictable land and permitting pipelines.

The policy does more than chase immediate deployments; it attempts to anchor an entire value chain: from semiconductor tooling and electronics manufacturing to AI training farms and downstream services. If implemented successfully, that stack could reduce hardware lead times and create local usage patterns that feed domestic AI for business and automation initiatives.

For boards and C‑suite leaders: treat this as an invitation to begin formal scoping, not a signal to rush blindly. Model TCO, stress‑test supply chains, and open conversations with hyperscalers and local partners now — because the tax code will help only if the lights stay on and the water keeps flowing.

Next step: commission a focused India compute TCO pilot (30–60 days) that benchmarks expected $/GPU‑hour, identifies two candidate states for deployments, and lists potential JV partners or reseller structures. The window for advantageous deals is opening fast — start scoping before site economics harden.