Mera joota hai Japani, lekin AI Hindustani

beyond server farms and data centres, and forging India's AI sovereignty through policy reform

Server Pool by Ulysse Gerkens & FARI / Better Images of AI / CC BY 4.0

Agriya Khetarpal

GCPP 43

The Takshashila Institution

April 2026

A policy analysis based on the Takshashila framework applied to the 21-year tax holiday for foreign cloud service providers utilising data centres as introduced in the
Union Budget, FY 2026-2027

Policy recommendations

• Sovereignty metrics
  • Contributions to open-weights models
  • Funding for Indian FOSS AI communities
  • Procurement of domestic open-architecture hardware (RISC-V)
• Local value additions
  • A minimum % workforce from Indian institutions on DCIM-aligned projects
  • Quotas for SMEs
  • Published upskilling programmes
• Environmental key performance indicators (KPIs)
  • PUE < 1.3
  • >80% renewable energy sourcing by 2035
  • Mandatory closed-loop cooling in water-stressed zones

Tiered tax system based on performance via a sovereignty and sustainability scorecard

Recommendation 1

Non-compliance: graduated tax clawbacks, but with a phase-in over 36 months with transparent reporting templates and industry consortia support

• Representatives from the central government, state governments, Ministry of Power, Ministry of Jal Shakti, environmental scientists, and representatives from affected marginalised communities
• Uniform national baseline for safety, security, and environmental impact
• Review state schemes for alignment with the national AI strategy, and hold a veto authority over sovereignty/ecology violations
• States that voluntarily align receive matching central grants

National Data Centre Council as a statutory body under MeitY

Recommendation 2

Karnataka's software ecosystem + Gujarat's renewable grid = a national compute corridor

Mandating open interoperability, local value addition, and environmental accountability that is not left to competitive state discretion

Goals

A co-investment fund for sovereign AI following on from Recommendation 1

Recommendation 2

Investment areas

• Domestic FOSS AI infrastructure, such as open-weights models, indic datasets, and AI governance frameworks by think tanks and research institutions
• RISC-V toolchain development aligned with advancing DIR-V programme and DHRUV64 chip design
• Cybersecurity auditing frameworks for ensuring data centre security standards are domestically controlled
• Regional AI innovation clusters  to redistribute innovation dividends

15%

of foregone tax revenue
to be directed to such a fund

Managed by an independent board with transparent investment criteria: interoperability with global standards such as the EU AI Act, and Indian governance oversight

• Rapidly emerging destination with 25% CAPEX support and power subsidies actively attracting new investment
• New Renewable Energy Policy 2025 removes capacity caps and enables hybrid renewable + BESS that directly addresses data centre power demands
• Faces a groundwater crisis with 85% of districts affected, and thus a case study in policy blind spots

• Home to Bengaluru (one of) India's largest existing data centre clusters with high concentrations of hyperscaler presence
• A mature state-level framework: a dedicated Data Centre Policy 2022–27 with sustainability targets
• Also, the site of India's most visible water stress crisis, i.e., a live cautionary tale of demand outpacing enforcement

Why did I compare Karnataka and Gujarat?

Without these amendments, the tax holiday accelerates infrastructure build-out while leaving India structurally dependent on foreign-controlled AI platforms.

Mera data bhi Hindustani

This policy is designed in the public interest,
but fails to define what the public interest is.