Mere data ke darbar mein

Agriya Khetarpal

LucknowFOSS 2.0

04 April 2026

India's AI data centre rush, and the FOSS enthusiast's hidden stake

The Environmental Impact of Data Centres in Vulnerable Ecosystems

Gloria Mendoza / Better Images of AI / CC BY 4.0

Follow along!

• A dedicated, secure facility that houses computing systems, servers, storage, and networking equipment
• ERP and IT infrastructure, web hosting and Content Delivery Networks (CDNs), edge computing, data storage, disaster recovery, and more
• Early example: US military's Electrical Numerical Integrator and Computer (ENIAC)

A glossary of terms

• A specialised data centre facility designed for computationally intensive, high-performance tasks of training and running inference for AI models.
• They are optimised for the parallel processing demands of AI workloads, typically utilising hardware such as AI accelerators (e.g., GPUs, TPUs) and high-speed interconnects.

A glossary of terms

• Massive cloud service providers – Amazon Web Services, Microsoft Azure, and Google Cloud Platform, and co.
• They offer highly scalable, on-demand computing, storage, and AI services. They operate colossal global data centres designed for high-performance data processing.
• Enable businesses to rent infrastructure and maintenance, and allow them to focus on efficiency and cost management.

A glossary of terms

A glossary of terms

A glossary of terms

• A sustainability metric used to determine the energy efficiency of a data centre
• It is expressed by dividing the total amount of power entering a data center by the power used to run the IT equipment within it, as a ratio, with the overall efficiency improving as the quotient tends towards 1.0.

A glossary of terms

•  A sustainability metric developed by The Green Grid to measure the efficiency of water usage in a data centre
• The ratio of annual water usage (litres) for cooling and humidification to the energy consumed by IT equipment (kWh)
• A lower WUE value indicates higher efficiency, supporting efforts to reduce environmental impact and improve water management

A glossary of terms

A glossary of terms

A headline every week

Data centres, vs. AI data centres

Leaf-and-spine network designs

Policies of the nation vs. states

State-level data centre policies (curated)

National-level data centre policies

• 150 billion litres consumed in 2025 → 358 billion litres by 2030
• A 100 MW facility: up to 800,000 litres per day
• 60–80% of Indian data centres projected to face high water stress this decade
• It is commonly known that any cities are already water-scarce!
• Example: Bengaluru's Devanahalli hub has groundwater extraction at 169% of permissible limits

Water woes

• 13 TWh consumed today → 57 TWh projected by 2030 (fivefold increase)
• India's grid: ~70% coal-powered, still
• A single hyperscale facility consumes as much power as 50,000–100,000 homes, which is a small city!
• PUE and WUE disclosures are not mandatory for operators of data centres

Pangs of electricity

Data colonialism

• India generates 20% of global data, stores only 3% domestically
• This gap is framed as a post-colonial wound, in that Indian data enriching foreign platforms, foreign economies
• "Build data centres here, keep the data home"
• "Atmanirbhar Bharat", i.e., a self-reliant India gets applied to digital infrastructure

our data, on our soil

• The five largest data centre operators in India's pipeline are Microsoft, Google, Amazon, Reliance, Adani/AdaniConneX
• Three of these are US corporations operating under the US CLOUD (Clarifying Lawful Overseas Use of Data) Act
• A 21-year tax holiday that returns zero corporate tax to the Indian exchequer
• Example: "AWS Mumbai" is not the same as Indian-controlled infrastructure

 But... who owns the infrastructure?

Data sovereignty is not achieved by the physical location of a server.

Contemporary cautionary tales: U.S.

Contemporary cautionary tales: the EU

In India

Dissecting the Lucknow AI City announcements

Extraordinary claims require extraordinary evidence

• No EIA has been filed on the Parivesh Portal yet. EIA Notification 2006 doesn't mention data centres.
•  UP has 3,823 MW of installed solar capacity, against its 22,000 MW target for 2027. That's ~17% of the target!
• STT GDC (Singapore-majority owned, Tata minority), Sify Technologies at HCL IT City
• no published engineering plan, no PPA for solar, no MW figure for renewable capacity, no green hydrogen feasibility study
• WUE in hot climates runs 1.5–2.5 L/kWh, for a 100 MW facility in Lucknow's summers (regularly 45°C+), that is 2–6 million litres of water consumed per day. Microsoft's Asia-Pacific operations report 1.65 L/kWh, and Lucknow is hotter than most AP locations.

Some realities to care about

Why should you care, as someone
interested in FOSS?

• NxtGen, a Bengaluru-based cloud provider serving clients like the Election Commission of India, built its enterprise cloud service SpeedCloud on OpenStack, an OSS infrastructure platform
• FOSS adoption enabled them to offer cloud services at prices ~80% lower than large proprietary competitors
• 1:50 Proprietary software to FOSS ratio
• Self-hosting Mattermost for internal communications — keeping company data off third-party SaaS platforms entirely
• "We would choose to use FOSS 10 out of 10 times"

On the "greenness" of software

Accountability as a FOSS value

• If the code is auditable, the infrastructure should be too
• Open data on energy and water is a FOSS value, and not just an environmental one
• Communities bearing the real costs deserve the same transparency we demand from software

Ending notes

Thank you!

"Water Day", by @Malicartoonist on Twitter