Cloud Ecosystem in India

1. At a Glance

• Cloud computing means using computing resources (storage, servers, software, processing) over the internet, as and when needed, instead of owning servers. Its benefits are lower cost, faster deployment, scalability, backup and recovery, and service continuity [1].
• India is building a domestic cloud ecosystem for data sovereignty, meaning national control over where sensitive data is stored and processed. It rests on five pillars: MeghRaj, data centres, the IndiaAI Mission, skilling, and policy (Cloud Selection Framework, DPDP Act) [1].
• Why it matters for UPSC: the topic connects e-governance (GS-II), IT infrastructure, AI and cybersecurity (GS-III), and data privacy (GS-II, Article 21 linkage). It is a recurring "Digital Public Infrastructure" theme.

2. Why in the News

• PIB published an explainer on India's domestic cloud ecosystem on 29 Oct 2026, during ESTIC-2026 (27–29 Oct 2026) [1].
• MeitY issued a Cloud Selection Framework in March 2026 to guide Ministries and Departments on choosing a cloud based on how sensitive their data is [1].
• BHASHINI moved to an Indian cloud run by Yotta Data Services, reported in Feb 2026 [1].
• The IndiaAI compute pool grew to 45,000+ GPUs [1][5].

3. Background & Evolution

• Rationale: speed up e-governance, optimise government ICT spending, and keep sovereign control over citizen data [1].
• Milestones:
• 2014: MeitY launches MeghRaj, the National Cloud Computing Initiative, also called GI Cloud [1][2].
• 2021: MeitY and NASSCOM launch FutureSkills PRIME for emerging-technology skills [1].
• 2023: DPDP Act, 2023 enacted. Its data residency provisions encourage local data centres and cloud services [1][9].
• 2024: Union Cabinet approves the IndiaAI Mission (outlay ₹10,371.92 crore). The original target was 10,000+ GPUs through PPP [1][7].
• Nov 2025: DPDP Rules, 2025 notified, with an 18-month phased compliance timeline [8].
• ~Dec 2025: 26 CSPs empanelled, 2,170 Ministries/Departments on MeghRaj [2].
• Feb 2026: 38,000+ GPUs and 1,050 TPUs available under IndiaAI [6]. BHASHINI migrates to an Indian cloud [1].
• Mar 2026: Cloud Selection Framework issued. Guidelines for Procurement of Cloud Services issued [1][3].

• Sep 2026: 30 empanelled CSPs and 2,330 departments on MeghRaj 2.0 [1].

• Related earlier infrastructure: NIC's National Data Centres and the State Data Centres [1].

4. Core Static Facts

5. Multi-Dimensional Analysis

Economic

• The pay-as-you-use model turns capital spending into operating spending and lowers the cost of public ICT [1].
• Data centre capacity grew from 375 MW (2020) to about 1.57 GW. The projected ~8 GW by 2030 points to large investment in power, land and construction [1][5].
• Subsidised GPUs (under ₹100 per hour) make compute affordable for start-ups and academia [6].

Geopolitical / Strategic

• Data sovereignty: keeping sensitive government and citizen data under Indian jurisdiction reduces exposure to foreign laws and supply-chain risks [1].
• BHASHINI's move to Yotta is an example: language datasets, AI models and citizen interactions all stay in India [1].
• A planned secure national GPU cluster of 3,000 next-generation GPUs is for sovereign and strategic AI applications [6].

Legal / Constitutional

• The DPDP Act, 2023 and the Rules, 2025 put informational privacy into practice. This follows from privacy being recognised under Article 21 [8][9].
• Data residency and conditional cross-border transfer shape where CSPs build their infrastructure [1].

Scientific / Technological

• Cloud is the base layer for AI. The GPU and TPU pools under IndiaAI are offered as a cloud-style shared facility [1][6].
• Security assurance comes from the STQC audit against the ISO 27001/27017/27018/20000 family [2].

Governance / Administrative

• The Cloud Selection Framework makes data classification a formal step. Sensitive workloads go to NIC or State Data Centres, and less critical ones can go to a hybrid setup with empanelled CSPs [1].
• MeitY empanelment through AMBUD and procurement through GeM standardise purchasing and reduce vendor-by-vendor appraisal [3][4].
• Bottlenecks (analytical): data centres need a lot of power and water, State Data Centres differ in capacity, and lock-in to large private CSPs is a risk.

Environmental (analytical)

• Growing towards ~8 GW of data centres raises questions about electricity demand, renewable energy sourcing and water for cooling [1].

6. Recent Developments (last 12–18 months)

• Nov 2025: DPDP Rules, 2025 notified. Full operationalisation of the Act with an 18-month phased compliance period [8].
• ~Dec 2025: 26 CSPs and 2,170 departments on MeghRaj [2].
• Feb 2026: 38,000+ GPUs and 1,050 TPUs under IndiaAI. A 3,000-GPU secure national cluster announced [6].
• Feb 2026: BHASHINI migrates to Yotta's Indian cloud [1].
• Mar 2026: MeitY Cloud Selection Framework and Guidelines for Procurement of Cloud Services [1][3].
• Aug 2026: data centre capacity at 1.57 GW. 237 projects have used IndiaAI compute (93.18 lakh GPU-hours) [1][5].
• Sep 2026: 30 CSPs empanelled and 2,330 departments on MeghRaj 2.0 [1].
• 27–29 Oct 2026: ESTIC-2026. PIB explainer on the cloud ecosystem released on 29 Oct [1].

7. Prelims Hooks

• MeghRaj, the National Cloud Computing Initiative (GI Cloud), was launched by MeitY in 2014 [1].
• MeghRaj offers IaaS, PaaS and SaaS on a distributed cloud architecture [1].
• CSPs are empanelled by MeitY after an audit by the STQC Directorate [2].
• STQC audit standards: ISO 27001, 27017, 27018 and 20000 [2].
• The CSP empanelment portal is AMBUD, and it accepts applications 24x7x365 [4].
• 30 CSPs were empanelled as of September 2026 [1].
• 2,330 government departments were on MeghRaj 2.0 as of September 2026 [1].
• NIC runs 4 National Data Centres and 24 Mini Data Centres [1].
• The Cloud Selection Framework (MeitY, March 2026) gives priority to NIC and State Data Centres [1].
• IndiaAI Mission outlay: ₹10,371.92 crore [1].
• 45,000+ GPUs are onboarded under the IndiaAI Mission [1].
• Data centre capacity: 1.57 GW (Aug 2026), projected at nearly 8 GW by 2030 [1].
• FutureSkills PRIME is a MeitY–NASSCOM initiative from 2021 [1].
• The cloud models allowed in government procurement include the Government Community Cloud, procured via GeM [3].
• BHASHINI migrated to an Indian cloud run by Yotta Data Services [1].
• "Serverless" computing still uses servers; the provider manages them [1].

8. Data Kept in India Is Not Always Under India's Control

• Where data sits and who controls it are two different questions
• Data residency means the data is physically stored inside India. The DPDP Act and the Cloud Selection Framework mostly deal with this question [1].
• Control is a separate question: which country's law can force the company that runs the server to hand the data over.

• A foreign-owned cloud company that runs a data centre in India still has to obey the law of its home country too. So data can sit in India and still be reachable from abroad.

• The Budget is inviting the foreign companies in

• Budget 2026–27 offers a tax holiday (a period with no tax on profits) to foreign cloud providers that serve customers from a MeitY-notified data centre in India [14].
• This brings in investment and capacity quickly. But it also means more of India's cloud will be run by companies whose head offices answer to other governments.

• This is why the Framework keeps the most sensitive data with NIC and State Data Centres, which the government owns, and sends only less critical work to private providers [1].

• Sovereignty stops at the chip

• The IndiaAI compute pool is built from GPUs that private partners rent to the government under PPP (public–private partnership) [1][7].
• Some of the software, data and buildings are Indian. The most advanced chips are not yet made in India. That is why the Semiconductor Mission is part of the same story.
• For your answer: "sovereign cloud" in India today means Indian control of the data layer, not of the whole stack.

9. The Case Against Keeping Data at Home — and Its Answer

• The strongest objection: location does not make data safe
• Security comes from encryption, audits and access rules, not from which country the server is in. The STQC audit against the ISO 27001/27017/27018 standards checks exactly these things [2].
• Forcing data to stay in India can raise costs. Departments lose the scale and price of global cloud companies, and they may get locked into fewer, smaller providers.

• India itself partly accepted this argument. The DPDP law moved away from blanket localisation (a rule that all data must stay in the country). It allows data to go abroad, except to countries the Centre puts on a notified "negative list" [9].

• What the objection gets right

• Insisting on a location for all data would be costly and would not add much security. India's own law does not do this.

• Putting priority on State Data Centres has a cost. Their capacity differs a lot from state to state, so some departments may end up on weaker infrastructure than a large private cloud could give them.

• Why the policy still stands

• The rule is graded, not blanket. Under the Cloud Selection Framework, data is first classified by how sensitive it is, and only the sensitive part must stay on government-owned infrastructure [1].
• For some data, legal reach matters as much as encryption: defence, policing, and citizen records such as language data from BHASHINI [1].
• A balanced conclusion: localisation is justified for a narrow set of sensitive data, and costly beyond that. The current design tries to draw that line, and the exam answer should say so.

10. 45,000 GPUs Is a Count of Machines, Not a Measure of Use

• "Onboarded" means available, not busy
• IndiaAI has 45,000+ GPUs onboarded [1]. By August 2026, 237 projects had used 93.18 lakh GPU-hours in total [5].
• A rough check: 45,000 GPUs running all day give about 10.8 lakh GPU-hours in a single day.

• So all the use so far is about what the full pool could deliver in around nine days. The pool grew in stages, so this is only a rough picture. But it shows the gap between buying compute and using it.

• Why this matters for a Mains answer

• The real test of the Mission is demand: are start-ups and universities actually using the subsidised compute (below ₹100 per hour [6])?

• Headlines about the GPU count measure supply. They do not tell us whether the ₹10,371.92 crore outlay [1] is producing models and research.

• The same caution applies to MeghRaj

• 2,330 departments are on MeghRaj 2.0 [1]. That counts how many departments have something on the cloud. It does not tell us what share of their work has moved, or whether costs actually fell.

11. The Electricity and Water Bill Nobody Has Fully Priced

• Power demand is set to jump
• The Central Electricity Authority (CEA) estimates that data centres will need about 17 GW of electricity by 2031–32 [13].
• Worldwide, data centres use about 1.5% of all electricity. In the US, the figure is about 4% [15].

• The IMF projects that AI-driven electricity use worldwide could reach 1,500 TWh by 2030, which is about as much as India uses in total today [15].

• What happens when the grid cannot keep up — the US warning

• The IMF modelled a case in which renewable capacity and power lines do not grow fast enough. US electricity prices could then rise by 8.6% [15].
• Under current policies, US carbon emissions would rise by 5.5% [15].

• The lesson for India: if data centres grow faster than green power and transmission lines, ordinary consumers may pay more, and more coal may be burned.

• Water is measured weakly

• The government says it has received no concerns about water use by data centres [12].
• "No complaints received" is not the same as "measured and found safe". Data centres cluster in cities that are already short of water.

• Groundwater use by data centres is governed by the general Jal Shakti guidelines of 2020 (amended in 2023), not by any rule written for data centres [11].

• The standards exist, but they are not yet a reporting duty

• BIS has published Indian Standards on PUE (Power Usage Effectiveness: the total power a building uses divided by the power its computers use; closer to 1 is better), WUE (Water Usage Effectiveness: water used per unit of computing energy) and carbon use [11].
• The government's reply to Parliament points to these standards and to general green-energy schemes [11]. It does not describe a rule that makes each data centre publish these numbers.

12. Who Should Do What Next

• MeitY should publish how much the compute is used, not just how much exists
• Publish GPU use by month, by sector and by type of user (start-up, academia, government).

• Without this, nobody can judge whether the GPU count [1] matches real demand [5].

• BEE and BIS should turn data centre standards into a reporting duty

• The PUE and WUE standards already exist [11]. The next step is to make large data centres report them every year, as a condition for incentives such as the Budget tax holiday [14].

• This links a tax benefit the government is already giving to a green condition.

• The Ministry of Power and CEA should plan data centre load with the grid

• The 17 GW estimate [13] should be built into transmission planning, using the Green Energy Corridor and Green Energy Open Access Rules [11].

• The IMF's US case shows prices rise when power lines lag behind demand [15].

• Jal Shakti and the states should set water rules by location

• Data centres in water-short areas should be required to use cooling methods that need little water. The Parliament reply itself mentions options such as direct-to-chip liquid cooling and immersion cooling [11].

• MeitY should state plainly which data foreign providers may host

• As foreign providers come in under the tax holiday [14], the Cloud Selection Framework's data categories [1] should say clearly what they may and may not host. This keeps "sovereign cloud" from becoming just a slogan.

13. Anchors for Answers

• Data: CEA estimates data centre electricity demand at about 17 GW by 2031–32 [13]
• Data: Data centres use about 1.5% of the world's electricity, and about 4% in the US [15]
• Data: 93.18 lakh GPU-hours used by 237 projects (Aug 2026), against 45,000+ GPUs onboarded [1][5]
• Report/Committee: IMF Working Paper "Power Hungry: How AI Will Drive Energy Demand" (WP/25/81, 2025): 8.6% US power price rise when grids are constrained [15]; Lok Sabha Unstarred Question 2990 (5 Aug 2026) on sustainable data centres [11]
• Law/Case: DPDP Act, 2023 (cross-border transfer allowed except to notified countries) [9]; K.S. Puttaswamy (2017), privacy under Article 21
• Comparison: United States: fast data centre growth without matching grid expansion raises power prices and emissions [15]
• Scheme: Green Energy Corridor and Green Energy Open Access Rules, for supplying clean power to data centres [11]; BIS LITD 31 standards on PUE, WUE and carbon use [11]; Budget 2026–27 tax holiday for foreign cloud providers using Indian data centres [14]

14. Mains Relevance

• GS-II: Government policies and interventions; e-governance applications, models, successes and limitations; transparency and accountability.
• GS-III: Infrastructure; awareness in IT and computers; indigenisation of technology; challenges to internal security through communication networks and cybersecurity.
• Possible questions: 1. Data sovereignty is now a part of national security. Discuss India's efforts to build a domestic cloud ecosystem in this context. (GS-III, 15 marks) 2. Examine how MeghRaj and the Cloud Selection Framework can make e-governance in India more efficient and more secure. What challenges remain? (GS-II) 3. Data centres are the backbone of the AI economy, but they put pressure on energy and water. Critically analyse India's data centre expansion. (GS-III)

15. Related Topics to Study Next

• IndiaAI Mission and its seven pillars: the compute pool is delivered as a cloud service.
• DPDP Act, 2023 and Rules, 2025: set the rules for data residency and cross-border transfer.
• Digital Public Infrastructure (Aadhaar, UPI, BHASHINI): these platforms run on sovereign cloud and data centres.
• Semiconductor Mission (ISM): hardware sovereignty complements cloud sovereignty.
• CERT-In and the cybersecurity framework: handle incident reporting and the security of cloud workloads.
• Puttaswamy (2017) and the right to privacy: the constitutional basis for data protection.
• Data centre energy demand and renewable energy integration: a link to environmental issues.
• GeM (Government e-Marketplace): the procurement channel for cloud services.

16. NCERT Concepts to Revise

• Sovereignty (Polity): The note's central theme is data sovereignty: national control over sensitive data, and why it stops short of the chip.
• Data localisation (Economy): Explains the rule that data must be stored inside India, and the note's debate on its costs and limits.
• Digital Personal Data Protection Act 2023 (Polity): India's main data protection law. It sets consent duties and conditions for sending data abroad.
• Right to privacy (Polity): Puttaswamy (2017) is the constitutional basis of data protection, linked to Article 21 in the note.
• Digital public infrastructure (Polity): Platforms like Aadhaar, UPI and BHASHINI run on this cloud, making DPI a recurring exam theme.
• e-Governance (Polity): MeghRaj and the Cloud Selection Framework aim at faster, cheaper, more secure delivery of government services.

17. Common Errors / Trap Areas

• Wrong ministry or body: MeghRaj and CSP empanelment are under MeitY. The audit is by STQC, not CERT-In. NIC runs the data centres [1][2].
• Old numbers: the CSP count went 12 → 23 → 26 → 30. MeghRaj departments went 2,170 → 2,330 [1][2].
• Data centre figures differ by source: a February 2026 PIB document said ~960 MW installed and 9.2 GW by 2030 [6]. The October 2026 release says 1.57 GW now and ~8 GW by 2030 [1]. Use the latest, and cite its date.
• GPU counts: the 10,000 target (2024) → 18,693 → 38,000+ → 45,000+. Don't mix up target and achievement [5][6].
• Service vs deployment models: IaaS, PaaS and SaaS are service models. Public, private and hybrid are deployment models. "Serverless" does not mean there are no servers [1].
• DPDP is not a blanket localisation law: cross-border transfer is allowed except to notified countries. Localisation can apply only to specified data of Significant Data Fiduciaries [8][9].