India AI & sector rulesUpdated 22 November 20269 min read

Data Localisation in India: Where Can You Host AI Workloads?

Where you can legally host AI workloads that touch Indian personal data: DPDP's negative list, Rule 13(4) for SDFs, RBI payment data, CERT-In logs and sectoral rules, in one decision table.

By Suhas Bhairav · Reviewed against the DPDP Rules, 2025 (G.S.R. 846(E), 13 Nov 2025)

DPDP approach
Negative list (s.16); no general localisation
Foreign State access conditions
Rule 15, by general or special order
SDF localisation
Rule 13(4), for data specified by the Central Government
Payment data
Stored only in India (RBI circular, 6 April 2018)
CERT-In logs
Rolling 180 days within Indian jurisdiction
DPDP cross-border rules apply from
13 May 2027

Quick answer

India has no general data localisation rule for AI workloads. The DPDP Act uses a negative list: section 16 allows transfers abroad except to countries the government restricts, and Rule 15 lets it impose conditions on making data available to foreign States. Localisation bites through Rule 13(4) for notified SDFs, RBI payment data storage, CERT-In's 180-day logs, and sectoral rules.

  • The DPDP Act has no general localisation requirement; section 16 allows cross-border transfer unless the government restricts a destination by notification.
  • Rule 15 lets the government set requirements on making personal data available to foreign States or entities they control.
  • Rule 13(4) can require a notified Significant Data Fiduciary to keep specified personal data and its traffic data in India.
  • Stricter sectoral rules prevail under section 16(2), notably RBI's requirement that payment system data be stored only in India.
  • CERT-In requires ICT system logs to be kept for a rolling 180 days within Indian jurisdiction, even if models run abroad.

Who this is for: CTOs and cloud architects, AI platform and MLOps teams, Fintech and payments companies, Data Protection Officers and privacy leads, Indian SaaS companies using foreign model APIs

Do you have to host AI workloads in India?

Usually not, but it depends on what data the workload touches and who you are.

The DPDP Act takes a negative-list approach. Section 16(1) lets the Central Government restrict transfers to specified countries by notification; everywhere else is permitted. Rule 15 adds that a Data Fiduciary transferring personal data abroad must meet any requirements the government specifies, by general or special order, about making that data available to a foreign State or entities under its control or agencies.

So a startup sending prompts containing customer names to a model API in another country is not breaking a DPDP localisation rule — there isn't one for ordinary fiduciaries. But four other sources of localisation can apply:

  1. Rule 13(4) — if you are notified as a Significant Data Fiduciary, personal data specified by the government (on a committee's recommendation) and its traffic data must not leave India.
  2. RBI payment data — payment system data must be stored only in India.
  3. CERT-In logs — 180 days of ICT system logs within Indian jurisdiction.
  4. Other sectoral rules — insurance records, telecom licence conditions, government data and MeitY cloud empanelment.

Section 16(2) confirms that where another Indian law provides a higher degree of protection or restriction on transfers, that law prevails. DPDP's cross-border provisions apply from 13 May 2027.

Decision table: where can this AI workload run?

Data in the AI workloadKey ruleHosting position
Ordinary customer personal data (non-SDF)DPDP s.16, Rule 15Abroad permitted unless destination restricted by notification; meet any Rule 15 orders
Data specified under Rule 13(4), held by a notified SDFDPDP Rule 13(4)Must stay in India, including traffic data about its flow
Payment system dataRBI circular, 6 April 2018Stored only in India; foreign processing allowed only under RBI's conditions
ICT system logs (any entity)CERT-In Directions, 28 April 2022Copy must be maintained in India for rolling 180 days
Insurance policyholder recordsIRDAI requirementsCheck IRDAI rules on records held in India
Telecom subscriber and network dataLicence conditionsCheck DoT licence terms
Government data on cloudMeitY empanelment and procurement termsUsually India-hosted on empanelled providers
Non-Indian individuals' data processed under contract with a foreign entityDPDP s.17(1)(d)s.16 does not apply; client contract and foreign law govern
Anonymised or aggregate dataOutside "personal data" if not identifiableFewer constraints, but test re-identification risk

Two caveats. First, no country has been restricted under section 16 and no SDF has been notified as of this writing; check MeitY and the Gazette. Second, the sectoral rows describe the general position — read the current regulator circulars for detail.

AI data flows that people forget are transfers

Localisation analysis often stops at "where is the database?" AI systems move data in more places.

  • Prompts to foreign model APIs. Every prompt containing personal data is processed where the model runs.
  • Embeddings. Vector representations of personal data stored in a foreign vector database are still derived from personal data; treat them as personal data unless you have strong evidence otherwise.
  • Fine-tuning datasets. Uploading training data to a foreign training service moves it abroad, often to multiple regions.
  • Observability traces. LLM tracing tools capture full prompts and responses; their SaaS backends may be abroad.
  • Support access. A vendor engineer abroad viewing your data is a cross-border access event.
  • Backups and disaster recovery. DR regions are frequently in another country.
  • Agent tool calls. An agent calling a foreign SaaS API with customer data creates a transfer.

What to record

For each flow, record the data categories, destination country, provider, contract basis and whether any localisation rule from the decision table applies. This becomes your evidence for DPDP cross-border transfer and feeds the data mapping inventory.

Architecture patterns for mixed constraints

Most organisations end up with some data that must stay in India and some that can travel. Three patterns handle that.

1. Split-plane architecture

Keep the data plane (databases, vector stores, logs) in an Indian region and call models through a gateway. Send only what each request needs. Payment data and SDF-specified data never leave; the gateway enforces that by classification tags.

2. Tokenise before you transfer

Replace direct identifiers with tokens before data leaves India, keeping the token vault in India. Rule 6(1)(a) explicitly recognises "virtual tokens mapped to that personal data" as a security measure. The model sees CUST_83F2, not a name and phone number. See masking and tokenisation for AI pipelines.

3. In-country inference

Run open-weight models in an Indian cloud region or on-premises for workloads touching localised data. This is the only clean option when the data itself, not just identifiers, is restricted.

Logs

Whatever the pattern, keep a copy of gateway and security logs in India for CERT-In's 180 days, and retain logs for at least one year under DPDP Rules 6(1)(e) and 8(3) from 13 May 2027. More detail: CERT-In rules for AI systems.

Common mistakes

  • Assuming DPDP requires localisation. It does not, for ordinary fiduciaries. Over-localising can cut you off from capable models for no legal reason.
  • Assuming no localisation anywhere. Payment, SDF-specified, sectoral and CERT-In log requirements are real.
  • Forgetting traffic data. Rule 13(4) covers traffic data pertaining to the flow of specified personal data, not just the data itself.
  • Ignoring DR regions. A compliant primary region with a foreign DR region is not compliant for localised data.
  • Not watching for notifications. Section 16 restrictions and Rule 15 orders can be issued by notification. Assign someone to track the Gazette.
  • Overlooking the s.17(1)(d) carve-out. Indian IT services firms processing foreign clients' customer data under contract are outside section 16. But security (s.8(5)) still applies.

Before 13 May 2027

Map every AI data flow, classify data against the decision table, implement tokenisation or in-country inference where needed, and put a Gazette watch in place for section 16 and Rule 15 orders.

If you also handle EU personal data

Indian companies serving European customers face the mirror problem. Under GDPR, transferring EU personal data to India is a restricted transfer under Chapter V, because India has no EU adequacy decision. Standard Contractual Clauses and a transfer impact assessment are the usual tools.

That leads some teams to host EU workloads in EU regions and access them from India. Note that remote access from India to EU-hosted personal data is itself treated as a transfer under GDPR, so EU hosting reduces but does not remove Chapter V obligations.

DirectionMain constraint
Indian personal data to abroadDPDP s.16 negative list, Rule 15, sectoral rules
EU personal data to IndiaGDPR Chapter V, SCCs, transfer impact assessment
EU personal data accessed from IndiaStill a transfer under GDPR

For details, read GDPR adequacy and India and EU data residency on Hetzner from India.

Further reading on suhasbhairav.com

Before 13 May 2027

Compliance checklist

  1. Map AI data flows

    Record prompts, embeddings, training data, traces, backups and support access with destinations.

    DPDP s.16

  2. Classify data against localisation rules

    Tag payment data, SDF-specified data and sectoral records that cannot leave India.

    DPDP Rule 13(4); RBI 2018 circular

  3. Keep CERT-In logs in India

    Maintain a 180-day log copy within Indian jurisdiction for all AI components.

    CERT-In Directions, 28 Apr 2022

  4. Tokenise before transfer

    Replace direct identifiers with tokens held in an Indian vault before calling foreign models.

    DPDP Rule 6(1)(a)

  5. Check DR and backup regions

    Ensure localised data is not replicated to foreign regions.

    RBI 2018 circular; Rule 13(4)

  6. Watch for notifications

    Track the Gazette for section 16 restrictions and Rule 15 orders.

    DPDP s.16, Rule 15

  7. Review vendor regions

    Confirm model, vector database and observability vendor regions and sub-processors.

    DPDP s.8(2)

Key dates

  1. RBI circular on storage of payment system data

  2. CERT-In directions including 180-day logs in India

  3. DPDP cross-border rules (Rule 15) and SDF obligations (Rule 13) commence

Answers

Frequently asked questions

Does the DPDP Act require data localisation?

No, not generally. Section 16 permits transfers abroad unless the Central Government restricts a destination by notification. Rule 15 lets it impose conditions on making data available to foreign States. Localisation applies only to personal data specified under Rule 13(4) for notified Significant Data Fiduciaries, and under stricter sectoral laws.

Can I use OpenAI or other foreign model APIs with Indian customer data?

DPDP does not prohibit it for ordinary fiduciaries, provided the destination is not restricted and you meet security, contract and notice duties. Sectoral rules may block it for payment data or other localised categories. Minimise and tokenise personal data before sending it.

Which data must stay in India?

Payment system data under RBI's 2018 circular; personal data specified by the government for notified SDFs under Rule 13(4); ICT logs for 180 days under CERT-In directions; and certain sectoral records under insurance, telecom and government rules. Check each regulator's current requirements.

Are embeddings personal data under DPDP?

Treat them as personal data if they are derived from personal data and can be linked back to an individual, for example because source chunks or identifiers are stored alongside. Storing embeddings abroad is therefore a transfer to assess under section 16 and any sectoral rule.

Has the government restricted any country under section 16?

No restriction notification under section 16 is reflected in the information available at the time of writing. Cross-border provisions apply from 13 May 2027. Check the Gazette and MeitY's data protection page for any notification before finalising architecture.

Does Rule 13(4) apply to my startup?

Only if your company is notified as a Significant Data Fiduciary, or included in a notified class, and only for the personal data the government specifies on a committee's recommendation. No SDF has been notified as of this writing; once notified, the restriction also covers related traffic data.

Legal references and sources

This guide explains the law for engineering and business teams. It is not legal advice; confirm how it applies to your organisation with qualified counsel, and check the Gazette for later notifications or amendments.

About the author

Suhas Bhairav

Suhas Bhairav

Systems architect and applied AI practitioner who designs production AI systems for Indian and global companies, with a focus on privacy-by-design under the DPDP Act and the EU AI Act.