Local Content Rules and Make in India for AI

A bid opens. Three platforms claim Make in India. One is an Indian-incorporated company with a model trained abroad and a runtime licensed from a US firm. One is a foreign OEM with an Indian assembly story for the server and a foreign model API. One is a DPIIT-recognised startup that wrote its orchestration in Bengaluru and uses open weights. All three put a flag on slide two. At most one of them has a local-content working that would survive a challenge.

The Public Procurement (Preference to Make in India) Order, issued by DPIIT and revised over the years, is the usual Union-level instrument. It speaks of local content, of Class-I and Class-II local suppliers, of purchase preference, and of minimum local-content thresholds that nodal ministries can notify for their items. GeM surfaces some of this as fields and filters. None of it is the same as Indian owned, data stays in India, or DPIIT recognised. Those are other articles.

This explainer is for procurement officers writing AI bids and for founders tempted to type 100% into a GeM box. Read the current order and any sector notification on dpiit.gov.in and the relevant ministry site. Thresholds and definitions move. This is the method, not a substitute gazette.

What the order is talking about

Local content is, in substance, the portion of a product or service that is domestic, calculated as the order defines — typically a residual after netting imported content from the price, with the order's own exclusions and formulae. You must use the definition in the order you are applying, not a vibe. Software and cloud make the formula uncomfortable. That discomfort does not let you skip it.

Class-I and Class-II local suppliers are classes based on crossing notified local-content percentages. Purchase preference, where it applies, is a mechanism that can let a higher-local-content supplier match or win against a lower one under stated conditions. The exact percentages and the matching rule live in the current order and in any item-specific notification. Do not copy a percentage from a 2017 news story into a 2026 AI bid. Nodal ministries may notify higher minimum local content for specific items. Check whether MeitY or another nodal body has a live notification that actually names the item you are buying. Do not assume all AI is 50% or any other number from a conference. Bidders self-certify in many processes, with later verification and penal consequences for false claims. That is why a casual 100% is reckless.

MII is a price-composition claim. Other good claims still need their own evidence.

Claim on a slide	What it is not	What you still have to calculate
Incorporated in India	Not automatically Class-I	Imported runtimes, cloud APIs, foreign labour in the price
DPIIT recognised startup	Not a local-content certificate	Same formula; recognition is a different preference family
Open-source model weights	Not automatically Indian content	What you added, where, and what you still license
Data stays in the SDC	A sovereignty / DPDP design choice	Irrelevant to MII unless the order's formula says so — usually it does not
Servers assembled in India	May help the hardware lot	Does not automatically cover the software lot in the same cart
Support team in Noida	Maybe part of a service calculation	Only the portion the order lets you count

How to ask for it in an AI bid without writing fiction

If you want MII to apply, say which order and which minimum local content you are using, and whether you are treating the buy as goods, services, or lots. Lots help: a GPU node can carry a hardware calculation; a licence can carry another. A blended 100% on a mixed cart is how you invite a complaint. Require a local-content working, not only a percentage box. Ask what was excluded, what was imported, and who signs. For software, insist on naming third-party runtimes and model APIs in the working. A silent API is imported content wearing a moustache. Do not set a minimum that only one bidder can meet unless you have a recorded market justification. On GeM, use the live MII fields rather than a scanned PDF that the portal will not evaluate.

Founders: do the working before the box

Sit with finance, not with marketing. List the bill of materials of the SKU: your engineering cost in India, Indian cloud or none, foreign APIs, foreign licences, imported laptops you should not try to count, support salaries. Apply the order's formula. If you cannot finish the working, you cannot type a number. If the number is lower than you wished, you have three honest moves: change the SKU (remove a foreign API), bid as Class-II or as non-local, or do not claim. You do not have a fourth move called round up for the portal. Keep the working. A later verification or a rival's complaint will arrive on a day you are busy.

1. Identify the live order and any item-specific notification.
2. Split lots if hardware and software would distort one percentage.
3. Demand a signed working, not a flag.
4. Name third-party runtimes and APIs in the working.
5. Use GeM's live MII fields when buying through GeM.
6. Verify or reserve the right to verify; do not treat the box as sacred.

Objections you will hear — and what to do with them

These are the lines that stall the file. Answer them in the room, then put the answer in the note. A spoken answer without paper will be forgotten by the next officer.

Software is weightless, so MII cannot apply.

The order applies to products and services as notified and as GeM implements. Weightlessness is not an exemption. It is a reason to calculate carefully.

If they are Indian-owned, that is enough.

Ownership is a different question — see the sovereignty cluster. MII is local content in the supply. An Indian-owned reseller of a foreign API can score poorly. A foreign-owned Indian development centre can score better on the formula. Do the working.

We will put 50% in every AI RFP.

Only if a current notification or the general order supports that number for this item. Invented thresholds are how you lose a challenge. Cite or do not write a number.

A week to make MII real on one indent

1. Day 1: download the current DPIIT order and any MeitY/item note.
2. Day 2: decide lots.
3. Day 3: draft the working template bidders must fill.
4. Day 4: align GeM fields and ATC language.
5. Day 5: legal reads consequences for false claims.
6. Day 6–7: file the method next to the Rule 149 page.

How this shows up in the file

Keep the order PDF you relied on, the lot split, the blank working template, and any GeM screenshots of MII flags. A later officer should be able to re-run the preference without calling you. Do not file a slide with a flag as evidence of Class-I.

A working template that forces honesty

Ask bidders to fill: SKU name; contract price basis; value of imported components and services (named: model API, runtime licence, foreign cloud, foreign labour); value of domestic components and services (named); exclusions they applied and the clause that allows the exclusion; resulting local-content percentage; class claimed; signatory and date. If a line is brand value or IP created in spirit, strike it. For open weights, ask them to put the commercial runtime and any paid inference stack in the imported column unless they can show a domestic paid component.

Preference without turning the bid into a lottery

Write how preference will be applied: which lot, which class, what happens on a tie, what happens if the working is missing. Missing workings should be non-responsive for the preference, not silently treated as Class-I. Rivals will check.

How this relates to sovereignty language

MII will not make a platform sovereign. A high local-content score with a foreign model API on the critical path is still a transfer architecture. A lower score with inference and logs under your control may be the better file. Use MII for what it is. Use the air-gap annexure for what it is. Use DPDP roles for what they are. One flag cannot do three jobs. If a political note demands both Class-I and no outbound, test whether any bidder can honestly offer both. If none can, change a requirement or change the politics. Do not force a 100% box.

What the next noting must contain

“Local Content Rules and Make in India for AI” belongs in a file, not only in a search result. A P2 Procurement should be able to point at one artefact that proves “local content AI procurement”: a packet capture, a processing schedule, a scored evaluation row, a dated notice, or a refusal rule. If the only evidence is a slide, you have a heading.

Make in India is a calculation and a class, not a tricolour on a slide. For software and models, the calculation is the part almost nobody has actually done. DPDP 2023 does not define sovereign AI and does not write a blanket localisation rule for every model hop. CERT-In’s 28 April 2022 directions still set specified incident clocks and 180-day log retention in India for in-scope events. The November 2025 AI governance text is guidance, not a statute. A Proprietary Article Certificate, when it is lawful, lives in GFR Rule 166 — not Rule 161.

Write three dated sentences under C3 GeM & Procurement: what was decided, which designation owns it after the next posting order, and when it will be re-checked. Unsigned sentences are souvenirs. Dated sentences are controls.

• Name the designation that owns “local content AI procurement”, plus a deputy.
• Attach one artefact a stranger can open next year.
• Name the instrument you are actually using — Act, direction, GFR clause, GeM term, or guideline paragraph.
• Leave unsourced percentages, GMV slides and house forecasts out of the noting.
• Revisit when the model, the SI, the notice, the region or the posting changes.

This article is informational field guidance, not legal or procurement advice. Confirm against the current General Financial Rules, Department of Expenditure OMs, live GeM terms on gem.gov.in, DPIIT orders and your counsel before you file it.