TDF and iDEX Grants for Sovereign AI

In a DRDO lab canteen in Bengaluru two founders argued about 'the defence grant'. One meant TDF — the Technology Development Fund, a DRDO scheme to grow indigenous defence technology with industry. The other meant iDEX — the DIO challenge path. They were about to file the same twelve slides on both portals. A scientist at the next table, who had sat on both kinds of reviews, said they would waste two committees.

This article keeps the schemes apart on purpose. Merging them is the most common sophisticated error we see in sovereign-AI decks. Both can fund work that later becomes an isolated agent on a defence network. Both can refuse you. Neither is a civilian SDC empanelment.

We will not publish grant ceilings as if they were frozen. TDF and iDEX pages and brochures state envelopes for given calls. Those envelopes have been amended and will be amended. Quote the call you are in.

Not legal, tax or grant-accounting advice. Read tdf.drdo.gov.in and idex.gov.in. If a consultant offers a combined 'TDF-iDEX kit', do not buy it.

Two schemes, two centres of gravity

TDF is DRDO's instrument to develop technologies that the defence research establishment cares about, with Indian industry including startups and MSMEs, in the Make in India spirit. The user flavour is a laboratory and a technology directorate. Success looks like a developed technology that can move toward induction or toward a production partner. The portal is tdf.drdo.gov.in.

iDEX is DIO's instrument to pull innovators into service-shaped problems through challenges. The user flavour is a service or organisation that wrote a problem statement. Success looks like a prototype that the user might take into a procurement pathway. The portal is idex.gov.in.

A sovereign-AI firm might honestly fit TDF if it is developing a technology — say, a verifiable isolated inference stack, a speech system for a constrained radio, a document intelligence method a lab has listed — rather than only wrapping a commercial agent. The same firm might fit iDEX if a DISC problem asks for a deployable capability. Fit is about the call, not about the adjective sovereign.

A comparison you can take into a founder meeting

Axis	TDF (DRDO)	iDEX (DIO / MoD)
Home	tdf.drdo.gov.in, DRDO	idex.gov.in, Defence Innovation Organisation
Typical pull	Technology development against DRDO / lab needs and calls	Challenge problems from services and organisations
What you are usually selling	A technology development project	A prototype against a problem statement
What success is not	Not automatically a production indent	Not automatically a production indent
IP and publications	Read the TDF agreement for that call — labs care about this	Read the iDEX / SPARK-style agreement for that challenge
Civilian marketing use	Easy to over-claim 'DRDO funded'	Easy to over-claim 'MoD approved'

Money, without a fake ceiling table

Both schemes publish financial support. iDEX official material has publicly described grant support through DISC and related windows, and a higher band for Prime-like windows. TDF calls state their own project cost norms. Repeating last year's crore figure here would make this article false the next time a brochure moves.

What you can put in a cash model today: 'If selected, support is milestone-based, incomplete against our full burn, and not a substitute for civilian revenue. We will not hire a standing defence team on a single application.'

What you cannot put: 'TDF always pays X, iDEX always pays Y, we will close the round on that.' Grant accounting, TDS, and whether GST applies to a particular disbursement are questions for your CA against the actual sanction. Do not let a slide decide them.

When 'sovereign AI' is a fit — and when it is a smell

Fit: you can develop or prototype a capability that runs isolated, does not train on the user's operational data without a written regime, and can be evaluated on a closed set. You can name the technology, not only the vibe.

Smell: you relabel a municipal helpdesk agent as 'battlefield decision superiority'. You propose a public-cloud LLM with a VPN. You refuse to discuss model provenance. You claim DPDP compliance as if it were a defence accreditation.

DPDP still matters for any personal data you touch. It is not the defence security process. ISO 27001 may help your hygiene story. It will not replace a lab's or a service's own assurance. Air-gap as practised in a state SDC is a starting vocabulary, not a clearance.

• Write a data-gravity sentence: whose data, whose metal, whose logs.
• Write a training sentence: we do not train on your operational data, or we do only here, like this, then we delete.
• Write an update sentence: media, hash, two-person load — or you do not belong in this neighbourhood.
• Write a human sentence: who still authorises the act the agent drafted.

Can you apply to both?

Yes, if the work packages are not the same grant paying twice for the same deliverable. Dual-use of public money on identical milestones is how you meet an audit. Keep ledgers. Tell both sides if asked. If a call forbids parallel funding, obey.

Do not copy-paste the proposal. A TDF reviewer wants a technology-development plan a lab can oversee. An iDEX reviewer wants a problem-solution and a user-shaped prototype. Same company, different essays.

Do not staff both as if both will hit. Probability-weight your founder time. One serious application beats two cynical ones.

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.

They are both MoD money, so the distinction is academic.

DRDO and DIO are different organisations with different contracts. Filing the wrong shape is how busy people reject you. The distinction is operational.

Give us the current maximum grant so we can model dilution.

Model a range from the live brochure and assume you will not get the headline number, and assume you will still need civilian cash. Dilution models that depend on a frozen ceiling are theatre.

Sovereign AI should automatically qualify.

No adjective automatically qualifies. A work plan does.

We can subcontract the whole TDF to a campus and take the logo.

Read the call's subcontracting and lead-performer rules. Passing through money while keeping the brand is how schemes close doors later.

A fortnight to choose a door

Choose one primary door this quarter unless you truly have two teams.

1. Days 1–3: read both portals. List live calls and problem or technology themes in a table.
2. Days 4–6: mark each live item fit / stretch / no. Stretch means you would have to invent a product.
3. Days 7–10: write the one-page data-gravity, training, update and human sentences. If you cannot, you are not ready for either door.
4. Days 11–14: pick one application. Assign a founder. Open a separate ledger code for any future grant. Brief the board that this is not revenue.

How this shows up in the file

Internal note: 'TDF and iDEX are distinct. We will not merge them in decks. We will quote financial envelopes only from the call PDF dated [date]. We will not describe grant selection as MoD or DRDO product approval. Parallel funding, if any, will be disclosed and not double-billed.'

Buyers on the civilian side should score the product in front of them, not the grant logo in the annexure.

This article is informational field guidance for Indian public institutions and their vendors, not legal, tax, procurement or engineering advice. Confirm the live circular, RFE, GCC, GeM term, state G.O. and your counsel before you file anything. Incentives, ceilings and portal screens change.

How to run the route without confusing the letterhead

“TDF and iDEX Grants for Sovereign AI” is a route problem. A P5 Startup/Vendor should know which legal person they are talking to — NIC, NICSI, a state IT corporation, iDEX, or a GeM buyer — and which paper that person can actually issue. Searching “TDF scheme AI grant” is not the same as being on a panel that can receive a work order.

TDF and iDEX both put public money under defence problems. They are not one programme. Compare portals, users, IP posture and prototype-versus-development before you file either. Empanelment letters are not purchase orders. DPIIT recognition is not a technical score. Reserved startup seats, if a notice writes them, are local to that notice. IndiaAI compute empanelment is not NICSI application-software empanelment.

Keep a warm evidence pack: CIN, GST, DPIIT, Udyam, financials, work-completion letters, architecture one-pager, DPA draft. Renewals are lost by people who treat the panel as a trophy.

• Screenshot the live RFE paragraph you are relying on, dated.
• Match the bidding entity name across every certificate.
• Do not mix iDEX, TDF, NICSI and GeM clocks on one tracker cell.
• Record the validity end date 90 days before it dies.

Close this loop before the next CAB

Put “TDF and iDEX Grants for Sovereign AI” on the next change-advisory or bid-opening agenda as a single line item with an owner. If it cannot earn a line item, it will not earn a control. The owner should be a P5 Startup/Vendor, not “the vendor.”

Revisit the item when the model, the GeM term, the region, or the SI changes. “TDF scheme AI grant” is not a one-time workshop. It is a watch item. Date the last check. Unsigned watch items are souvenirs.