← All digests

2026-04-01

India AI Digest — Wednesday, April 1, 2026

  • MeitY circulates a Second Draft Amendment to the IT Rules (March 30): continuous on-screen AI labels, three-hour deepfake takedowns, and a Rule 3(4) that conditions intermediary safe harbour on compliance with future MeitY directions.
  • Sarvam AI is reported to be finalising a $300–350M round at a $1.5B post-money valuation, with Bessemer leading and Nvidia, Amazon, HCLTech and Prosperity7 as strategic participants — the largest single capital infusion into a pure-play Indian AI company on record [TBV: round not yet closed].
  • Position movements: regulatory_clarity 0 (India, hypothesis-bound), capital_availability +1 (India).

MeitY circulates Second Draft IT Rules Amendment; safe-harbour now conditioned on MeitY directions

The Ministry of Electronics and Information Technology placed the draft IT (Intermediary Guidelines and Digital Media Ethics Code) Second Amendment Rules, 2026 in the public domain on March 30, 2026, with a comment window now extended to May 7. The draft sits on top of a February 20, 2026 amendment that already established the Synthetically Generated Information (SGI) framework — three-hour takedown obligations for AI-generated content on government notice, two hours for impersonation cases, mandatory provenance metadata.

What this means. The substantive moves in this draft are three.

First, on AI-generated content, the labelling standard tightens. Earlier rules required SGI labels; the draft requires that the label be continuous and clearly visible throughout the duration of visual content. The shift is from disclosure-at-load to disclosure-as-watermark. For Indian platforms hosting user-uploaded video — short-form social, OTT comments, e-commerce reviews with generated assets — the engineering work is non-trivial. The provenance-metadata layer (C2PA-style) was already on the roadmap from February. The continuous-overlay layer is new.

Second, and more structurally, Rule 3(4) of the draft requires intermediaries to comply with any clarification, advisory, order, direction, SOP, code of practice or guideline issued by MeitY, with such compliance treated as a precondition for safe harbour under Section 79 of the IT Act. The form matters. Earlier safe-harbour conditioning ran through statutory rules and judicially-tested takedown processes. The draft moves a meaningful share of the compliance surface to executive instruments that can be issued, modified, and withdrawn outside of rulemaking. The Internet Freedom Foundation and several legal commentators have flagged this as the most consequential single change in the draft. Whether this clears the consultation in its current form, or returns narrowed, is the open question.

Third, the draft extends Part III of the Code of Ethics — historically applied to digital news publishers — to news and current-affairs content posted by individual users on social media. This is not directly an AI rule, but it sits in the same instrument and changes the regulatory perimeter that AI-generated political and current-affairs content lives within.

The dual read holds. Read generously, the draft fills in operational specifics for an SGI regime that India has chosen to lead on globally — first major economy to set enforceable deepfake-takedown timelines, first to require continuous AI labelling, first to wire provenance metadata into platform compliance. India is using the IT Rules instrument to do what other jurisdictions have done by legislation; speed is the trade-off the instrument enables. Read skeptically, Rule 3(4) is the structural risk: a compliance standard that adjusts by executive direction is hard to plan against, hard to litigate against, and asymmetric in burden. Indian product startups operating an AI-feature surface absorb the same uncertainty as global incumbents who can staff a full compliance practice against it.

This is a draft, not a notification. The May 7 deadline runs over the next five weeks and the final form is genuinely undetermined.

India angle. The implications cluster across the consumer AI stack.

Consumer generative-AI products in India. Indic chatbots, image and video generators, voice-cloning products. Watermarking and provenance metadata work that was being scoped after February 20 now needs a continuous-overlay implementation as well, on the assumption the draft passes substantively as written. For Indic-language products specifically, the label text, font support across scripts, and accessibility on small-screen devices are all engineering questions without good off-the-shelf answers yet.

Indian platforms hosting third-party AI content. Short-form video, marketplace listings with generated imagery, customer-support transcripts with synthetic voice. The takedown-window mechanics put real-time content moderation in scope; the safe-harbour conditioning under Rule 3(4) makes it harder to scope what enough compliance looks like in advance. Build the moderation infrastructure for the strictest plausible read.

SI layer (TCS, Infosys, Wipro, HCLTech). Client conversations on generative-AI deployments in India now have a more concrete regulatory layer to design against. The compliance practice becomes a meaningful differentiator versus global AI services firms entering the Indian enterprise market.

Indian foundation-model labs. Direct exposure is limited — model training and weights are not the regulatory target — but downstream products built on Indian models inherit the obligations. The Sarvam, Krutrim, Karya, and AI4Bharat-derived product layer will need to ship watermarking and provenance-aware generation by default. AI4Bharat's open-source releases face a separate question on whether and how the compliance infrastructure travels with the weights.

What this is not. Not the final form of the rules. The draft is open to consultation through May 7, and Indian rulemaking on synthetic content has revised in response to industry feedback before — the February 2024 advisory walkback being the closest reference point. Treating the draft as settled is premature.

Source: MeitY draft Second Amendment Rules, 2026 (PDF), placed in public domain March 30, 2026. Secondary coverage: MediaNama, Internet Freedom Foundation, LiveLaw.

Confidence: High on the existence and broad provisions of the draft; medium on Rule 3(4)'s downstream effect since rulemaking is mid-process.

Sarvam AI reported to be raising $300–350M at $1.5B valuation; Bessemer leads, Nvidia and Amazon participate

Bloomberg reported on April 2, 2026 that Sarvam AI is finalising a Series B of approximately $300–350 million at a post-money valuation of $1.5 to $1.55 billion. Bessemer Venture Partners is reported to lead. Nvidia, Amazon, HCLTech, and Prosperity7 Ventures are reported as strategic participants. Glade Brook Capital is in talks to enter the round at a later close [TBV — round not yet closed at the time of reporting]. Sarvam was selected by IndiaAI Mission to build a sovereign LLM in April 2025 and showcased its Sarvam-30B and Sarvam-105B foundation models at the India AI Impact Summit in February.

What this means. If the round closes near the reported terms, this is the largest pure-play Indian AI company funding to date — past the Krutrim 2024 unicorn raise on size, and at a valuation set against shipped foundation models rather than against a pre-product story. The substance diagnostic for Sarvam over the past two years has been the lead anchor case: AI4Bharat lineage on the technical bench, shipped models with detailed disclosure, third-party reproduction of the tokenizer-efficiency claims, and a deployment ecosystem now spanning speech, vision, agents and infrastructure. The disclosure quality is materially different from what the Indian foundation-model field looked like in early 2024.

The investor composition is the second-order signal. Nvidia and Amazon are strategic — compute supply and cloud distribution respectively. HCLTech is the SI-layer connection that bridges Sarvam's foundation work into Indian and global enterprise deployment pipelines. Prosperity7 is sovereign-adjacent capital. Bessemer is the pure-financial lead. The composition reads as a deliberate stack: financial conviction, compute access, distribution, and SI integration in a single round. This is the shape a foundation-model round takes when the company is positioning to ship into production at enterprise scale rather than to extend the research runway.

The skeptical read is the round-status caveat. Bloomberg's reporting is on a deal that is in late-stage diligence, not closed. Indian AI funding rounds at this scale have leaked, repriced, and recomposed before. Reading the announced composition as final ahead of the wire transfer is a class of error the field has made repeatedly. The capability story Sarvam is shipping into is real; the capital story should be re-verified at close.

The other piece worth holding: a $1.5B valuation for an Indian foundation-model company is structurally below the global frontier-lab valuation tier (Anthropic, OpenAI, xAI, Mistral) by an order of magnitude or more. The framing "India's frontier AI champion" risks confusing scale of capital with scale of model. Sarvam's case — Indic-optimised foundation models that change the unit economics of Indic-language production deployment — does not need frontier-tier capital to play. The round size is large for the Indian market and adequate for the strategy. It is not a frontier-lab round.

India angle. Three categories of read.

Indian AI capital ecosystem. A funding round of this size and composition resets the bar for what Indian foundation-model companies can raise, and which strategic participants are reachable. The presence of Nvidia and Amazon as named strategic investors in an Indian foundation-model round, at a stage where the financial lead is non-Indian, is a precedent. Indian deep-tech funds and sovereign-adjacent capital sources will be evaluated against this composition over the next funding cycle. Whether the round produces follow-on capital availability for the next tier of Indian foundation-model companies (BharatGen, Karya, Two Photon AI, Tech Mahindra's Project Indus continuation, others) is the empirical question.

Compute and infrastructure access. The Nvidia participation is the operational signal. Sarvam's compute pipeline now has direct supplier alignment in addition to the IndiaAI Mission GPU allocation. For other Indian foundation-model labs negotiating compute access through the IndiaAI Mission tender process or through cloud providers, the negotiating reference point shifts.

SI-layer integration. The HCLTech participation matters more than its quantum suggests. Sarvam's foundation models routed through an Indian SI's enterprise distribution is a different deployment model than direct API sales. If HCLTech operationalises this — Sarvam-on-HCL into BFSI, healthcare, government accounts where global API access faces residency or sovereignty constraints — the unit economics for Indic-language enterprise AI in India becomes meaningfully different. This is the deployment architecture that the IndiaAI Mission policy framing has been pointing at; this is one of the first commercial-stack instantiations of it.

What this is not. Not validation that India has built a frontier-tier foundation-model company. The valuation tier, the parameter scale, and the benchmark positioning of Sarvam's models are all below frontier. The case for the round is narrower and stronger: an Indian-language-specialised, enterprise-deployable foundation-model layer with a credible technical bench and a stack-aligned investor base. That is a real position. It is not the same position as Anthropic or OpenAI.

Source: Bloomberg, April 2, 2026. Corroborating coverage: Outlook Business, BW Businessworld. Sarvam has not issued a primary statement on the round at the time of writing.

Confidence: Medium. Round size, valuation, and investor list are reported, not yet primary-source confirmed; round is described as in late diligence rather than closed. The capability and substance picture for Sarvam is high-confidence on independently-shipped artefacts.

Two-item digest. The week's other live thread — the IndiaAI Mission's ongoing fund-release shortfall (Rs 400 crore released against a five-year outlay of Rs 10,372 crore, surfaced in Rajya Sabha replies) — is being reported from April 4 onward and falls outside the ±2 day window for this date. It will get its own item in the next available digest.