2026-04-29

India AI Digest — Wednesday, April 29, 2026

The Maharashtra cabinet approved AI Policy 2026 on April 29, targeting ₹10,000 Cr in investment and 1.5 lakh jobs by 2031, with provisions for at least 2,000 GPUs as a "Compute as a Service" pool, six AI Excellence Centres, five AI Innovation Cities, and grants of up to ₹1 Cr per startup (₹1.25 Cr for women-led).

Anthropic is weighing preemptive offers to raise roughly $50B at an $850B–$900B valuation, per Bloomberg reporting on April 29; the round has not been accepted and the company's run-rate is reported at roughly $40B annualised.

Position movements: regulatory_clarity +1 (Maharashtra), compute_infrastructure +1 (Maharashtra, conditional on execution), foundation_model_capability 0 (Anthropic — global frontier consolidates, not a measurable Indian-built movement).

Maharashtra cabinet approves AI Policy 2026

The Maharashtra cabinet, chaired by Chief Minister Devendra Fadnavis, approved the state's AI Policy 2026 on April 29, 2026. The policy targets over ₹10,000 crore in investments and more than 1.5 lakh jobs by 2031. Headline provisions include at least 2,000 GPUs of state-pooled compute exposed to government departments as a "Compute as a Service" platform, six AI Excellence Centres, five AI Innovation Cities, training for two lakh youth and professionals, financial support of up to 20% of AI implementation costs for 5,000 MSMEs, grants of up to ₹1 crore per startup (up to ₹1.25 crore plus 25% additional support for women-led startups), 100% stamp duty waiver, and electricity rate concessions for up to ten years for qualifying investors. Fadnavis framed the document as periodically revisable, citing his own estimate that "about 70% of jobs will be reshaped by AI" in coming years.

What this means. Maharashtra is the second large state, after Karnataka and ahead of Tamil Nadu's draft, to publish a coherent AI policy with both a compute-pool provision and a startup-grant instrument in the same document. The structural choice worth naming is the bundling of capital subsidy, stamp duty waiver, and electricity concessions — the standard industrial-policy toolkit — with AI-specific provisions like the GPU pool and ethical-AI framework. That signals the state treating AI as an industrial sector to be incentivised, not as a technology-policy domain to be regulated.

The 2,000-GPU floor is the operationally interesting figure. It sits well below IndiaAI Mission's national pool of roughly 38,000 GPUs and well below private hyperscaler India-region capacity, but it is the first state-level commitment of meaningful scale. Whether it translates into accessible compute for state-funded research and citizen-service workloads, or into a procurement vehicle that ends up locked to a few large vendors, depends on the access model the implementing rules specify. None of the present coverage names the procurement mechanism, the access-pricing band, or whether allocations will be open to non-Maharashtra startups.

The grant-and-subsidy stack — ₹1 crore per startup, the women-founder uplift, the 20% MSME implementation subsidy — is at a scale where it can move some early-stage decisions but cannot substitute for venture capital. For a deep-tech AI play, ₹1 crore is perhaps two months of senior-engineering payroll. For an MSME applying off-the-shelf AI to existing operations, 20% capital subsidy on a well-bounded implementation is materially helpful. Treat the two instruments as serving different cohorts, not as a continuum.

The "first state with a separate ethical AI framework" framing in the announcement is harder to evaluate without the framework text. Cabinet-approved policies in India routinely carry forward such positioning lines that subsequent rules notification dilute or reshape. The verifiable observation is that the policy commits to the framework; the substance lands when the framework itself is published.

India angle. State-level AI policy fragments the regulatory surface that the source-named central bodies — TPEC and AIGEG — are currently drafting against. Builders and deployers operating across multiple states will now navigate a state-specific layer in addition to the central one. Whether that fragments compute access decisions (e.g., will a Tamil Nadu startup access Maharashtra's GPU pool on the same terms?) or accelerates the field overall depends on how state policies converge or diverge in implementation.

For Indian SI-layer companies (TCS, Infosys, Wipro, HCL) and the AI-services cohort, Maharashtra's grant instruments are not directly relevant — these companies are well past the eligibility bands. The policy matters to them through its enterprise-adoption levers: BFSI, manufacturing, and government-services demand pull-through funded by the subsidies. Mumbai-headquartered enterprises absorbing the 20% implementation subsidy on AI projects creates predictable, planned demand the SIs can sell into.

For Indic-language AI work, the policy's specific mention of a State AI Data Exchange focused on Marathi and regional languages is the part to watch. Marathi data quality and coverage in existing Indic foundation-model work (AI4Bharat's IndicTrans, Sarvam's models) is uneven. A state-funded data-curation effort, if it produces release-grade corpora rather than internal datasets, would be a measurable contribution to Indic capability beyond what private labs alone have produced.

What this is not. Not a national AI policy. Maharashtra's framework operates within and below DPDP, MeitY advisories, and any future central AI legislation. State governments have limited sovereignty over the kinds of obligations that determine whether AI deployments are permissible — DPDP cross-border transfer rules, MeitY's still-pending stricter framework, sectoral regulators' AI guidance. The policy is a state's industrial bet on AI, not a regulatory instrument of the kind that would resolve the planning ambiguity flagged in earlier digests.

Source: Business Standard, April 29, 2026. → link

Confidence: medium — cabinet approval and headline provisions confirmed across multiple reports; specific allocation mechanisms and the ethical-AI framework text are pending publication.

Anthropic weighs $50B raise at up to $900B valuation

Bloomberg reported on April 29, 2026 that Anthropic is considering preemptive offers to raise roughly $50 billion at a valuation in the $850 billion to $900 billion range. The reporting is sourced to people familiar with the matter; Anthropic has not accepted any offer. A $900B mark would more than double Anthropic's previous round (reported as a $30B raise at $380B in February 2026) and edge past OpenAI's $852B post-money from earlier in 2026. Bloomberg cites annualised revenue run-rate at roughly $30B as last announced, with reporting indicating it has since moved closer to $40B, alongside a count of more than 1,000 enterprise customers each spending over $1M annualised.

What this means. Two things are happening at once and they do not perfectly co-move.

The valuation conversation is about competitive positioning in an unsettled top-of-stack. Anthropic, OpenAI, and Google's frontier-model bet are increasingly read by capital as a small set of commoditising contenders rather than a single winner. Investors offering $900B for Anthropic are betting that the frontier-model market accommodates two or three companies of OpenAI-comparable scale, not one. Whether that bet proves correct is the conditional under all the headline numbers.

The revenue figures, separately, are the part that has changed the most in the last six months. A run-rate moving from announced $30B toward reported $40B in the span of a quarter is the kind of growth that historically accompanies platform-scale repricing. The 1,000+ enterprise customers spending $1M+ each — Anthropic's own statement — is the under-discussed structural fact: this is now an enterprise revenue base, not a developer-API curiosity. The valuation argument rests on whether that enterprise contract base compounds at platform-scale economics or settles into a more typical SaaS growth curve.

The cautious read is that preemptive funding offers at this scale are partly information about what investors are willing to pay rather than what the company is worth. The aggressive read is that the revenue trajectory genuinely supports the multiple. Both readings can be true at once for different windows of the same outcome.

India angle. Two cross-stack implications matter for Indian builders.

SI-layer dependency on the frontier-model duopoly. Infosys announced an Anthropic enterprise partnership earlier in 2026 (covered in prior digests) and a separate OpenAI partnership. TCS's data-centre buildout has OpenAI as anchor tenant. If Anthropic and OpenAI both clear $850B+ valuations on enterprise revenue compounding, the Indian SI integration partnerships are sitting on a revenue base that is now a meaningful percentage of global enterprise AI spend. The strategic exposure runs both ways: pricing power for the frontier-model labs as they scale; revenue-share predictability for the Indian SIs whose offerings are integrated downstream.
Cross-border inference and data residency. Indian BFSI, healthtech, and government deployments using Claude or GPT class models still face the same DPDP cross-border-transfer question regardless of the API provider's valuation. A larger, better-capitalised Anthropic is more likely to invest in India-region inference capacity sooner than a smaller one. Whether the new round funds India-region buildout specifically is the variable Indian deployers should watch in the round announcement, when it lands, rather than the headline number.

For Indian foundation-model labs — Sarvam, AI4Bharat, Krutrim — a $900B Anthropic is competitive context, not direct competition. The category Indian labs operate in (Indic-optimised, cost-efficient, often open-weight) is structurally distinct from the frontier-tier API business at this scale. The capital gap widens; the workload overlap remains narrow.

What this is not. Not a closed round. Bloomberg's reporting names "considering offers" and "very early stage." The announcement that lands could differ materially from the band currently reported, and the pace at which the round closes — TechCrunch sourced two-week timing in follow-up reporting — will be visible directly. Treat the valuation band as a directional reading until the term sheet is public.

Source: Bloomberg, April 29, 2026. → link

Confidence: medium — round under discussion, not closed; valuation band and revenue figures from secondary reporting and prior Anthropic disclosures.

Position movements

Dimension	Direction	Magnitude	Why
regulatory_clarity	+1	2	Maharashtra publishes a concrete state-level AI policy, reducing planning ambiguity for state-resident builders even as the implementation rules are pending.
compute_infrastructure	+1	1	Maharashtra commits to a 2,000-GPU pool exposed to departments as Compute as a Service; movement is conditional on procurement-and-access mechanism.
capital_availability	+1	1	Maharashtra startup grants up to ₹1 Cr (₹1.25 Cr for women-led) marginally widens early-stage capital floor for state-resident AI startups.
foundation_model_capability	0	1	Anthropic round signals continued capital concentration at the frontier; gap to Indian-built foundation models widens at the level of capital, not yet at the level of capability.