India AI Thematic Report
India AI: share prices have reset even as policy and private funding expand the compute-and-data foundation—creating a 12–24 month set-up for clearer, product-like AI revenue to matter more.
India’s AI story is moving in two directions at once. On the ground, capacity is being built quickly: more affordable compute, larger public datasets, and faster pathways for startups and institutions to test real use-cases. In listed markets, the closest “AI-linked” companies are still mostly IT services and engineering firms—business models where AI often shows up first as productivity and delivery efficiency, not as a standalone product line.
That gap is why the recent price action looks confusing at first glance. Across a basket of listed Indian technology and engineering names with meaningful AI exposure, one-year returns are broadly negative and highly dispersed (Exhibit 1). This is less about “AI enthusiasm disappearing” and more about investors asking a harder question: where does AI translate into measurable revenue, and who keeps the benefit (customer vs vendor)?
Policy is now changing the input economics. Parliamentary disclosures show IndiaAI’s compute program has scaled from 34,381 GPUs (July 2025) to 38,231 GPUs (December 2025), with subsidised pricing disclosed around ~₹65 per GPU-hour on average (and ₹92 per GPU-hour for H100). That scale is large enough to change experimentation costs, even if it does not automatically show up in listed earnings immediately. The near-term constraint is shifting from “can you access compute?” to “can you build distribution, data rights, and governance that buyers trust?”
EXHIBIT 01
Listed India Tech/Engineering AI Exposure: one-year returns show Broad Weakness and High Dispersion
Source : Screener (market data), compiled
EXHIBIT 02
IndiaAI Outlay Split by Pillar: Where Policy is Putting Money (₹ Crore)
Source: Lok Sabha disclosure (IndiaAI pillar allocations), compiled
Industry Structure
India’s AI market is easiest to understand as three stacked layers, each with different economics and different timelines for showing up in public earnings.
1) Infrastructure & Policy Layer: The Shared Foundation
IndiaAI is building the enabling layer: compute access, datasets, skills, and ecosystem financing, plus early guardrails under “Safe & Trusted AI.” The practical takeaway is simple: testing and prototyping costs are coming down, which increases the number of credible teams that can build India-relevant AI applications.
Parliamentary disclosures also reference AIKosh (beta launched March 2025), described as hosting 890+ datasets alongside models/toolkits. That matters because in many verticals (public sector services, Indian languages, local compliance), data availability and quality can be a bigger gating factor than model selection.
What policy builds vs what companies must earn
2) Private Application Layer: Outcome-First Businesses
Private funding is increasingly focused on applications that can prove ROI. TechCrunch (citing Tracxn data) reports Indian AI startups raised just over $643m across 100 deals in 2025, largely in early and early-growth stages—consistent with an application-first market. NASSCOM estimates cumulative funding in Indian GenAI startups at ~$990m by H1 CY2025, up ~30% YoY—meaning momentum is real, but still small relative to global leaders.
One signal worth noting: Google and Accel announced a co-investment program—up to $2m per startup for at least 10 early-stage AI startups—which aligns with a world where distribution and platform access matter as much as model performance.
3) Listed Monetisation Layer: Services-led Exposure Today, Product-like Exposure Tomorrow
Listed “AI exposure” is mostly indirect today. It tends to show up through:
IT services (AI-enabled delivery, automation, copilots embedded in projects)
Engineering services (embedded intelligence in auto/industrial programs)
A smaller set of product/analytics businesses where AI can become a repeatable, packaged offering
A useful way to organise listed names is into four groups:
Large IT services: strong cash generation; AI often shows first as productivity and efficiency
Faster-growing mid-cap services: more operating leverage; more sensitive to demand and sentiment
Engineering/embedded: tied to program cycles and customer concentration
Products/analytics: closer to repeatable revenue if distribution and retention are proven
EXHIBIT 03
Three-Year Returns: Evidence of Divergence within the same “AI Exposure” Bucket
Source : Screener (market data), compiled
Where private AI is most likely to show up in listed revenue (12–24 months)
EXHIBIT 04
Valuation Premia are Concentrated: EV/EBITDA Spread Highlights where Expectations are Already Priced
Source: Screener (valuation), compiled
EXHIBIT 05
Operating Margin Template by Segment: A Simple Lens on Business-Model Differences
Source: Screener (financials), compiled
Structural Inflection
What is changing now (and why it matters)
Compute access is scaling to meaningful levels. IndiaAI disclosures show 34,381 GPUs onboarded (July 2025) and 38,231 GPUs (December 2025), with subsidised pricing disclosed around ~₹65 per GPU-hour on average and ₹92 per GPU-hour for H100.
Datasets are becoming more usable. AIKosh is described as hosting 890+ datasets, reducing a common bottleneck for India-specific deployment.
Share prices are forcing discipline. The one-year drawdown is broad (Exhibit 1), while three-year dispersion remains meaningful (Exhibit 3). The market is rewarding evidence—conversion rates, renewal behaviour, and credible disclosure—rather than the label alone.
Capital is becoming more selective. Ownership shifts suggest sponsorship is concentrating rather than lifting the whole basket (Exhibit 8).
The path to public markets is starting to appear. Fractal Analytics filed a DRHP with SEBI (Aug 25, 2025), while Qure.ai communicated IPO intent in ~two years (May 2025), indicating early formation of public-market readiness in AI/analytics (Exhibit 9).
EXHIBIT 06
Compute Democratization Milestones: A Dated View of Policy Build + Ecosystem Signals
Source: Lok Sabha disclosures; major ecosystem announcements, compiled
EXHIBIT 07
Anchor Diagnostic: Growth, Valuation, and One-Year Returns do not move in a Straight Line
Source: Screener (market + financial data), compiled
EXHIBIT 08
Ownership Rotation: Foreign Ownership Changes are Concentrated, not Uniform
Source: Screener (shareholding trends), compiled
IndiaAI KPI dashboard: what to watch for real earnings translation
Capital Cycle Panel
Public
Listed markets are in a “prove it” phase. One-year returns are weak and uneven (Exhibit 1), and the growth/valuation relationship is not clean (Exhibit 7). Investors are focusing on demand visibility, disclosure credibility, and whether productivity gains stay with vendors or are passed through to clients.
Private
Private funding is selective and skewed to earlier stages, supporting outcome-led applications rather than capital-heavy model build. The distribution signal (platform-linked capital and partnerships) is consistent with a next phase where go-to-market and data rights matter more than raw compute access.
IPO
The first signs of a public-market pathway are emerging. Fractal Analytics filed a DRHP with SEBI (Aug 25, 2025), and Qure.ai has communicated IPO intent on a multi-year timeline (May 2025), highlighting the early formation of “IPO-ready” AI/analytics businesses (Exhibit 9).
EXHIBIT 09
Early Public-Market Signals: AI/Analytics Issuers moving toward Formal Pathways
Source: SEBI filings; Reuters reporting, compiled
Valuation & Capital Implications
1) Valuation Premia now need Clearer Proof
The key shift is not that the AI theme is weakening—it is that markets are demanding clearer proof of how AI becomes durable earnings. Exhibit 7 is the simplest way to see this: strong multi-year growth does not automatically protect one-year returns when investors worry about conversion, pricing, and cost pass-through.
2) Ownership Concentration Raises the Bar for Disclosure and Execution
Ownership shifts (Exhibit 8) suggest sponsorship is becoming more concentrated. That tends to increase the premium placed on consistent disclosure (bookings, retention, and margin outcomes) and on evidence that companies can invest in product-like offerings without destabilising cash generation.
3) Policy-Driven Supply Expansion Increases Competition in Applications
As compute and data access become easier, competition shifts to distribution and trust: who has enterprise channels, proprietary workflows, and compliance readiness. Over 12–24 months, this typically expands the overall market, but it also makes undifferentiated offerings harder to defend.
Services economics: simple indicators that AI is improving the business model
Risks & Invalidation
Execution risk (policy): onboarding is necessary; utilisation, allocation transparency, and operational throughput determine impact.
Demand risk: pilots may not convert to scaled programs if ROI is unclear or budgets tighten.
Pricing risk: if productivity gains are competed away, AI can lift delivery efficiency without lifting earnings quality.
Disclosure risk: if “AI revenue” remains non-standard and unauditable, markets may continue to discount the theme in listed proxies.
Governance risk: safety, privacy, and auditability requirements can raise costs and lengthen sales cycles in regulated verticals.
Invalidation trigger: if large IT services growth re-accelerates meaningfully without clearer AI-linked commercial metrics and retained productivity, the near-term rotation toward more product-like earnings becomes harder to defend; similarly, if policy-led compute scaling stalls materially, the broader ecosystem timeline extends.
CTA