The Second Layer

On the private infrastructure India built while no one was looking

In 2025, a single payments rail in India processed more money than the entire annual economic output of France.

UPI – the Unified Payments Interface – handled 228 billion transactions worth nearly $3.66 trillion in the calendar year. France’s nominal GDP, by comparison, stands at approximately $3.16 trillion. The numbers are not quite the same currency, but the order of magnitude is striking enough to survive the conversion: India has built a payments infrastructure that moves, in a single year, more value than one of the world’s largest economies produces in that same period.

This is, by any measure, an extraordinary fact. And yet it is only the beginning of the story we want to tell.

UPI is the road. What interests us is who built the intelligence about how to use it – and what that intelligence is now worth.

Layer One: What the Government Built

The story of India’s digital infrastructure is, in its first chapter, a story about the state doing something remarkable. In the span of roughly a decade, the Indian government assembled a digital public infrastructure of a scale and sophistication that has no precedent in economic history.

Consider the numbers. Over 1.4 billion Indians are enrolled in Aadhaar, the world’s largest biometric identity system, which now processes over 27 billion authentication transactions per year. Mobile connections stand at 1.12 billion, with over 92% classified as broadband. Average monthly data consumption per user has surged from 2.8 GB before 2016 to 27.5 GB by 2024 – a near ten-fold increase in less than a decade, driven by Jio’s disruptive entry and the rapid rollout of 4G (and, now, 5G) networks. India today has over 800 million internet users, a number that continues to grow at 6–8% annually.

In 2025, UPI surpassed Visa in daily transaction volume, processing over 640 million transactions a day and accounting for approximately 49% of all global real-time payments. ONDC – the Open Network for Digital Commerce – grew from 1 million monthly transactions in early 2023 to over 18 million by October 2025, connecting over 500,000 sellers across 1,200 cities. Running through all of it is a gig and platform economy that now employs approximately 12 million workers, a figure projected to reach 23.5 million by 2030.

This is Layer One. It is visible, celebrated, and extensively documented. The IMF has recognised UPI as the world’s largest retail fast-payment system. Delegations from across the world visit India to study the India Stack.

We do not want to diminish any of this. It is genuinely extraordinary. But it is not, in our view, where the most interesting value is being created.

Governments build roads. The companies that learn how to navigate those roads with intelligence that maps every pothole, every shortcut, every pattern of congestion often end up worth more than the roads themselves.

Layer Two: What Private Companies Built on Top

Sitting on top of India’s public digital infrastructure, largely invisible to the naked eye, is a second layer of infrastructure that is, in many ways, more valuable and far harder to replicate.

This second layer is not made of fibre or spectrum or government mandates. It is made of data – a decade’s worth of transactions, preferences, behaviours, and patterns accumulated by a handful of companies that were present at the right moment and had the patience to stay. It is made of physical networks built using that data. It is made of operational knowledge encoded into systems that took years to calibrate.

No government policy created it. No competitor can buy it. It accrued, quietly and cumulatively, through millions of individual interactions with Indian consumers across thousands of geographies, in a market of extraordinary complexity.

Our portfolio companies are, each in their own way, custodians of this second layer. And the market, in our view, has not yet fully reckoned with what it is worth.

Delhivery: The Map No One Else Has

Logistics in India is a problem that defeats easy description. The country has 19,000+ serviceable pin codes. Delivery density varies by orders of magnitude between a south Mumbai neighbourhood and a small town in eastern Uttar Pradesh. Road infrastructure, last-mile access, customer availability windows, return rates, cash-on-delivery behaviour – each of these variables looks different depending on where you are, and none of them can be reliably inferred from a satellite image or a government database.

Delhivery has spent years building the map. Not a geographic map – those exist. An operational map: which routes are reliable at which times, which pin codes have high return rates for which product categories, which delivery windows maximise first-attempt success in which areas. This is the kind of intelligence that can only be accumulated through hundreds of millions of actual deliveries, made by actual drivers, in actual conditions.

A new entrant to Indian logistics can hire drivers and lease warehouses. They cannot, in any reasonable timeframe, replicate what Delhivery knows about how goods actually move across India. The financial expression of this is a widening cost advantage. Delhivery’s cost per shipment continues to decline relative to peers – not because it is cutting corners, but because its network intelligence allows it to operate more efficiently at every node. In a category where cost is ultimately destiny, this gap is compounding.

Policybazaar: A Decade of Insurance Behaviour

India’s insurance market is chronically under-penetrated – less than 4% of GDP, against a global average closer to 7%. The reasons are well-understood: low trust in intermediaries, complex products, and a consumer base historically approached through agents with misaligned incentives. What is less well understood is the nature of the data asset that has accumulated at Policybazaar as the dominant aggregator across this market.

When a consumer buys insurance through Policybazaar, the company learns something. It learns how sensitive that consumer is to price at different income levels. It learns which product features were decisive. It learns the geography, the life stage, the existing coverage. It learns, critically, whether the policy is renewed – and if not, why not. Multiply this across more than 90% of India’s online insurance traffic, sustained over a decade, and you have something that no traditional insurer and no new entrant possesses: a comprehensive, longitudinal dataset of Indian insurance behaviour.

This dataset is not on any balance sheet. It does not appear in any standard valuation metric. But it is the reason Policybazaar’s customer acquisition costs are lower than any competitor’s – because it knows, better than anyone, which customer is worth acquiring, at what price, and how to retain them. It is the reason the renewal book, now growing at scale, generates gross margins that are structurally different from anything visible in the early years of the business.

Lenskart: The Store That Knew You Before You Walked In

Eyewear in India was, for most of its commercial history, a category defined by low trust and inconsistent quality. The neighbourhood optician knew his craft but not his customer. There was no brand with the scale to generate data, and no data to generate competitive advantage.

Lenskart changed this through a sequence of moves that, in retrospect, looks inevitable, but required genuine conviction at each step. The company first built its digital presence, capturing prescription data, style preferences, and purchasing behaviour at scale. Most retailers would have stopped there. Lenskart used the data to go offline – but not in the way traditional retailers go offline.

When Lenskart opened physical stores, it did not open them where real estate was cheapest or footfall was highest. It opened them where its data told it customers already existed: micro-markets where order density justified a presence, neighbourhoods where consumers had already demonstrated willingness to buy. Each new store opened with a pre-existing customer base. The data had done the market research that traditional retailers pay consultants to approximate.

The physical presence then did something the digital channel could not: it built trust. For a product as personal as eyewear – worn on the face, requiring precise measurement, touching vision itself – the ability to try a frame matters enormously. That trust, built in stores, drove repeat purchases across both channels.

Centralised manufacturing completed the model. By producing at scale through its own facilities, Lenskart achieved mass customisation at low cost. Every lens ground to prescription, every frame finished to specification, but through a process that grew cheaper as it grew larger. The result is a business with structurally expanding margins, a loyal subscription base through its Gold programme, and a physical retail network that any competitor would need years and billions of rupees to replicate.

When Lenskart listed in November 2025, it brought to public markets something genuinely rare: a business that had redesigned an entire category from the ground up, using data as the primary architectural material.

Ixigo: The Intelligence Layer on Indian Railways

Indian Railways carries over 23 million passengers a day – more than the combined population of Sweden, Denmark and Norway. It is the world’s fourth-largest rail network, and navigating it – understanding availability, predicting waitlist movement, knowing which trains actually run on time – is a problem of extraordinary complexity.

Ixigo has spent years accumulating the intelligence layer that sits on top of this system. Its waitlist prediction models, built on hundreds of millions of historical booking outcomes, are materially more accurate than anything a first-time entrant could construct. Its understanding of travel demand patterns across India’s tier-2 and tier-3 cities – which routes surge during which festivals, which fare classes move first – is encoded into systems that took years to calibrate. This is why Ixigo commands the loyalty of India’s value-conscious traveller in a way that well-funded competitors have found difficult to displace. The product is better because the underlying intelligence is deeper.

The Second Layer in the Age of AI

There is a version of the AI story that is, at its core, a story about commoditisation. Large language models trained on public data can replicate many things: customer service interactions, content generation, generic recommendations. The businesses most exposed to this version of the story are those whose value proposition rests on capabilities that AI can approximate at low cost. That is a real risk for many companies. It is not, in our view, a risk for the businesses in our portfolio.

The reason is precisely the second layer. AI systems are only as valuable as the data they are trained on. Public data – scraped from the internet, drawn from open databases, synthesised from available sources – produces general intelligence. What it cannot produce is specific intelligence about Indian insurance renewal behaviour in tier-2 cities, or optimal last-mile routing in the 19,000+ pin codes of the Indian logistics network, or the prescription and style preference patterns of Lenskart’s ten million customers, or the waitlist dynamics of Indian Railways across fifteen years of booking history. That data does not exist in any public corpus. It exists only inside our portfolio companies, accumulated through years of transactions that no model can reconstruct from the outside.

This means the AI era, far from eroding the second layer, is likely to amplify its value. As AI tools become cheaper and more powerful, the scarce input shifts from the algorithm to the data. A company that owns a decade of proprietary behavioural data and now has access to world-class AI infrastructure to extract signal from it is in a structurally stronger position than it was five years ago – not a weaker one. Policybazaar can build underwriting models that no insurer could replicate without its renewal book. Delhivery can optimise routing in ways that are unavailable to any competitor without its delivery history. Lenskart can predict what a customer will want before they know themselves. The second layer, in short, is the raw material that makes AI compounding rather than commoditising for these businesses.

We are already seeing early expressions of this. Each of our portfolio companies is actively deploying AI across its operations – in fraud detection, in demand forecasting, in personalisation, in cost optimisation. In each case, the AI is being trained on data that only that company possesses. The competitive advantage is not the model, which any well-resourced competitor can license. The competitive advantage is the data the model is trained on, which no competitor can replicate. This is, we believe, one of the most important and underappreciated dimensions of the investment case for India’s leading digital franchises.

Why This Matters Now

We have held these businesses through a period when markets were not pricing the second layer. The 2022 correction was brutal for new-age digital stocks across India – and in some cases, entirely justified. There were businesses in the broader category that had no second layer: companies with scale but no intelligence, with users but no retention, with revenue but no improving unit economics. Markets were right to reprice those.

What markets were less precise about was the distinction between businesses that were structurally challenged and businesses that were structurally sound but temporarily mispriced. Our portfolio companies fell into the second category. The data assets, the network intelligence, the physical infrastructure built on top of digital insights – none of this disappeared during the correction. In many cases, it continued to compound.

What is changing now – and this is the development that most excites us as we write this letter – is that the second layer is beginning to express itself in the financial results. Operating leverage, the long-promised reward for building fixed costs early, is arriving. Unit economics are improving not because these companies are cutting their way to profitability, but because the intelligence they built is making them more efficient at scale. Incremental customers cost less to acquire. Renewal rates compound. The gap between reported profitability and underlying economics, which was the source of so much confusion during the 2021 listings, is narrowing. One of the best examples of this is PB Fintech.

For years, we told investors to look past the P&L and into the cash flows. We are now watching the P&L begin to tell the same story the cash flows have been telling for some time.

The market’s historical habit of discounting these businesses – applying cautious multiples to uncertain economics – was, in many cases, a rational response to genuine uncertainty. That uncertainty is resolving. The economics are becoming legible. And legible economics, for dominant franchises with genuine second-layer infrastructure in large and growing markets, attract a different kind of investor at a different kind of multiple.

We are not making a market call. We are observing a structural shift in the quality and visibility of earnings being produced by businesses we have spent years understanding.

Looking Ahead

The public infrastructure India built is extraordinary. Aadhaar. UPI. ONDC. A broadband network reaching 800 million people. A gig economy employing 12 million. These are achievements of genuine historical significance.

But infrastructure, by its nature, becomes available to everyone. The companies that will generate the most durable returns are not those that merely rode the infrastructure, but those that used it to build something on top that is uniquely theirs – a second layer of intelligence, data, and operational knowledge that compounds quietly, accrues over years, and cannot be replicated by a well-funded new entrant in any reasonable timeframe.

That is what our portfolio companies have built. That is why we own them. And that is why we believe – with greater conviction today than at any point since inception – that the most interesting part of this story is still ahead.

Best Wishes,

Ravi Srivastava