COMPETITIVENESS AND POLICY · 4 JUNE 2026 · 12 MIN READ

India's Startup Boom Is Real. Its Innovation Deficit Is Larger.

India has 191,000 DPIIT-recognised startups and ranks just behind the US and China by raw count. However, its position drops to 22nd per capita and 4th in output quality (per StartupBlink). Despite $11 billion in venture funding in 2025, much of India's ranking is driven by scale rather than startup quality—though the numbers still reflect a rising technology powerhouse. This is reflected, to some extent, in withdrawals from the Indian equity market over the last two years; foreign portfolio investors withdrew a record $18 billion from Indian equities in 2025 alone — the largest annual outflow in the country's capital market history. South Korea, whose economy is less than a third of India's size, overtook India to become the world's sixth-largest equity market in early 2026. Its market capitalisation surged 86% in a single year.

These two facts — the startup formation boom and the capital flight — are not in contradiction. These are some thoughts on how they are connected.

 Tararua Economics’ mapping of India's active startup universe — drawn from DPIIT recognition data cross-referenced against MCA filings — puts the operational number at between 85,000 and 105,000 entities. The headline 191,000 figure includes dormant registrations, pre-revenue shells, and single-founder operations. The commercially active population is roughly half. Whilst this is still a substantial ecosystem, deeper insights come when one looks at how it is structured. The two largest groups by entity count are Technology-Enabled Process Execution (18,000–22,000 entities) and Consumer Applications (15,000–20,000 entities). Together, they account for roughly 40% of the commercially active universe. SME software and marketplace platforms add another 20,000–30,000. The AI-Centric category — the one that might credibly claim original technology creation — sits at 5,000 to 7,000 entities, of which fewer than 300 have raised meaningful institutional capital. B2B Industrial and Deeptech, the category most likely to generate genuinely differentiated IP, numbers 8,000–12,000 entities and is, by the report's own assessment, the segment where the funding gap relative to formation is widest. 

In short, the ecosystem skews heavily towards process automation, friction removal, and consumer convenience. The technology is real. The innovation, in the deepest sense of the word, is thinner than the numbers suggest.

Here is the uncomfortable observation I keep returning to. Much of what India calls its startup innovation is, on close inspection, a sophisticated redeployment of labour arbitrage — the same structural bet that built the IT services industry, now applied to the digital consumer economy. Quick commerce scales because delivery riders are cheap. Platform-based services — home cleaning, plumbing, food preparation — aggregate affordable labour and present it through a digital interface. Consumer apps monetise a demographic of 65% under-35 users whose time costs less than infrastructure automation would. The AI layer being inserted into most of these businesses is largely friction reduction and personalisation — applying existing models to existing workflows. I do not dismiss this. These platforms deliver real economic value. Zepto's $1.3 billion raise in 2024 reflected genuine market demand. Consumer services attracted $1.8 billion in funding. The unit economics of fintech consumer platforms — Zerodha, Groww — have proved durable. But none of this constitutes original technology creation that migrates the productive frontier. It constitutes technology-enabled service delivery at Indian cost structures.

One analyst put it with precision I have not been able to improve upon: "India has become obsessed with services because they are easy to roll out, do not require deep R&D, and involve no IP — only processes that can always be replicated and improved. Indian IT services companies have developed wizards to aid their service delivery, but have rarely attempted to build products." This observation was made about IT services in 2025. It could have been made about India's startup ecosystem as a whole.

In my time working with financial institutions in emerging markets, the observation that stays with me is that the sectors that most need next-generation technology transformation in India — banking, insurance, healthcare, retail, and automotive — are being served by platforms that digitalise existing processes, not ones that reimagine or innovate the underlying operating model. Take banking, for example: India has built the India Stack, UPI, and the world's most sophisticated real-time payments infrastructure. What it does not have — with the partial exception of a few neo-banking and credit-decisioning players — is a cohort of companies building the next generation of core banking, trade finance, and risk architecture for the Indian and emerging market context. The momentum is focused on processing layer digitisation, but the intelligence layer is getting lost. The same story continues on in healthcare with 5,000+ healthtech entities focused on appointment booking, pharmacy delivery, and teleconsultation — friction removal. The diagnostics and AI-driven clinical decision support segment (Qure being the standout exception) remains thin relative to India's disease burden and the scale of the opportunity in secondary and tertiary care. The consistent pattern across these sectors is the same: digitalise distribution, automate front-end friction, leave the productive core of the industry unaltered. It has some value, but it is not the productivity transformation these industries need, nor is it sufficient for India to claim its position in the emerging economic landscape.

 There is a history here that deserves honest examination because it illuminates how India has periodically stood on the threshold of original enterprise software creation — and how it consistently gets lost. In the late 1990s and early 2000s, I worked with some of the best software India produced, a genuinely original cohort of enterprise financial technology companies. i-Flex Solutions built Flexcube – a core banking platform that ultimately reached 700 installations in 125 countries, winning The Banker's Core Banking Solution of the Year. Polaris Software, born from Citibank's India operations, built a suite of financial technology products for the global BFSI segment. Infosys built Finacle. These were not process outsourcing businesses. This was IP that competed directly with Temenos, Misys, and Sungard in global markets and was beating the best in the industry. But then most of them cashed out. Oracle acquired i-Flex in 2005. Flexcube became Oracle Financial Services and ceased to exist as an independent Indian entity. Polaris eventually spun out its products division as Intellect Design Arena — one of the few survivors of the original cohort still operating as an independent, India-headquartered financial technology product company. The rest of the landscape was absorbed, displaced, or retreated to service delivery.

I am not suggesting that those acquisitions were errors — in isolation, they were rational transactions for the companies involved. But taken together, they represent a structural moment at which India's most promising enterprise software IP creation migrated into global incumbents' portfolios rather than scaling into independent global challengers. What it tells me is that capability exists. However, the institutional and capital infrastructure to sustain independent IP creation at a global scale does not.

What is striking is that the current generation of Indian enterprise tech — the 18,000–22,000 process execution companies and the 5,000–7,000 AI-centric entities — is not obviously building towards a moment that would produce the next i-Flex. The funding flows confirm this. Enterprise tech attracted $1.8 billion in 2024, but the dominant trend in that category is incumbents retrofitting AI layers into existing process tools, rather than new entrants building category-defining architectures.

 For my sins, during the dot-com era, I had the opportunity with EDS and Kearney to gain decent exposure to boardroom conversations at Indian tech firms and to know that this is not a failure of capability. India's engineers are genuinely world-class — the GCC (Global Capability Centre) expansion in Bengaluru and Hyderabad is driven precisely by global firms recognising the technical depth. The failure lies in incentive architecture and institutional design. The returns in building services are faster, more predictable, and less capital-intensive than those from building products. Founders respond to incentives. Investors price risk. The result is a rational equilibrium that is structurally suboptimal at the national level.

 A comparison to Korea may seem, on the surface, gratuitous and inappropriate, but I think it is instructive and spans a time horizon that matters. South Korea's semiconductor industry exported $173.4 billion in goods in 2025 — a record high, representing 20.8% of the country's total export volume. In March 2026, semiconductor sales surged 151.4% year-on-year to $32.83 billion in a single month. SK Hynix reported a 49% operating margin. The AI infrastructure cycle — the demand for high-bandwidth memory, advanced chips, and AI-capable server components — has rewarded, at an extraordinary scale, the technology bets Korea began making in the 1980s and deepened over every subsequent decade. The result is that South Korea's listed companies surged to a $5 trillion market capitalisation in early 2026, overtaking India's $4.8 trillion. India's economy is more than twice the size of Korea's. India has a far larger talent base. India is growing faster in GDP terms. And yet global capital markets have just repriced Korea above India — not because Korea's macroeconomics are superior, but because Korea is positioned inside the AI hardware cycle and India is positioned outside it.

What the capital markets are doing, in their blunt way, is revealing a structural reality that macro growth numbers tend to obscure: in a technology-driven investment cycle, it is not the size of your economy but the nature of your technology base that determines where global capital allocates. Korea made the deep tech bet when it had India's skill base and less than India's scale. It absorbed significant capital risk over decades — through Samsung's near-collapse in the 1990s, through the dot-com correction, and through the 2008 cycle. The compounding is now visible in one year's worth of AI-driven export data.

Foreign portfolio investors withdrew a net $18.4 billion from Indian equities in 2025 — the largest annual outflow on record and the second-largest in absolute terms in Indian capital market history. This followed a dramatic October 2024, in which Rs 114,445 crore of net equity selling occurred in a single month. The immediate causes are well-documented and real: rupee depreciation, stretched valuations relative to earnings growth, dollar strength, and global portfolio rebalancing towards US AI-adjacent equities. These are not illusions. They are the immediate drivers that any FPI risk committee would cite. But there are deeper structural factors, too, that sit underneath the visible ones. The global AI capital cycle is concentrating in hardware — semiconductors, HBM, AI infrastructure — and in large-cap software platforms directly monetising AI. India's listed market has almost no exposure to either. It lacks listed semiconductor companies. Its IT services majors — TCS, Infosys, Wipro, HCLTech — are facing structural displacement of their services delivery model by the same AI tools their clients are deploying. The consumer-facing platforms are richly valued relative to earnings. What India has in demographic scale and domestic demand, it lacks, from the perspective of a global technology investor repositioning around the AI structural shift.

 

I want to be precise about what this argument is not claiming. It is not claiming that Indian technology is unsophisticated. It is not claiming that no original innovation is occurring. And it is not claiming that the current trajectory is irreversible. There are genuine signals of deeptech formation beginning to accelerate. The B2B Industrial and Deeptech cohort numbers 8,000–12,000 entities. The IndiaAI Mission, PLI schemes for semiconductors and EVs, and state-level deeptech accelerators are beginning to close the structural funding gap — slowly, but observably. Krutrim, Sarvam AI, and QpiAI represent genuine foundation model and AI infrastructure bets being made in India by Indian founders. The 209 AI-in-SaaS companies founded in 2024 was the highest single-year formation rate in a decade. Advanced manufacturing startups have increased nearly tenfold over a five-year period. The Global Innovation Index shows India's rank rising from 81st in 2015 to 40th in 2023. Patent filings grew 31.6% in 2022 — the highest growth rate globally. These are real improvements, not statistical noise.

But the rate of improvement and the absolute level of R&D intensity sit too far apart for comfort. India improved from 81st to 40th in the Global Innovation Index over eight years. Korea sits at the top of the R&D intensity table, with no signs of softening its commitment. The gap between where India is and where Korea was when it made its deeptech bets is navigable — but navigating it requires a different level of policy urgency and private-capital risk appetite than is currently evident. 

Indian founders are not choosing consumer apps over deeptech because they lack the capability to do deeptech. They are choosing consumer apps because the payback timeline is shorter, the market is visible, the funding path is better-trodden, and the exit options — IPO, strategic sale — are more legible. A founder considering whether to spend seven years building an original healthcare diagnostics platform or eighteen months building a consumer health app has a rational individual calculus that points towards the app. The aggregate of those rational individual decisions produces the category mix we see in the formation data. The same logic applies to capital. Investors who allocated to the Indian edtech wave — over $10 billion over the last decade — and absorbed the BYJU's collapse now have an institutional memory of long-cycle consumer-tech risk. The rational response is to prefer faster-cycle business models. Deeptech gets underfunded not because investors are unsophisticated but because the risk-adjusted return profile is genuinely inferior in the current incentive environment. 

Changing this requires mission-oriented intervention — patient capital, government procurement commitments that de-risk the revenue line for early-stage deeptech companies, and R&D tax incentives calibrated to drive private sector participation towards 50% of GERD, rather than the current 36%. India's government has pointed in this direction with the PLI schemes and the IndiaAI Mission. The instruments exist. The scale and urgency are the question. India's startup ecosystem is, by any reasonable global benchmark, impressive. The formation velocity, the domestic capital deepening, the profitability inflection — 31% of surviving Indian startups reported positive EBITDA in 2024, up from 18% in 2021 — these are genuine structural improvements. The problem is not that India lacks a startup ecosystem. The problem is that the ecosystem is optimised for the wrong outputs at this stage of the country's development. Labour arbitrage dressed in AI tools produces employment and convenience. It does not produce the productivity transformation that banking, healthcare, insurance, and manufacturing require. And it does not produce the technology IP that global capital markets, in the current investment cycle, are pricing at a premium.

The Korea inflection point is not a distant analogy. It happened in living memory, in a country with comparable starting conditions, and its consequences are now visible in a single year's worth of semiconductor export data. The AI hardware cycle has made the technology investment thesis explicit in a way that no amount of macro growth narrative can obscure.

I watched the i-Flex story unfold — a genuinely world-class enterprise software product, built in India, deployed across 125 countries, then absorbed into a US technology incumbent. That arc has repeated, in quieter forms, across the Indian enterprise technology landscape for two decades. The question I find myself sitting with is whether the current generation of Indian deeptech formation — the Krutims, the Sarvams, the AI diagnostic platforms — will find the institutional and capital environment to compound independently, or whether the structural incentives will, again, route the best outcomes toward acquisition rather than independent scale.

The talent is unambiguously there. The question has always been the architecture around it.

 

#India #Technology #Innovation #Startups

Discuss this analysis. Tararua Economics briefs governments, sovereign institutions and investors on these questions. Contact