What this article argues
- A service carries no lock. Ola reproduced Uber inside a year because the only things to reproduce were an app, a driver network and a discount budget.
- The money explains the behaviour. India spends 0.64% of GDP on research and development, and private industry funds 41% of that against 77% in China.
- Temperament explains nothing. UPI, Aadhaar and Chandrayaan-3 all shipped first from India, and each had a buyer who committed for a decade.
Open your phone in Lucknow tonight. You can hail a cab, order biryani, get a litre of milk in eleven minutes, and pay for all three straight from your bank account without touching a card. Four of those services were invented somewhere else and rebuilt here at a fraction of the price. The rail underneath the payment was built here first, and the rest of the world is still copying it.
That split runs through most of Indian technology. It has a cause you can read off a budget line.
The pattern, with dates
Uber took its first paid rides in San Francisco in 2010. Ola launched in Mumbai in December of the same year, and the gap has never been that small again. Amazon opened in 1994 and Flipkart in 2007. Seamless was taking restaurant orders online in New York in 1999, Grubhub scaled the model from 2004, and Swiggy started in Bengaluru in 2014. Getir invented the ten-minute grocery in Istanbul in 2015;7 Blinkit put ten-minute delivery into twelve Indian cities in August 2021, and Zepto came out of stealth that November.
One line on the chart runs backwards. NPCI switched on UPI in 2016 with 21 banks and 373 transactions in its first month. Brazil launched Pix in 2020. In May 2026 UPI cleared 23.2 billion transactions worth ₹29.9 trillion and now carries close to half of the world's real-time payment volume.6
Swipe the chart sideways to see all of it.
A service has nothing to hold
Consider what Bhavish Aggarwal had to reproduce to build Ola. An app, a payments integration, a call centre, and enough discount capital to keep drivers logged in through the first eighteen months. Uber owned none of the cars, so there was no asset to acquire and no process to reverse-engineer. The product is a coordination layer sitting on vehicles, kitchens and spare rooms that belong to other people.
That structure carries a consequence founders in Bengaluru understood before investors in Menlo Park did. The defensible part of a marketplace is local density: drivers in Koramangala, riders in Gomti Nagar, restaurants within four kilometres of both. Density does not travel. Uber's lead in San Francisco bought it nothing in Bengaluru, so a local team with local money could take the local market while the original was still raising its Series C. India produced Ola, Swiggy, Zepto and Flipkart because the category rewards whoever stands closest to the customer.
Run the same test on a hard-technology product. Copying an EUV lithography machine requires ASML's patents, its supplier network around Veldhoven, and the tacit knowledge of engineers who have spent fifteen years debugging tin-droplet plasma sources. Copying semaglutide requires the molecule, the peptide manufacturing line, and a decade of trial data. The distance between first and second is measured in patents and process knowledge, and neither expires because someone shipped a better app.
A patent is a licence to be slow. A marketplace is a licence to be fast. India has been very good at the second thing for twenty years, in part because it was never asked to be good at the first.
The habit the IT industry taught
India's most successful technology industry sells hours. TCS started writing software for clients in 1968 and Infosys in 1981, and the export services sector they built now bills more than $250 billion a year. That industry has produced world-class delivery engineering, and it has produced almost no owned intellectual property, because the contract hands the client the code. Three generations of Indian engineers, managers and investors learned the same lesson from it: technology is a service you bill for, and the customer keeps the asset.
The habit outlived the business model. When Indian capital moved into products after 2010, it moved into products with the same shape as a service contract, where revenue starts in month three and the risk sits in execution rather than in physics.
India does not lose the copy
Copying carries no shame in this argument, and the Indian versions are often better than the originals. Getir invented quick commerce, then went bankrupt in Austria in 2022 and France in 2023 and retreated to Turkey. Blinkit now runs more than 1,500 dark stores across more than 100 Indian cities, Swiggy Instamart 1,062 across 127, and Zepto over 1,000 across 40.8 Making a ten-minute delivery pay on a basket worth a few hundred rupees is a harder operations problem than making it pay in London, and Indian teams solved it at a population density no Western market offers.
The bill for that solution is visible too. A retailers' union counted more than 200,000 neighbourhood shops closed in a single year as dark stores spread.8 What none of it produces is a technology that someone in another country has to license.
Where the money is
The Economic Survey 2025-26 put India's gross expenditure on research and development at 0.64% of GDP. The United States spends 3.48%, China 2.43% and South Korea 4.91%.1 India has held between 0.6% and 0.7% for more than a decade while GDP more than doubled, so the rupee figure has climbed and the share has not moved.3
Swipe the chart sideways to see all of it.
Composition matters more than the total. Private industry funds 41% of Indian research. In China the business share is 77%, in the United States 75% and in South Korea 79%.1 Most Indian research is therefore paid for by the government, and government research in India stops where the paper is published.
| Country | Business share of R&D | GERD as % of GDP |
|---|---|---|
| South Korea | 79% | 4.91% |
| China | 77% | 2.43% |
| United States | 75% | 3.48% |
| India | 41% | 0.64% |
Swipe the table sideways to see all columns.
The Survey names the failure point without euphemism. Indian institutions perform well at technology readiness levels 1 to 3, the stage of papers and laboratory demonstrations. An industrial economy sells at levels 7 to 9. The stretch between them, prototyping and piloting and characterisation, is where Indian work stalls, and the Survey calls it the valley of death.2 Nobody is funded to cross it. A CSIR laboratory has no mandate to build 500 units and place them in hospitals, and a fund with a four-year horizon will not pay for three years of characterisation before the first invoice.
Venture capital repeats the pattern. Indian startups raised $10.5 billion across every sector in 2025, down 17% year on year, across 1,518 rounds.4 Deep tech has climbed from 4% of Indian venture and private equity activity in 2016 to about 15% in 2025, which is real movement.5 Nasscom and Zinnov then found that 91% of that deep-tech money went to AI companies. The category widened on paper and stayed software underneath.
What the patent count says
The direction of travel is good, and the honest version of this argument has to say so. Filings by Indian applicants grew 19.1% in 2024, a sixth consecutive year of double-digit growth. Indian residents crossed half of all applications at their own patent office for the first time in 2023. India now sits 38th on the Global Innovation Index, up from 81st in 2015.3
Then adjust for the size of the economy. WIPO puts India's resident patent applications per unit of GDP at 381. China's figure is 4,977.9 Chinese residents filed more than 1.5 million patent applications in 2023; Indian residents filed about 50,000.
Swipe the chart sideways to see all of it.
The slope is right. The base is thirteen times smaller, and a base that small takes two decades of compounding to close.
The exceptions that break the culture argument
The comfortable version of this essay blames temperament. Indian engineers want a salary, Indian families want stability, Indian founders want the safe copy. Three counterexamples make that hard to sustain.
ISRO reached Mars orbit on its first attempt in 2014, which no other space agency has managed, for about ₹450 crore. Chandrayaan-3 landed near the lunar south pole in August 2023, the first mission from any country to do so, for about ₹615 crore. UPI, Aadhaar and CoWIN were all built here first and studied afterwards by central banks and health ministries elsewhere.
Each of those projects had one thing in common. A buyer committed for ten years and never asked what the return looked like in quarter three. ISRO has the Government of India. NPCI had the Reserve Bank and a mandate. Give the same engineers that buyer and they ship things nobody has shipped before.
The contentment sits in the capital, not in the character. A founder builds what somebody will fund, and in India the person who would fund eight years of silicon before the first sale has barely existed.
What would have to change
Money has started to move. The government operationalised a ₹1 lakh crore Research, Development and Innovation fund in November 2025 and capitalised the Anusandhan National Research Foundation with ₹14,000 crore from the centre.3 The India Deep Tech Alliance has committed $2.5 billion, including $1 billion earmarked for AI startups over three years.5 Micron's assembly and test plant at Sanand and the Tata-PSMC fab at Dholera put silicon capacity on Indian soil for the first time. The Economic Survey has proposed Translational Research Centres to pay for the prototyping stage that kills most Indian research.2
All of that is input. The output test is narrow and slow. By 2035, does an Indian company own a process, a molecule or a machine that a competitor in Shenzhen or Eindhoven cannot reproduce without a licence? Nothing in the app economy answers that question, however well it runs.
Tonight the phone in Lucknow works better than the phone in San Francisco. Almost none of what makes it work belongs to anyone here. Changing that costs a decade of somebody's patience, and the decade has to be paid for before it starts.