The Middleman's Republic

IIn which the author examines, with as much good humour as the subject permits, why the world's most populous nation imports its ingenuity by the pallet, exports its geniuses by the planeload, and taxes at 28 per cent anyone foolish enough to attempt originality in between.

Or: How We Learnt to Stop Inventing and Love the Container Ship

Prologue: The Parable of Karthik's Circuit Board

A friend of mine — let us call him Karthik, because that is his name and he has given up being embarrassed about this story — decided last year to build an electronic device. Nothing seditious. A smart air-quality monitor, of all things: a gadget to inform Indians, with laboratory precision, exactly how unbreathable their air is. A growth market if ever there was one.

Karthik is an engineer of the capable, unglamorous sort — the kind who reads datasheets for pleasure and believes, touchingly, that a good product is mostly a matter of building it. He began, as a patriot might, by attempting to source his components locally. What followed was an education.

The first Indian distributor did not reply to his email. The second replied after nine days to inform him that the particulate sensor he wanted carried a minimum order quantity of five thousand units; Karthik needed twenty. The third quoted a price roughly triple the global rate for what turned out, upon inspection, to be a unit of dubious provenance with the original markings sanded off. A fourth gentleman on SP Road offered, with the weary magnanimity of a man who has seen too many dreamers, to “arrange anything from China, two weeks, cash.”

Which is when Karthik, at 1:40 in the morning, messaged a supplier in Shenzhen directly. The reply arrived in fourteen minutes. It contained the datasheet, the price in three currency options, a gentle correction to his footprint layout, and a question about whether he would prefer conformal coating. Within forty-eight hours a factory engineer — an actual engineer, employed by the actual factory — had reviewed his Gerber files free of charge and flagged two design-for-manufacture issues. Ten assembled prototypes were on a DHL flight within nine days.

They then arrived in India, where the real adventure began.

The parcel that had crossed the Pearl River Delta, the South China Sea, and two international borders in four days proceeded to spend the next three weeks in the custody of Indian customs. There was a query about BIS registration — for prototypes, mind you, ten units, not for sale, clearly declared. There was the small matter of the duty stack, which on electronics of this sort runs to roughly 28 per cent once basic customs duty and IGST have finished with you. And there was, eventually, the delicate cough of the customs house agent, who suggested that a modest “clearance facilitation charge” might resolve the BIS ambiguity rather faster than correspondence would.

Karthik paid. Everybody pays. Transparency International's Global Corruption Barometer found that 39 per cent of Indians had paid a bribe for a public service — the highest rate in Asia, comfortably ahead of Cambodia — and that nearly half of those who paid were asked to. On the watchdog's 2025 Corruption Perceptions Index we scored 39 out of 100, ranked 91st, which at least has a pleasing numerical symmetry to it.

Here is the detail I cannot get past. The government of the People's Republic of China — a state with which we share a disputed border, a trust deficit, and the occasional fatal skirmish — moved Karthik's goods with courtesy and speed. The government of his own republic held them hostage for tea money.

Now, one anecdote is not data, and I would not hang a thesis on Karthik's twenty sensors. But I have come to believe his parcel is a diagnostic instrument of some precision. Trace its journey and you trace the whole architecture of the problem: an ecosystem that cannot supply an inventor, a bureaucracy that suspects him, a duty regime that taxes his curiosity as though it were contraband, and — across the water — a rival that has industrialised the very act of helping strangers build things. This essay is an attempt to follow that parcel all the way down.

Part I: The Pilgrims of Pazhou

Twice a year, in spring and autumn, a great migration takes place. Tens of thousands of Indian businessmen board flights to Guangzhou for the China Import and Export Fair — the Canton Fair — held in the vast Pazhou complex, 1.55 million square metres of exhibition space, 75,700 booths, 4.65 million products on display. The most recent spring edition drew a record 314,000 overseas buyers from 220 countries. India has long ranked among the top three buyer nations at the fair, behind only Hong Kong and the United States — an entire subcontinent's entrepreneurial ambition, queuing politely at the temple of somebody else's manufacturing.

So established is this pilgrimage that a full-service industry has grown up around it. Indian tour operators offer Canton Fair packages with visa handling, hotel blocks near Pazhou, and — my personal favourite — daily Indian dinners prepared by an accompanying Maharaj, in vegetarian, Jain, and non-vegetarian variants. One may thus source Chinese electronics for one's Indian brand without once compromising one's dietary purity. Civilisational continuity, it turns out, is negotiable on the factory floor but non-negotiable at the buffet.

And what are the pilgrims buying? Not machinery to make things with, by and large. They are buying finished innovation — products already designed, engineered, tooled, tested, and certified by someone else, awaiting only a logo. Shoes, smartwatches, earbuds, air fryers, massage guns, fairy lights, yoga mats: the entire visible universe of Indian e-commerce, pre-assembled in Guangdong. The buyer's contribution to the value chain is a sticker and a marketing budget.

I do not say this to sneer at traders. Trade is honourable, arbitrage is ancient, and the man who spots a margin between Yiwu and Yamuna Vihar is doing his job. The indictment is not of any individual middleman; it is of an economy in which middleman-ship has become the dominant expression of entrepreneurial energy — in which a generation's cleverest commercial minds are deployed not in making better products but in finding, freighting, and rebranding somebody else's.

The numbers are unsentimental about this. Surveys by Jungle Scout have found that over 70 per cent of Amazon sellers source their products from China; by 2024, sellers based in China had themselves crossed 50 per cent of the marketplace. Our own trade ledger tells the same story with fewer decorations: Indian imports of electronics, telecom instruments, circuit boards, and battery components from China surged 16 per cent to $131.63 billion in 2025–26, widening the bilateral deficit to $112.6 billion. That is not a trade relationship; that is a dependency with paperwork.

Consider the case study nobody can stop citing, because it is simply too perfect. boAt — India's most beloved audio brand, the toast of every festive sale — captured a 37 per cent share of the personal audio market by 2020, thrashing the Chinese brand realme, which limped in at 8 per cent. How did a plucky Indian upstart defeat the Chinese? By manufacturing in China, of course. For years, roughly nine in ten boAt products rolled out of Chinese factories; the brand's genius lay in design briefs, distribution, and marketing that spoke fluent Indian. To the company's genuine credit, it has since shifted assembly onshore through a joint venture with Dixon Technologies under the government's production-linked incentive scheme, and now advertises that three-quarters of its products are made in India. The assembly, that is. The components — the drivers, the chips, the cells, the little beating hearts of the things — still, overwhelmingly, speak Mandarin.

There is a genre of national self-congratulation that calls this “Make in India.” A more honest label, for much of it, would be Mark in India — the affixing of domestic insignia to foreign engineering. And before anyone objects that Apple does the same: yes, precisely, Apple does the same, except that Apple designed the phone. The design, the silicon, the operating system, the patents — the parts of the value chain where the margins and the power live — belong to Cupertino. Our champions, with honourable and growing exceptions, own the sticker and the ad jingle.

The Canton Fair, in other words, is not the disease. It is the X-ray.

Part II: Ctrl+C, Ctrl+V, Series A

If hardware is imported wholesale, software — our supposed national genius — has largely been imported conceptually. Recite the honour roll of Indian consumer tech and you are reading a translation exercise: Ola for Uber, Flipkart for Amazon, Oyo for every budget-hotel aggregator that came before it, Zomato and Swiggy for the global food-delivery template, Blinkit and Zepto for the quick-commerce playbook road-tested in Chinese cities years earlier.

My favourite specimen, for sheer poetry, is Paytm. It was described for a decade as “India's Alipay,” and the description was rather more literal than the describers intended: Alipay's parent, Ant Financial, bought a 25 per cent stake in Paytm's parent company back in 2015 — its first big cheque in India. The homage was so faithful that the original acquired a quarter of the tribute. (Ant finally exited in August 2025, offloading its last 5.84 per cent for about ₹3,980 crore, nudged out as much by New Delhi's allergy to Chinese cap tables as by any strategic epiphany.)

Now — and this is where the argument must be handled with care, because the lazy version of it is wrong — copying is not the sin. Everybody copies. China's own tech pantheon began life as a tribute act: Baidu was the Google clone, Alibaba the eBay-cum-Amazon clone, Tencent's QQ a pixel-faithful ICQ, Meituan a Groupon knock-off that later ate the entire category. Imitation is how a developing economy learns to ship at scale. It is a stage.

The only question that matters is whether you graduate from the stage. China graduated with extreme prejudice. The country of the Google clone now produces BYD, which outsells Tesla; CATL, which sets the global pace in batteries; DJI, which owns the world's drone market outright; Huawei, which holds more 5G patents than any firm alive despite half a decade under sanctions; and DeepSeek, of which more presently. The Australian Strategic Policy Institute's Critical Technology Tracker measured the arc with brutal economy: across 2003–2007, China led in 3 of 64 tracked frontier technologies; across 2019–2023, it led in 57 of 64, with the United States ahead in just 7. That is not catching up. That is lapping the field while the field debates whether running is culturally appropriate.

India, meanwhile, remains enrolled in the copying stage like a student who keeps deferring graduation for the campus food. And the reason is not a deficiency of neurons — it is the orientation of capital. Indian tech startups raised $9.1 billion in 2025, up a healthy 23 per cent; e-commerce and fintech topped the table, as they always do. Deep tech attracted roughly $2.3 billion of that — a 37 per cent jump, celebrated with much trumpetry — except that by the industry's own accounting, AI ventures constituted 84 per cent of those startups and 91 per cent of that funding, with average cheques that would embarrass a Shenzhen seed round. One Bengaluru founder confessed toThe Ken the prevailing anxiety about the new money: that it becomes “a pot of money that VCs use to create fake deep-tech versions of themselves.”

When the commerce minister, Piyush Goyal, stood up in April 2025 and needled Indian founders for doing dukaandari — shopkeeping; delivery apps and ice-cream brands — while Chinese firms built semiconductors and robotics, he was mobbed, and the mobbing was half deserved, because the founders' rejoinder was unanswerable: with whose money, minister? The capital for a seven-year silicon play barely exists here; the state's own research foundation (of which, again, more presently) took two years to spend its first rupee. A billion consumers reward the app that reaches them by Thursday far more reliably than the lab that files patents by 2031. Blaming founders for following the incentive gradient is like blaming water for flowing downhill and then holding a press conference about the moral fibre of water.

The tell, as ever, is what happens when someone does attempt the hill. In January 2025, a Chinese lab most people had never heard of released a reasoning model, trained on deliberately hobbled export-grade GPUs, that matched OpenAI's best and wiped roughly a trillion dollars off American tech valuations in a single trading session. DeepSeek's founder, Liang Wenfeng, had already explained his constraint the year before: money was never the problem — “Bans on shipments of advanced chips are the problem.” China, denied the best hardware, responded by making the best hardware less necessary.

And India's response to the DeepSeek moment? A debate. A truly magnificent debate — op-eds, panel discussions, parliamentary questions — about whether India could build such a thing, conducted at precisely the moment Washington's short-lived AI diffusion rule had filed us under Tier 2, with a cap of some 50,000 high-end GPUs through 2027. (The rule was rescinded within months, but the filing cabinet had spoken.) China spent the DeepSeek moment demonstrating escape velocity from hardware dependence. We spent it litigating our seat allocation on someone else's aircraft.

Which makes the exception worth naming with genuine respect. Sarvam AI, a Bengaluru startup handed 4,096 subsidised H100s under the IndiaAI Mission, shipped in early 2026 two foundation models — a 30-billion and a 105-billion-parameter mixture-of-experts pair — trained from scratch, on Indian compute, released open-weight under Apache 2.0, reportedly on a budget in the region of $50 million. One must be honest about the full arc: Sarvam's earlier “sovereign” model, released in mid-2025, was a fine-tune of Mistral — the flag was Indian, the base coat was French — and the wags were merciless. But the 105B is the real article, and it proves the point cuts both ways: the talent can do it, when someone actually points the money and the compute at the hill. The scandal is not that Sarvam exists. The scandal is how nearly it didn't.

Part III: The Original Sin — An Education Designed by Its Detractors

To understand why the system points the way it does, one must exhume the man who aimed it. In 1835, Thomas Babington Macaulay — a gentleman whose acquaintance with Indian learning was inversely proportional to his confidence in dismissing it — produced his infamous Minute on Indian Education, declaring that a single shelf of a good European library outweighed the entire literature of India and Arabia, and prescribing an education system to manufacture a class of persons “Indian in blood and colour, but English in taste, in opinions, in morals, and in intellect.” The purpose was never mass enlightenment. It was the production of clerks: a compliant intermediary caste to run the Company's paperwork at local wages, while knowledge “filtered down” to the masses in some perpetually postponed future. (A certain member for Thiruvananthapuram has prosecuted this case at book length, and I refer the reader to him for the full charge sheet.)

The colonisers eventually left. The blueprint, regrettably, stayed and got promoted. In 1945 the Sarkar Committee, asked how a soon-to-be-free India should build technical capacity, recommended not mass vocational training, not universal schooling, but four elite institutions modelled explicitly on MIT. Thus the IITs — the first opened in 1950 at Kharagpur, on the site of a colonial detention camp, a symbolism everyone admired and nobody examined. Independent India looked at Macaulay's inverted pyramid, and with the best intentions in the world, gilded the top of it.

The gilding continues. The Institutes of National Importance — IITs, IIMs, NITs — absorb roughly 18.2 per cent of the national higher-education budget while enrolling under 0.76 per cent of its students; the state spends about eighteen times more per INI student than per ordinary one. We built the penthouse before the plinth, and we are still ordering chandeliers.

Meanwhile, at the plinth: the ASER surveys, year after patient year, document a foundational catastrophe with the persistence of a metronome. In 2024, 44.8 per cent of Grade 5 children in government schools could read a Grade 2 text; 30.7 per cent could manage basic division. Among rural youth aged 14 to 18 — the demographic dividend itself, the ones on the posters — a quarter cannot fluently read a Grade 2 passage in their own language, and more than half stumble on a three-digit-by-one-digit division sum. These are not children failing school. This is a school system failing to occur.

Why does it fail to occur? Partly because it frequently does not show up. The famous J-PAL experiment in rural Udaipur found teacher absence rates around 44 per cent; when researchers issued tamper-proof cameras and tied salaries to photographic proof of attendance, absence promptly halved to 21 per cent. Professional duty, it transpires, responds admirably to surveillance and money — a finding that tells you everything about the trust architecture of the Indian state, in both directions. And even when present, teachers report spending vast tracts of the day on mid-day-meal ledgers, census duties, and WhatsApp demands from the block office, teaching for perhaps half the mandated hours. The bureaucracy audits the dal with more rigour than the division.

Above this hollowed base, we then built — nothing. The vocational middle, the technician class that carried China's manufacturing miracle on its shoulders, simply does not exist here at scale: roughly 2 per cent of Indian students at secondary level and above receive vocational education, against about 25 per cent in China. The results are darkly comic. L&T, the country's flagship engineering conglomerate, publicly bemoans a shortage of 45,000 workers and engineers; the same year, 11,000 candidates — among them PhDs, engineers, and MBAs — applied for 15 peon and clerk posts in Madhya Pradesh. A shortage of employable skills and a glut of unemployable degrees, coexisting in the same labour market, is not a paradox. It is a report card.

One table, because the divergence deserves to be seen in columns:

Both countries stood at exactly the same mark in 1987 — 62 per cent of workers in the fields. One built literacy and technicians and moved a third of a billion people up the complexity ladder into factories. The other built entrance examinations. The World Inequality Lab, which assembled these figures, describes the strategies as top-down versus bottom-up; I would describe them as building a ladder versus building a diving board.

Part IV: Research & Development, or the Absence Thereof

Let us now speak of R&D, that thing we hold conferences about.

The headline number, freshly tabled in Parliament this July: India's gross expenditure on research and development reached a record ₹2.45 lakh crore in FY24 — which sounds stirring until one notices it amounts to 0.84 per cent of GDP, up from 0.82 the year before. At this rate of ascent we shall reach our own stated target of 2 per cent by roughly the heat death of the universe, or 2035, whichever the ministry concedes first. China spends 2.69 per cent of a much larger GDP. South Korea spends nearly 5. The United States, about 3.5. Brazil — Brazil — manages 1.19. We are outspent, as a share of national income, by a country whose principal exports include telenovelas.

There was one genuinely structural bright spot in the data: for the first time, private industry (₹1.27 lakh crore) outspent government (₹1.18 lakh crore), crossing 51.8 per cent of the total. Progress — though in Korea, China, and America the private share sits above 70 per cent, and much of our public spend is defence and space line items rather than broad industrial research. Indian industry, the minister himself lamented while tabling the figures, must wean itself off imported technology. One pictures the assembled captains of industry nodding gravely before flying to Guangzhou.

What does the money we do spend produce? Volume, mostly. India now ranks third or fourth globally in sheer research output — over 300,000 papers a year, ahead of the UK and Germany — and ninth in citations, with a national H-index of 925 against America's 3,213. We have, in other words, perfected the academic equivalent of the white-label import: papers that look like research, formatted like research, indexed like research, and read by nobody. A respectable share of this is outright rot: study after study has identified India as the world capital of predatory publishing, home to over a quarter of the outfits running multiple pay-to-publish “journals” — the inevitable harvest of a promotion system that counts papers instead of reading them. The University Grants Commission created an approved-journals list to stem the tide, then found the fraudsters simply migrated, as fraudsters do, one loophole ahead of the circular.

Patents tell the same story in a different register. Indian resident filings have grown at double digits for six consecutive years — genuinely commendable — and yet the intensity gap remains a chasm: China files roughly 4,900 resident patent applications per $100 billion of GDP; India files 381. Historically, some 90 per cent of central research funding flowed to the elite institutes, while the state universities educating 95 per cent of students were left to run laboratories on prayer and Sellotape. PhD stipends arrive with the punctuality of the monsoon in a drought year; the nation's doctoral students have had to protest in the streets for the timely release of their own fellowships, an activity that consumes precisely the hours one might otherwise spend on, say, research.

Into this landscape the government has now launched its grand correctives, and here one must be scrupulously fair, because the design is genuinely good and the early execution genuinely alarming. The Anusandhan National Research Foundation (ANRF) was established by statute in 2023 as the apex funder of Indian science. A parliamentary committee reported this March that the foundation utilised zero of its ₹2,000-crore annual budget in 2023–24, zero again in 2024–25, and 61 per cent in 2025–26. A research foundation that took twenty-four months to disburse its first rupee has at least achieved something no Indian lab ever has: perfect capital preservation. Alongside it now sits the ₹1 lakh crore Research, Development and Innovation Fund — Cabinet-approved July 2025, launched by the Prime Minister that November, first cheques issued to second-level fund managers in May 2026, with the Technology Development Board and BIRAC receiving ₹2,000 crore apiece to on-lend as patient capital. By the standards of Indian officialdom this ten-month sprint constitutes indecent haste, and the officials deserve the compliment. The question — the only question — is whether the money reaches teams doing genuinely hard things at DARPA-ish speed, or whether it curdles into a fund-of-funds that launders consumer-app risk appetite through a deep-tech label. The pipes are laid. We await the water.

Part V: The Ledger of Departures

And now the cruellest column in the national accounts.

Begin with the study that should be laminated and nailed to the door of every education ministry office: economists Choudhury, Ganguli, and Gaulé tracked the top scorers of the IIT Joint Entrance Examination — the most ferocious academic filter on the planet — and found that 36 per cent of the top 1,000 had emigrated. Among the top 100, 62 per cent. Attending one of the original five IITs added a further five percentage points to the probability of departure, courtesy of alumni networks wired straight into American graduate schools and Silicon Valley. Read that plainly: we operate the world's most competitive examination in order to determine, with exquisite statistical precision, whom to gift to California. The state spends eighteen times the average per elite student and then waves from the tarmac. It is the most rigorously merit-based export programme in human history.

We have even learnt to celebrate it. Every time an Indian-origin executive ascends at Alphabet or Microsoft or IBM, the national press erupts as though the export of our finest minds were a trade surplus. Sundar Pichai and Satya Nadella are magnificent men and wretched metrics: their triumphs are Mountain View's operating income, not Chennai's.

Set this against the eastern ledger. Chinese and Indian STEM PhDs in America historically stayed at nearly identical, sky-high rates — both near nine in ten. But the Chinese line has begun to bend homeward while the Indian line stays flat around 80-plus per cent. The haigui — the “sea turtles,” returnees swimming home — have become a torrent: returning graduates have more than doubled since 2018, up another 12 per cent in 2025 alone. And it has reached the very peak of the distribution. CNN counted at least 85 established and rising scientists departing American institutions for Chinese ones since the start of 2024, more than half of them in 2025. The roster reads like a fantasy-league draft: Charles Lieber, former chair of Harvard chemistry, took up Tsinghua's highest faculty rank in Shenzhen in April 2025 — convicted in Boston, feted in Guangdong. Omar Yaghi joined Tsinghua in Beijing that July to build an AI-and-chemistry institute, then won the Nobel Prize in Chemistry in October. China recruited a chemist in the summer and collected his Nobel by the autumn; even our IPL auctions are not that efficient. Gérard Mourou, the 2018 physics laureate, went to Peking University. Berkeley's Yang Dan to Beijing; Berkeley's Sun Song, a Fields Medal contender, to Zhejiang; Harvard's Liu Jun to Tsinghua. Some of these scientists were pushed as much as pulled — America's China Initiative and funding chaos did Beijing's recruiting for it — but the destination tells you what mattered: there was a landing strip.

That is the entire asymmetry in one word. A Chinese researcher at Stanford contemplating home sees a funded state laboratory, a faculty package with equipment money, spousal hiring, and frontier employers of the DeepSeek, BYD, and Huawei class. An Indian researcher of equal calibre contemplating home sees an assistant professorship at a salary his Bay Area rent would laugh at, a grant that may arrive after the equipment's warranty expires, an import duty on his instruments, and a customs officer with a delicate cough. Talent does not flow toward flags. It flows toward the place where it canbuild. Coming home to China is a career move. Coming home to India is a sacrifice, and we have the audacity to be sentimental about it.

Part VI: The Bureaucratic Sublime

Threading through every section above, you will have noticed, runs a single connective tissue: the low-trust state. It deserves a brief chapter of its own, if only to admire the craftsmanship.

The Indian administrative apparatus was engineered by a colonial power to extract and to suspect, and independence changed the personnel without changing the posture. The licence raj was formally executed in 1991; its ghost promptly reincarnated as the compliance raj. The state assumes the citizen is a tax evader, the importer an under-invoicer, the researcher a procurement risk, the teacher a truant — and in fairness, having starved every one of them of trust for two centuries, it has manufactured no small number of each. So it responds with fortifications: BIS registrations, EPR certificates, import-export codes, valuation queries, utilisation certificates, and the discretionary power of ten thousand gatekeepers, each of whom has learnt that a gate is a toll booth waiting to be recognised. The customs “revaluation” — in which an official simply decides your goods are worth more than the invoice says, unless persuaded otherwise — is not a bug in this system. It is the system's compensation package.

The cost is not the bribe. The bribe is trivial; ask Karthik. The cost istime, and the risk premium on time, and the entire category of enterprise that never begins because its founder can price the friction in advance. A nation cannot run an AI buildout, a semiconductor mission, and a deep-tech fund through the same choke-points it uses to harass a man importing twenty air-quality sensors. The state's own flagship projects feel the drag: the Tata fab at Dholera, the ₹91,000-crore centrepiece of the Semiconductor Mission, was announced at the 28-nanometre node with first silicon promised for December 2026; the reported plan has since slipped to 90 nanometres — two generations coarser — with commercial output drifting toward 2028. Starting at mature nodes is perfectly defensible engineering. Slipping two nodes and eighteen months between the annual report and the press release is perfectly Indian scheduling.

We rent compute, meanwhile, with genuine competence — the IndiaAI Mission has stood up some 38,000 GPUs for researchers and startups, and that is honest, useful work. But rented compute is a treadmill, not a territory. The machines depreciate, the export rules mutate with each American administration, and the tier we are filed under is decided in a city where we do not vote.

Part VII: The Exceptions That Prosecute the Rule

Honesty now demands the counter-brief, because India has built world-class things, and pretending otherwise would be its own farrago of distortions.

UPI is the genuine article: roughly 228 billion transactions in 2025, over 21 billion in a single month, recognised by the IMF as the world's largest retail fast-payment system by volume. It is a masterpiece — of state-built public infrastructure. Standards, rails, protocol design: the things a competent bureaucracy can decree into existence. It is not a semiconductor, not a molecule, not a foundation model; it required no seven-year private bet on unproven physics. Celebrate it, absolutely — and notice which muscle it exercised.

ISRO landed Chandrayaan-3 near the lunar south pole for about ₹615 crore — less than the budget of a mediocre Hollywood film, a fact every Indian uncle can recite from memory. Glorious. Also: an island. ISRO is a mission-mode enclave of excellence floating in a state that underfunds nearly everything around it, and its celebrated frugality is, examined coldly, the optimisation you perfect when the topline never grows. We are the world champions of doing more with less because less is all we have ever appropriated.

The pharmaceutical industry vaccinated a respectable fraction of the human species — through what is, definitionally, process innovation on molecules whose patents have expired. Extraordinary manufacturing. Other people's invention.

And Sarvam, the exception already saluted, which broke pattern only after the state aimed 4,096 GPUs and a mission mandate directly at it. The pattern across all four is uncomfortably crisp: India excels wherever the game isexecution, standards, cost-engineering, and scale — and thins out precisely where the game is invention. The exceptions do not refute the thesis. They cross-examine it, establish its boundaries, and then, rather devastatingly, corroborate it.

Part VIII: What Is Actually To Be Done

Diagnosis without prescription is merely elegant complaining, and Delhi has enough of that. So, concretely — the dials a serious government (and a serious industry) would turn, in descending order of leverage:

1. Make the RDI Fund behave like DARPA, not like a department.

The ₹1 lakh crore is real; the May 2026 first cheques are real. Now publish the metric that matters: median time from application to money-in-bank, quarterly, for every second-level fund manager. Milestone-based tranches, programme managers with authority to kill and to double down, and an explicit quota for pre-revenue hardware, materials, and biology. If in three years the portfolio is 84 per cent AI wrappers, the experiment failed and someone should say so aloud.

2. A green channel for R&D atoms.

Duty-free import of prototypes and research equipment up to a sane annual cap per registered R&D entity; BIS exemption for non-retail development units; a 72-hour customs clearance service-level for anything flagged R&D, with the clearance times published. Karthik's twenty sensors should clear Nhava Sheva faster than his pizza order. This is a stroke-of-pen reform; it requires no money, only the surrender of discretion — which is, of course, why it is hard.

3. Drag private R&D from 52 to 70 per cent.

Restore a weighted tax deduction for in-house research (the old Section 35(2AB) logic, killed without replacement), add a patent-box rate on income from domestically developed IP, and tie PLI extensions to component-level value addition — drivers, cells, and boards, not just final assembly. boAt proved assembly follows incentives; now point the incentives one layer deeper into the bill of materials.

4. Build the landing strip.

A returnee package worth the name: five hundred laboratories at ₹5–10 crore in guaranteed start-up funding, five-year fellowships at globally embarrassing-to-refuse stipends, dual academic-industry appointments, spousal placement, and — critically — procurement autonomy, so the returning scientist buys her mass spectrometer without a GeM tender odyssey. China did not moralise its diaspora home; it out-bid the alternative. Sentiment is not a compensation structure.

5. Fix the plinth, finally.

The NEP's target of vocational exposure for half of all learners must be executed as infrastructure, not circulars: funded skill labs, industry-linked ITIs with the 400 credential mills stripped out and the survivors capitalised, and Teaching-at-the-Right-Level pedagogy scaled with the same seriousness we bring to entrance-exam coaching. The camera experiment taught us the ugly, useful truth: accountability works. Deploy it and stop flinching.

6. Kill the paper mill.

Promotions and PhDs assessed on a small number of works, read by actual humans, with citation impact weighted over counts — and criminal-adjacent consequences for predatory publishing rings. China ran precisely this crackdown; the sky did not fall; the citations rose.

7. Measure time itself.

Publish, every quarter, the state's latency dashboard: median grant disbursement time, median customs clearance for R&D goods, median BIS certification, median stipend delay. What gets measured gets minimised. What stays hidden stays for sale.

None of this is exotic. Every item has a working precedent somewhere between Seoul, Shenzhen, and Arlington, Virginia. The binding constraint is not knowledge and has never been money. It is the willingness of a low-trust state to extend trust first — to its scientists, its founders, its importers, its teachers — and to accept the fraud losses of trust as a cost of doing civilisation, rather than accepting the stagnation losses of suspicion as the price of control. China, an authoritarian surveillance state, contrives to trust its builders more than the world's largest democracy trusts hers. Sit with that sentence for a moment. I have been sitting with it for months.

Epilogue: The Parcel, Revisited

Karthik's monitor works beautifully, by the way. The firmware is written in Bengaluru. The enclosure was designed in Bengaluru. The boards are assembled by a contract manufacturer in Shenzhen, because after the third customs adventure he ran the arithmetic and the arithmetic won; his next production batch will ship to customers in Dubai and Singapore first, because their borders behave like doors rather than tollbooths. He remains, in every way that matters, an Indian founder. His supply chain has simply emigrated ahead of him — the way the JEE toppers did, the way the citations did, the way the components at Pazhou always already had.

The tragedy of Indian innovation, I have come to think, is not that we cannot invent. The IITians running half of Silicon Valley, the Nobel-calibre scientists, the from-scratch foundation model built on a shoestring — the raw material is embarrassingly abundant. The tragedy is that we have constructed, with two centuries of diligence and at colossal public expense, a machine for ensuring we needn't: an education that filters instead of building, capital that distributes instead of inventing, a border that taxes curiosity, and a state that trusts no one and is trusted by no one in return. The machine is not broken. It is functioning precisely as designed. It was simply designed, in 1835, by a man who wanted clerks.

He got us. The question of the next twenty-five years is whether we have the nerve to want something else.

Sources & Further Reading

History & education architecture

• T.B. Macaulay, Minute on Indian Education (2 February 1835).
• Report of the Sarkar Committee on Higher Technical Institutions (1946); Institutes of Technology Act (1961).
• S. Tharoor, An Era of Darkness: The British Empire in India (Aleph, 2016).
• N.K. Bharti & L. Yang, The Making of China and India in the 21st Century — World Inequality Lab Working Paper 2024/24; PDF.
• ASER Centre / Pratham — ASER 2024 national findings; see also asercentre.org and Pratham ASER page.
• E. Duflo, R. Hanna & S.P. Ryan, “Incentives Work: Getting Teachers to Come to School”, American Economic Review 102(4), 2012.
• Working-paper analyses of Institutes of National Importance budget concentration (~18.2% of higher-education spend; under 0.76% of enrolment); L&T labour-shortage coverage and Madhya Pradesh peon-post episode (contemporary press).

R&D, academia & policy

• DST / parliamentary replies (July 2026): GERD ₹2.45 lakh crore; 0.84% of GDP; private share 51.8% — covered in contemporary Rajya Sabha / Lok Sabha reporting.
• OECD Main Science & Technology Indicators (China, US, Korea, Brazil R&D/GDP comparators).
• Predatory publishing — arXiv:2003.08283 (Scopus country-level analysis).
• WIPO, World Intellectual Property Indicators (resident filings; patents per $100bn GDP).
• Parliamentary Standing Committee on Science & Technology reporting on ANRF utilisation (March 2026); Careers360 / Business Standard coverage of RDI Fund disbursements (May 2026).

Startups, capital & the copying question

• NASSCOM–Zinnov, India Tech Startup Landscape 2025 ($9.1bn funding; deep-tech ~$2.3bn).
• Piyush Goyal “dukaandari” remarks — Business Standard (Startup Mahakumbh, April 2025).
• Ant Group exit from Paytm — Reuters; Economic Times (August 2025).
• ASPI, Critical Technology Tracker (3/64 leads in 2003–07 → 57/64 in 2019–23).
• DeepSeek-R1 / January 2025 market reaction — Reuters / FT contemporary coverage; Liang Wenfeng interview (36Kr / “Waves,” 2024).
• Sarvam AI — Sarvam-30B / Sarvam-105B release coverage (Feb–Mar 2026); IndiaAI Mission GPU allocation (4,096 H100s).

Hardware, trade & customs

• Canton Fair 139th session — PR Newswire (314,000 overseas buyers; 1.55m m²; 75,700 booths).
• Jungle Scout / Marketplace Pulse / Statista — Amazon seller-sourcing surveys (>70% from China; Chinese sellers >50% of marketplace, 2024).
• boAt China-manufacturing / Dixon PLI shift — Quartz/Scroll and TechRadar contemporary coverage.
• Transparency International — Corruption Perceptions Index 2025 (India 39/100, rank 91); India country page; Global Corruption Barometer — Asia (2020 bribery rates).
• MeitY / BIS Compulsory Registration Scheme and EPR import-compliance documentation.
• Tata Electronics / India Semiconductor Mission — Dholera fab announcements vs subsequent node/schedule coverage (Varindia and contemporaries).
• US AI-diffusion export rule (Jan 2025; India Tier 2) and May 2025 rescission — contemporary Reuters / White House coverage.

Talent flows

• P. Choudhury, I. Ganguli & P. Gaulé — “Top Talent, Elite Colleges, and Migration” (NBER WP 31308); Journal of Development Economics.
• CSET (Georgetown), Trends in U.S. Intention-to-Stay Rates of International PhD Graduates (NSF Survey of Doctorate Recipients).
• CNN investigation of scientists departing US institutions for China (2024–25); Chinese Ministry of Education returnee (haigui) statistics.
• Chemistry World / Boston Globe / Nature — Charles Lieber to Tsinghua SIGS (Apr 2025); Omar Yaghi to Tsinghua (Jul 2025; Nobel Chemistry Oct 2025); Mourou, Yang Dan, Sun Song, Liu Jun appointments.

The exceptions

• NPCI / IMF — UPI 2025 transaction volumes; IMF recognition as world's largest retail fast-payment system by volume.
• ISRO / contemporary coverage — Chandrayaan-3 mission cost (~₹615 crore); see Chandrayaan-3 overview.
• IndiaAI Mission compute updates (~38,000 GPUs empanelled/deployed).

Alternate titles considered and affectionately shelved: “Mark in India” · “A Farrago of Middlemen” · “The Clerk Factory” · “Which Way the River Runs.”

On this site

• The Arranged Marriage of the Century — Part 2: why Japan and India are each other's missing half.
• The Rope Sellers Buy a Rope Machine — Indian IT as the middleman pyramid meeting the machine that automates pyramids.
• The Rope Sellers — accountability moats, body shops, and who still gets paid to be wrong.