Invention and Instantiation
Invention is the first reduction to practice of a technical principle that had not previously existed in a working artifact. Instantiation, which is the word I will use throughout for what is loosely called manufacturing, is the reproduction of an already solved design at a specified unit cost, yield and cadence. In ordinary Maldivian conversation these two are treated as points on a single scale of technical competence, with China somewhere ahead of the United States on that scale and moving further ahead every year, and the claim I make in this essay is that they are not points on one scale at all. They are two different businesses with opposite payoff structures, opposite tolerances for failure, and different institutional requirements at every level from bankruptcy law to the physical layout of a supplier park. A country that admires the output of one arrangement while legislating the conditions of the other will end with neither. I would try first to establish the distinction on economic grounds, then test it against the record in semiconductors, aviation, energy and medicine, then give the Chinese arrangement its strongest possible case and answer it, then admit the American failure that my argument has to survive, and only after all of that state what the Maldives can and cannot import from either.
The economics of invention are convex and the economics of instantiation are concave, and almost everything else follows from that single asymmetry. I would set the two payoff structures out plainly before applying them, since the remainder of this essay is an application of them and nothing more.
Let us call set X the set of all attempts at invention, and let each element of X have a payoff. The distribution of those payoffs is not normal. Most elements pay zero or less than zero, a small number pay back the capital, and a very small number pay several hundred times the capital, and the expectation of the set is carried almost entirely by that last group. Correlation Ventures examined roughly 21,000 venture financings between 2004 and 2013 and found that about 65% percent of them returned less than the money put in. An investor facing that distribution does not attempt to raise the average quality of his attempts by controlling them more tightly, because the loss he can suffer on any one attempt is bounded at his capital while the gain is not bounded at all. He attempts instead to increase the number of attempts and to shorten the time it takes to discover that an attempt has failed. Failure, in this business, is not waste. It is the price of the option, and the option is the product.
Now let us call set P the set of all production runs of a design already known to work. Here the distribution is inverted. The upside is bounded, because the best possible outcome is that every unit meets specification at the design cost, and there is nothing better than that. The downside is not bounded, because a process excursion can destroy a year of margin, a recall can destroy a company, and in the industry I am most familiar with, a single escaped defect can destroy an aeroplane. The rational operator of set P therefore spends his effort on the suppression of variance. He standardises, he documents, he audits, he removes discretion from the line, and he treats an unexplained deviation as a defect even when the unit passed. That is to say that the disciplines which make a man excellent at instantiation are precisely the disciplines which would make him useless at invention, and the reverse holds with equal force. The engineer who cannot tolerate an unexplained result will never invent anything, and the engineer who is comfortable with unexplained results should not be allowed near a production line.
It is essential for my argument to pause here and insist that this is a statement about institutions and incentives rather than about peoples. Nothing in what follows depends on any claim about American creativity or Chinese diligence as national characteristics, and I regard such claims as lazy. The same Chinese engineer who cannot get a speculative project funded in Hefei invents freely in Sunnyvale, and the same American firm that invented the technology loses the ability to build it within 20 years of moving the work offshore. What differs is the arrangement each engineer is standing inside, and arrangements can be described, priced, and in part copied. That is the whole practical value of the analysis.
The American advantage in invention rests on a small number of mechanisms which are unglamorous, mostly legal, and almost never mentioned when Maldivians discuss why the Americans keep inventing things. Each of them can be stated plainly, each of them has a date and a statutory reference, and none of them requires any appeal to national character.
The first is the price of failure. Chapter 11 and Chapter 7 of the United States Bankruptcy Code permit a firm to fail, discharge its obligations, and permit its founders to raise money again within a few years. In jurisdictions where insolvency carries criminal exposure, or where it ends a man's standing permanently, the rational entrepreneur selects safe projects. He is behaving correctly given his incentives, and the society that punishes him for failing has quietly purchased a portfolio of low-variance projects and given up the tail. It should be understood that this is not generosity toward failed businessmen. It is the state buying option value at a discount.
The second is the enforceability of employment restraints, which sounds like a technicality and is in fact the reason Silicon Valley exists in California rather than in Massachusetts. Section 16600 of the California Business and Professions Code voids covenants not to compete. Massachusetts enforced them. AnnaLee Saxenian's Regional Advantage (Harvard University Press, 1994) traced the divergence of Route 128 and Santa Clara County to exactly this, that engineers in California could leave and take their knowledge with them, and the firms that resulted were denser, more numerous, and collectively far more inventive than the vertically integrated Massachusetts minicomputer companies that legally retained their people and went extinct with them. Eight engineers walked out of Shockley Semiconductor in 1957 and founded Fairchild, and out of Fairchild came Intel, Advanced Micro Devices (AMD), National Semiconductor and a long list of others. In a jurisdiction with enforceable non-competes those eight men would have stayed, or gone into another line of work, and the transistor would have diffused through the economy a decade more slowly.
The third is compulsory diffusion of foundational patents. The 1956 consent decree that settled the United States antitrust action against AT&T confined the company to common carriage and required it to license its existing patents to domestic applicants at nominal royalty. The transistor patents were in that portfolio. A monopolist was thereby forced to hand the most valuable invention of the century to everyone who asked for it, and the American semiconductor industry is in a real sense a creature of an antitrust settlement rather than of the free market it is usually credited to.
The fourth is public money spent on results that no market would have paid for in advance. The National Institutes of Health disburse on the order of 47 billion dollars a year and the National Science Foundation roughly 9 billion, and the Bayh-Dole Act of 1980 (Public Law 96-517) allowed the universities receiving that money to own and license the patents arising from it, which converted a large federal research expenditure into a pipeline of firms. The Defense Advanced Research Projects Agency (DARPA) operates on a different principle again, in which a program manager holds real authority for a fixed term and is expected to terminate his own programs when they fail. DARPA funding produced packet switching and thereby the internet, and it also placed an award on the order of 25 million dollars with a small company called Moderna in 2013 to work on messenger RNA therapeutics, 7 years before anybody needed a coronavirus vaccine in a hurry.
The fifth is immigration, and here the numbers are not ambiguous. Theodore von Kármán, who founded the Guggenheim Aeronautical Laboratory and effectively American supersonic aerodynamics, was Hungarian. Igor Sikorsky, who gave the world the practical single-rotor helicopter, was born in Kyiv. Andrew Grove of Intel was Hungarian, Jensen Huang of Nvidia was born in Taipei, Sergey Brin was born in Moscow, and Katalin Karikó, whose work on nucleoside-modified messenger RNA made the vaccines possible, was Hungarian and was demoted at the University of Pennsylvania for pursuing it. The National Foundation for American Policy has repeatedly found that a majority of American startups valued above 1 billion dollars have at least one immigrant founder. A country that imports the ambitious surplus of every other country's education system, and then lets them keep their own equity, has arranged a permanent subsidy to itself that no industrial policy could replicate.
I have now established the mechanisms. The record is where they have to be tested. Semiconductors are the cleanest case because the chain of invention is documented to the day. The point-contact transistor was demonstrated at Bell Telephone Laboratories on 16 December 1947 by Bardeen and Brattain, working under Shockley. Jack Kilby at Texas Instruments built the first working integrated circuit in September 1958. Jean Hoerni at Fairchild produced the planar process in 1959, and Robert Noyce built the planar integrated circuit on top of it, which is the object every chip on earth is descended from. Mohamed Atalla and Dawon Kahng demonstrated the metal-oxide-semiconductor field-effect transistor at Bell Labs in 1959, and that device, not the bipolar transistor, is what the entire digital world is actually made of.
Then it becomes important to look at what happened to the manufacturing of those inventions, because the honest review of this does not flatter the American arrangement. Extreme ultraviolet lithography, the technology that makes present-day logic possible, was developed under a consortium formed in 1997 in which Intel, Motorola and AMD funded work at Lawrence Livermore, Lawrence Berkeley and Sandia national laboratories. The physics and the early engineering were American and publicly funded. The machine is Dutch. ASML now sells every EUV scanner in existence, at something over 150 million dollars each for the standard systems and considerably more for high numerical aperture, and the leading-edge logic those machines produce is patterned in Taiwan and South Korea. The United States invented the process and captured the design layer and the tool software, and lost the tool and the fabrication.
Aviation gives the same lesson with more nuance, and since it is my own field I will be exact about it. The Wright brothers' contribution was not the aeroplane in the loose sense. Their contribution was three-axis control, protected in United States Patent 821,393 of 22 May 1906, arrived at through a wind tunnel they built themselves in 1901 after they concluded that the published Lilienthal lift coefficients were wrong, and coupled with the first correct treatment of the propeller as a rotating wing rather than as a marine screw. That is invention in the strict sense, the reduction to practice of a principle nobody had instantiated. What followed is the part most people never hear. The Wrights spent the next decade litigating, the patent war suppressed American aircraft development so effectively that the United States entered the First World War with no combat aircraft of its own design worth flying, American squadrons flew French SPADs and Nieuports, and the deadlock was only broken by the government-brokered cross-licensing agreement of 1917. Invention without an institutional arrangement for diffusion produces a museum piece and a lawsuit.
The jet engine was not American at all. Frank Whittle filed his patent in Britain in January 1930, and Hans von Ohain's engine flew in the Heinkel He 178 on 27 August 1939. The swept wing was German, presented by Adolf Busemann at the Volta Congress in Rome in 1935 and collected by American investigators at the end of the war, after which it appeared in the B-47 and the F-86. What the United States did with all of this was industrialise it, which is to say that General Electric turned Whittle's design into the I-A, and later the American manufacturers took the high-bypass turbofan from concept to the TF39 for the C-5 in 1968 and the JT9D that made the 747 economically possible in 1969. Digital fly-by-wire was first flown by NASA on a modified F-8 on 25 May 1972 using a spare Apollo guidance computer, and it was Airbus, in Toulouse, that turned it into a certified civil product with envelope protection on the A320 in 1988. Anyone who wants to argue that invention and commercial capture are the same thing has to explain these two events.
Space is where the American mechanism shows most clearly, because the mechanism and the failure are both visible. SpaceX failed three consecutive Falcon 1 launches and succeeded on the fourth, on 28 September 2008, with the company nearly out of money. A first stage was landed and recovered on 21 December 2015 and a recovered stage was reflown on 30 March 2017. No amount of process discipline would have produced that result, because the intermediate states of the program were indistinguishable from incompetence, and only a financing arrangement that priced failure as the cost of an option could have carried it. The Global Positioning System (GPS) is the opposite case and equally instructive, a system with no commercial customer at all, funded by the United States Department of Defense from the first Block I satellite in 1978 to full constellation in 1995, and then given away, and the entire logistics, aviation, agriculture and telecommunications industries were rebuilt on top of it.
Medicine and biology is the next one. Katalin Karikó and Drew Weissman published the nucleoside modification result in Immunity in 2005, and were largely ignored for 15 years. Jennifer Doudna and Emmanuelle Charpentier published the CRISPR-Cas9 mechanism in Science in 2012 and Feng Zhang demonstrated it in mammalian cells in early 2013. Charpentier is French, Karikó is Hungarian, and the firm that turned her work into a product first was German. The American system did not have a monopoly on the minds. It had the arrangement that let those minds find money, and that difference is the whole of the matter.
I will now give the Chinese arrangement the same treatment, because the usual Maldivian account of it is a fairy tale in either direction, and both the awe and the contempt are in my view uninformed. Four mechanisms carry almost the whole of the Chinese advantage, and all four are as prosaic as the American ones.
The first mechanism is the number of trained production engineers. China produces first university degrees in engineering on the order of one and a half million a year against roughly two hundred thousand in the United States including computer science. Process engineering is labour of a particular kind, unglamorous, cumulative, and requiring a great many competent people rather than a few exceptional ones, and a country with seven times the supply of such people will win that contest under any political arrangement whatever.
The second is agglomeration, and this is the one that engineers underrate and economists overrate for the wrong reasons. In the Pearl River Delta a hardware designer can obtain an injection-moulded prototype, a modified flexible printed circuit and a revised motor winding within days of asking, because the vendors are within an hour's drive and hold the tooling. In a dispersed supply chain the same three changes take weeks. Design quality is not a function of the intelligence applied to a problem so much as of the number of build-test cycles completed before the design freezes, and cycle time is therefore the binding constraint on hardware development. Shenzhen did not out-think anybody. It compressed the iteration loop by roughly an order of magnitude, and an order of magnitude in cycle time is worth more than any amount of cleverness applied once.
The third is the cost of capital for chosen sectors. State-directed lending, provincial land grants and local government financing vehicles have driven the effective cost of capital in targeted industries close to zero and in some cases below it. This is normally described as unfair, and whether it is unfair is beside the point I am making. Capital priced at zero changes which projects clear the hurdle rate, and capital-intensive, long-payback, thin-margin instantiation is exactly the class of project that a normally priced capital market rejects and a subsidised one accepts. The Americans did not lose solar manufacturing because their engineers were worse. They lost it because a 15 percent hurdle rate says no to a factory whose return is 4 percent, and correctly so, from the point of view of the private investor.
The fourth is the learning curve itself, which is a real and quantifiable engineering phenomenon rather than a slogan. Wright's law holds that unit cost falls by a constant percentage for every doubling of cumulative production, and for photovoltaic modules that percentage has been on the order of 20% per doubling for four decades. Cumulative volume is the input to that function. A political system able to direct enormous volume into a sector is therefore able to buy movement down the learning curve directly, and a country that cannot generate volume cannot buy it at any price.
The silicon solar cell was invented at Bell Laboratories in 1954 by Chapin, Fuller and Pearson, at about 6% efficiency. The passivated emitter and rear cell architecture that dominates the modern market came out of Martin Green's group at the University of New South Wales in the 1980s. Neither of those facts is Chinese. What is Chinese is that module prices fell from roughly 4 USD per watt in 2008 to something in the range of 10 to 15 cents per watt by 2024, and that the International Energy Agency's 2022 assessment of solar supply chains found China holding more than 80% of global manufacturing capacity at every stage and above 95% for wafers. The invention was American, the decisive cell refinement was Australian, and the instantiation that made solar the cheapest source of electricity in most of the world was Chinese. Anyone who wants to say the Chinese only copy has to explain away the 80%, and anyone who wants to say the Chinese did it all has to explain away Chapin, Fuller, Pearson and Green.
Batteries repeat the pattern exactly. Lithium iron phosphate as a cathode material came from John Goodenough's group at the University of Texas at Austin, published by Padhi and colleagues in 1997, and Goodenough took the Nobel Prize for it in 2019 at the age of 97. Contemporary Amperex Technology and BYD made it manufacturable, and BYD's blade cell of 2020 is a genuine packaging innovation which raised volumetric efficiency by removing the module level entirely. Pack prices fell from roughly 1200 USD per kilowatt-hour in 2010 to about 150 USD in 2024 on BloombergNEF's survey. BYD passed Tesla in quarterly battery-electric deliveries in the fourth quarter of 2023, 526,409 against 484,507. These are real achievements and I decline to belittle them, because belittling them would be as unserious as being awed by them.
High-speed rail is the purest instantiation case in the record. China ran a technology transfer tender in 2004 and bought from Kawasaki, Siemens, Alstom and Bombardier, produced the CRH series under licence, absorbed the technology, and then built more than 45,000 route-kilometres of high-speed line, which is more than the rest of the world combined by a wide margin. The Fuxing trainsets that followed are substantially indigenous. No new physics was involved at any point. What was involved was the ability to sustain a construction program of that magnitude for 2 decades, and no democratic state with land acquisition constraints and 4 or 5 year electoral cycles can do it, which is a genuine advantage of the arrangement and must be recorded as one.
Drones are the case where China invented a product rather than a technology. DJI was founded in 2006 and took roughly 70% of the world consumer market. Every component in a Phantom existed already, the microelectromechanical inertial sensors, the brushless motors, the lithium polymer cells, the GPS receiver and the video links, and none of them were Chinese in origin. What DJI did was combine them into an object nobody had built and integrate the flight control law well enough that an untrained person could fly it. Schumpeter called that a new combination and treated it as innovation proper, and he was right to. It should be noted that this is the strongest evidence against my own thesis.
Aviation is where the Chinese arrangement meets its limit. The COMAC C919 first flew on 5 May 2017, was delivered to China Eastern in December 2022 and entered service on 28 May 2023. It is a competent conventional narrowbody. Its engines are CFM International LEAP-1C units, which is to say American and French, its fly-by-wire system is Honeywell, and a substantial part of its avionics, actuation and environmental systems are Western. The indigenous CJ-1000A engine remains in development. Deliveries have run in the low 10s per year against an Airbus A320-family production rate of roughly 50 to 60 aircraft per month. The reason is not Chinese incompetence. The reason is that the mechanism which makes Chinese manufacturing overwhelming, cumulative volume driving a learning curve, is unavailable in a product where global demand is perhaps 2,000 units a year across all manufacturers, where certification is adversarial and evidence-based, and where the physics of a high-bypass turbofan's hot section is a materials and process problem that cannot be shortened by building more of the wrong thing. Single-crystal turbine blade casting with directional solidification is knowledge that lives in a few 100 people and a small number of foundries, and it took Rolls-Royce, General Electric and Pratt & Whitney 50 years each to acquire it. The learning curve is not a magic wand. It is a function of cumulative volume, and where the volume is small the function does very little.
One might rightly say that all of this is antiquarian. The Americans invented the transistor, the solar cell and the lithium cathode, and so what, since the Chinese now make all 3 cheaper, faster and in some cases better, and the money and the industrial power went to the maker. Value accrues to whoever holds the physical capacity, and a patent certificate hanging in a Delaware office keeps nobody's lights on. That is the argument, and it is the argument most educated people actually hold, so it deserves a full answer.
The answer has three parts. The first is that value does not in fact accrue to the assembler in the general case. Stan Shih of Acer described this in 1992 as the smile curve, in which margin sits at the two ends of the chain, in design and brand, and collapses in the middle where assembly sits. Kraemer, Linden and Dedrick measured it on the iPhone and found Apple capturing well over half of the retail price as gross profit while Chinese labour captured something under 2%. The Chinese solar industry dominates global module manufacturing and has spent much of the last 2 years selling below cost, with the largest producers posting losses, because a commoditised product with subsidised entrants converges to a price that recovers nothing above marginal cost. Dominating the manufacture of a commodity is not the same thing as earning from it, and a policy that confuses market share with margin has confused an activity with a return.
The second part concerns the chokepoint, and here the events of October 2022 settled the argument empirically. When the United States imposed export controls on advanced logic equipment, the binding constraint turned out to be a handful of firms holding positions no volume of manufacturing could substitute for, ASML in lithography and Cadence, Synopsys and Siemens in electronic design automation. An entire national manufacturing base of extraordinary competence was gated by 4 or 5 companies whose product is knowledge. That is what invention buys, and it does not buy it every day, but the option pays exactly when it matters. The Netherlands, a country of 18 million people, holds more leverage over the semiconductor industry than any manufacturing base on earth, and it holds it because of one firm's position in one physically difficult machine. A small country should read that fact very carefully.
The third part is the one that matters most for policy. Instantiation capability is rentable and invention capability is not. Any state with capital can hire a Chinese contractor to build its bridge, and its port, and the Maldives has done so. No state can hire the ability to have thought of something first. From this then it follows that a small country's scarce resources should be spent on the capability that cannot be purchased in the market, and the capability that can be purchased should in fact be purchased, on the best terms available, from whichever supplier offers them. The correct posture toward Chinese construction capacity is that of a well-informed buyer, and there is nothing shameful in being a buyer, provided one knows what one is buying.
I would be writing propaganda if I stopped there, so the American failure has to be stated plainly. Manufacturing knowledge is largely tacit. It lives in the hands of process engineers and in the accumulated corrections of a particular line, and it cannot be fully written down, which means it cannot be fully transferred and cannot be recovered once lost. Gary Pisano and Willy Shih called the resulting shared base of skills and suppliers the industrial commons in the Harvard Business Review in 2009 and argued that the United States had let it erode to the point where whole categories of invention were no longer available to it, because you cannot invent a manufacturing process for an industry you no longer have. Intel's failure at the ten-nanometre node while Taiwan Semiconductor Manufacturing Company went past it is the clearest demonstration. Intel had the money, the physicists and the patents, and it did not have the yield learning.
Aviation supplies the more painful evidence. Boeing designed the 737 MAX, and the manoeuvring characteristics augmentation system (MCAS), in an organisation that had subordinated engineering judgment to schedule and to a certification strategy aimed at avoiding simulator training. Lion Air 610 was lost on 29 October 2018 with 189 people aboard and Ethiopian 302 on 10 March 2019 with 157. The door plug that departed Alaska Airlines 1282 on 5 January 2024 left the factory without the 4 bolts that retain it, and the paperwork for the removal that preceded the omission did not exist. That is not a design failure. That is the failure of exactly the variance suppression discipline I described earlier as the core competence of instantiation, inside the country that invented the aeroplane. The distinction between invention and instantiation is not a claim that Americans are good at one and therefore safe. It is a claim that they are different competences, and the second one can be lost by a nation that has the first.
Therefore the honest verdict on the two arrangements is that each is efficient at one half of the technical economy and each is now visibly paying for having neglected the other half. China is spending enormous sums to buy the invention half, through its research funding, its returning-scholar programs and its semiconductor funds, and it has begun to succeed in places, in quantum communications with the Micius satellite in 2016, in sodium-ion cells, in electric drivetrains, and recently in the efficiency frontier of large language models. The United States is spending enormous sums to buy back the instantiation half through subsidised fabrication capacity. Both are attempting to import a mechanism they let go of, and both are discovering that it is slow and expensive.
The error I am attacking is not admiration of Chinese manufacturing. Admiration is appropriate and I have expressed some of it above. The error is a specific logical move, which is the inference from an observed outcome to a policy prescription without possession of the mechanism that produced the outcome. Richard Feynman gave his 1974 Caltech commencement address the title Cargo Cult Science, and described islanders who built runways out of packed earth, lit fires along the sides, put a man in a hut with two wooden pieces on his head for headphones and bamboo for antennas, and waited for the aeroplanes to land. Everything was correct in form. Nothing landed, because the arrangement that had made aeroplanes land was not the runway.
Let us make it concrete and local. Let us suppose Sharaah, a ship builder in R. Innamaadhoo, who visits Guangdong, sees a five-axis machining centre producing propeller blanks at a rate he cannot match, and returns convinced that the machine is the reason. He borrows from BML against his house, imports the identical machine, installs it, and discovers within the year that he has bought the artifact and not the mechanism. He has no tool-and-die shop within 4000 kilometres that can regrind his cutters in two days, no second machine to run when the first is down, no volume of work to justify a full-time programmer, no local supplier of the alloy at the tolerance the machine assumes, and no 300 other boat yards nearby competing with him and thereby teaching him. The Guangdong workshop's productivity was never a property of the machine. It was a property of everything within an hour's drive of the machine. Sharaah has bought a photograph of a factory.
That is exactly what a state does when it observes Chinese industrial results and demands that its politicians reproduce them. The demand takes a predictable form in our public discussion. Somebody observes that the Chinese built the Sinamalé Bridge, which they did, at a project cost of roughly 220 million USD, financed with a Chinese Ministry of Commerce grant, a China Eximbank concessional loan, and about 12 million USD of our own budget, opened on 30 August 2018 and handed over that November. Somebody observes that the same period produced a 3,400 metre runway at VIA built by Beijing Urban Construction Group. From these observations the conclusion is drawn that we should adopt the Chinese way of doing things, and the person drawing it usually cannot state what the Chinese way of doing things is beyond the fact that things get built.
Consider what is actually being inferred. The visible first-stage effect is a bridge, which is real, useful, and photographs well. The second-stage effects arrive later. The concessional loan entered its repayment phase, the Eximbank obligation was among those restructured under the G20 Debt Service Suspension Initiative in July 2020, external debt reached roughly 4.4 billion dollars by the end of 2024 against a GDP of well under 7 billion, and on 2 April 2026 the state settled the 2021 sukuk at 524.68 million dollars, comprising the 500 million principal and 24.68 million in accumulated profit at a rate of about 9 to 10 percent, paid out of the Sovereign Development Fund and foreign reserves. Loan repayments in the first four and a half months of 2026 came to 8.67 billion MVR, against 5.15 billion for the whole of 2025. Whether that was the right borrowing to have done is a separate argument on which reasonable people differ, and I am not making it here. The point is narrower and harder to escape. The generation that admired the artifact is not the generation that pays for it, and a public debate conducted entirely in the currency of visible artifacts will systematically overbuy them, because the man who cuts the ribbon and the man who services the debt are never the same man and frequently not even the same party.
Now the second and more damaging half of the error. Even where the money is sound, the model cannot transfer. The 2022 census counted about 500,000 people resident in the Maldives of whom about 380,000 were Maldivian citizens. The mechanism that makes Chinese manufacturing formidable is cumulative volume acting on a learning curve inside a dense supplier network. We have no volume, we have no supplier network, and no policy can create either, because both are functions of population and land area and we have neither in the quantities required. To pursue an industrial strategy modelled on the Chinese one is to purchase every cost of that model, the subsidised capital, the protected state enterprises, the tolerance for loss-making champions, while being structurally incapable of receiving the benefit that justifies those costs.
The American model transfers no better, and I want to be even-handed about this, because I have heard people that admires Silicon Valley with exactly the same lack of curiosity about mechanism. A venture portfolio strategy requires that 65% of attempts fail and be absorbed. In a country of about 380,000 people, the cohort of people capable of running a serious technical venture in any given decade is perhaps a few 100, and a 100 failures does not represent a healthy portfolio, it represents the destruction of the entire cohort's balance sheets and, in a society this small, their standing. Convexity requires a large number of independent trials, and a small population cannot generate them. The American arrangement is not superior in the abstract. It is superior at a scale we do not have.
Having spent this long on demolition I owe a positive position, and it rests on distinguishing between importing a result and importing a mechanism. Results are artifacts, bridges, factories, industrial parks, national champions, and they can be bought and are usually bought at the wrong price for the wrong reason. Mechanisms are arrangements of incentive and information, and they are cheap, unphotogenic, and almost entirely absent from our policy discussion.
The first is the price of failure inside the state. Both the American and the Chinese arrangements, in different ways, permit a project to be declared dead and the information from its death to be used. In our administrative culture a failed initiative ends a career, and the observable consequence is that no initiative is ever declared failed. Projects are extended, rebranded and quietly abandoned without an evaluation, which means the state pays the full cost of the failure and receives none of the knowledge, which is the worst available combination. The reform is unglamorous and costs almost nothing, being a requirement that every project above some threshold have a written prior statement of what would count as failure, a scheduled evaluation against it, and an explicit rule that an honest negative evaluation carries no career penalty while a concealed one does. Nothing in that requires Chinese capital or American venture funds.
The second is absorptive capacity, in the sense Cohen and Levinthal gave the term in Administrative Science Quarterly in 1990, that a firm's ability to use external knowledge is a function of the related knowledge it already holds. A turnkey contract delivers an asset and transfers no capability, because there is nobody on our side of the table who could have built the thing and therefore nobody who learns anything by watching. The corrective is a standing rule that no major technical acquisition proceeds without a named domestic counterpart team with the competence to specify, audit and accept it, and with the contractual right to be present during design and construction. This costs a fraction of 1% of a project and it is the only line item that produces a person instead of a structure.
The third is regulatory competence as a national asset. Adopting European rules wholesale, which we largely do, transfers a text and not an inspectorate, and the text is worthless without people who can interpret it against a case nobody anticipated. A small state cannot invent an airframe and it can absolutely hold a first-rate oversight authority, because oversight scales with the number of qualified inspectors rather than with population or capital, and the ICAO's effective implementation score under the Universal Safety Oversight Audit Programme is a published, comparable, purchasable-by-effort number. There are few investments available to us with a better ratio of international standing to cost.
The fourth is the deliberate selection of domains where small scale is not disqualifying. The learning curve punishes us wherever volume decides, and it is silent wherever knowledge, regulation, environment or accumulated operational data decide. We operate the largest seaplane fleet in the world, on the order of 60 aircraft at TMA alone, in an environment with island geography, water aerodrome operations, daylight-limited seaplane rules and a passenger mix that exists nowhere else on this scale. Nobody on earth holds more operational data on that class of operation than Maldivian operators, and almost none of it has been converted into published standards, training products, certification services or safety research that could be sold to every developing water-aerodrome operator from Indonesia to Greece. The same holds for reef and lagoon engineering, for island logistics and for the operating systems of high-end island hospitality. These are domains in which our smallness is irrelevant and our specificity is the asset.
The fifth is the honest treatment of the human capital equation. Scholarships without absorptive employers are a foreign aid program run by the Maldives for the benefit of richer countries. The binding constraint on return migration is not patriotism, it is whether there exists an organisation at home in which a trained person's competence is used rather than resented, and that is a question about the internal governance of our state enterprises rather than about education policy. It should be understood that this trade-off has a real cost on the other side, since raising the technical standard of appointments necessarily reduces the number of appointments available for political distribution, and there is no version of this reform in which that does not happen.
In conclusion, I have argued that invention and instantiation are distinct economic activities with opposite payoff structures, that the American arrangement is optimised for the convex one through the low price it puts on failure and the legal and fiscal machinery that supports it, that the Chinese arrangement is optimised for the concave one through scale, supplier density, subsidised capital and the learning curve, that neither is a national virtue and both are institutional arrangements which can be described and to a limited degree copied, and that the Maldivian habit of inferring policy from foreign outputs is a category error. The country does not have to choose between being awed by Beijing and being awed by Washington. It has to stop being awed, which is a harder discipline than either, and start asking of every foreign success the only question that has ever produced a transferable answer, which is not whether it worked but by what mechanism it worked and whether we possess the conditions that mechanism requires. Where we do, we should copy it exactly. Where we do not, we should buy the output from whoever sells it cheapest and spend our scarce capability on the small number of things in which 400,00 people, sitting on one of the most operationally difficult nations in the world, are already the global authority and have never once thought to say so.