Isaac Asimov's Foundation & Why China/Singapore Want More Manufacturing Jobs
Manufacturing creates 27 service jobs per 100 workers. That matters
In Isaac Asimov’s Foundation series, the psychohistorian Hari Seldon develops mathematical models proving the Galactic Empire is collapsing. His equations detect patterns invisible to individual observers: peripheral regions losing technical knowledge, productive capacity concentrating in the capital Trantor then atrophying, social cohesion dissolving despite aggregate prosperity.
The Empire’s decay follows predictable stages. Trantor, the administrative capital, grows to house over forty billion inhabitants: bureaucrats, financiers, and administrators consuming resources but producing nothing. As Asimov describes it: “All the land surface of Trantor, 75,000,000 square miles in extent, was a single city. The population, at its height, was well in excess of forty billions. This enormous population was devoted almost entirely to the administrative necessities of Empire.”
Meanwhile, peripheral regions that once sustained the Empire lose technical knowledge systematically. Salvor Hardin, mayor of the Foundation’s colony on Terminus, describes the pattern: “We’re receding and forgetting, don’t you see? Here in the Periphery they’ve lost nuclear power. In Gamma Andromeda, a power plant has undergone meltdown because of poor repairs, and the Chancellor of the Empire complains that nuclear technicians are scarce. And the solution? To train new ones? Never! Instead they’re to restrict nuclear power.” The decay spreads: first they can’t maintain their systems, then they forget the principles, finally they treat remaining equipment as religious artifacts.
Trantor’s leadership dismisses Seldon’s similar warnings and put him on trial. The statistics look fine. Economic activity concentrates in the capital. Markets boom. Administrative employment soars. Why worry about distant manufacturing regions when the core economy thrives?
Seldon sees what aggregate numbers miss. At his trial for treason, he states plainly: “The fall of Trantor cannot be stopped by any conceivable effort. It can be hastened easily, however.” He explains to the imperial court: “The fall of Empire, gentlemen, is a massive thing, however, and not easily fought. It is dictated by a rising bureaucracy, a receding initiative, a freezing of caste, a damming of curiosity—a hundred other factors.”
Civilizations that stop producing and start only consuming, administering, and financializing don’t transition smoothly. They collapse. Not uniformly, but from the periphery inward. By the time the center notices, reconstituting lost productive capacity requires centuries.
Paul Krugman credits Foundation with inspiring his economics career. Economists love Seldon’s insight that mathematical models can predict civilizational patterns. What some miss is what doomed Trantor: the belief that administration and finance can replace production.
The Contrarian Bets
In 2018, Singapore (a city-state that successfully transitioned to services, with 73% of GDP from finance, trade, and logistics) announced Manufacturing 2030, a strategy targeting 50% growth in manufacturing value-add by 2030. Not preserving heritage industries. Aggressively expanding advanced manufacturing.
Meanwhile, China, having achieved the service transition that development economists prescribed, reverses course. After a decade experimenting with financialization and property speculation, Xi Jinping declares manufacturing the foundation of future development. “The real economy should not be abandoned, nor should the traditional industries within it.”
Two of the world’s most analytically sophisticated governments betting against academic consensus.
Development theory offers a clear path: agriculture/mining → manufacturing → services. Every economics textbook explains this progression. The advanced economies that abandoned manufacturing in the 1980s-2000s believed they were following natural economic evolution. Reagan-era policymakers shifted focus to services during recession. Clinton administration economists called the 1990s “the Fabulous Decade” as services boomed and manufacturing declined.
Singapore and China didn’t reject this logic casually. They independently discovered something about the service transition that conventional metrics didn’t detect. One learned through systematic research, one through crisis experience. Their analysis identified patterns in demographic data, employment multipliers, and competitive foundations that aggregate productivity statistics missed entirely.
The academic establishment pushes back hard. Leading economists at Peterson Institute argue that even eliminating the entire U.S. manufacturing trade deficit would increase manufacturing’s employment share by only 1.7 percentage points. Bruegel’s analysis concludes “the manufacturing jobs boom isn’t happening” and warns policymakers are “setting themselves up for failure.” IMF research argues that services can drive productivity growth just as well as manufacturing, and the decline in manufacturing jobs has contributed little to rising inequality in advanced economies.
The consensus: Services drive productivity growth. Trying to restore manufacturing jobs is nostalgic futility.
But what if the conventional metrics are measuring the wrong things?
What They Discovered
The Multiplier Asymmetry
Singapore didn’t set out to challenge development theory. As a small city-state facing survival constraints, it obsessively measures every economic relationship because mistakes show up immediately in a 280-square-mile economy. In 2018, the Ministry of Trade and Industry commissioned research to understand sectoral spillovers.
The numbers revealed an asymmetry that surprised the researchers. Manufacturing creates $0.29 in additional services for every $1 million in value-added. Services create only $0.02 in manufacturing. A 14.5x asymmetry. Employment followed the same pattern: 100 manufacturing jobs create 27 service jobs. 100 service jobs create only 3 manufacturing jobs.
Singapore’s policymakers checked whether this was a local quirk. U.S. data validated a similar pattern. Manufacturing multipliers range from $1.92 to $2.74 for every dollar of output, the highest of any economic sector. Retail and wholesale trade generate only $0.54 and $0.58 respectively.
Manufacturing doesn’t just create manufacturing jobs. It creates demand for precision logistics, specialized financing, engineering support, IP protection, and corporate R&D. These aren’t generic service jobs. They’re high-skill, high-wage positions requiring specialized expertise. Services mostly create demand for more services. Marketing agencies need graphic designers. Consultants need other consultants.
That 14.5x reveals what conventional metrics tend not to focus on: At high development levels, advanced manufacturing and sophisticated services are complements, not substitutes. Without manufacturing, services become generic rather than specialized.
Israel’s experience validates this finding. Between 1975 and 2005, Israel’s Office of the Chief Scientist enforced strict requirements that state-funded R&D be manufactured domestically. Products developed with public money had to be produced in Israel. Intellectual property couldn’t be transferred abroad without repayment of 3-6 times the original grant.
Between 1995 and 2011, high-tech employment more than doubled from 98,000 to 215,000 workers. Israeli economists quantified the multiplier effect: each employee in high-tech manufacturing created two additional jobs in non-manufacturing industries. Each R&D center employee created only one-third of an additional job.
When Given Imaging pioneered endoscopy capsule technology with $5 million in state R&D funding, the IP retention requirements forced a choice: take state money and build here, or forfeit the investment. The required technology didn’t exist domestically. Given Imaging built it anyway, establishing a production facility in Yokneam Illit that eventually employed nearly 1,000 workers.
High-tech exports grew from 14% of manufactured exports in the 1980s to 54% by the early 2000s. The multiplier worked exactly as Singapore’s research predicted: manufacturing created sophisticated services, not the reverse.
The Demographic Signals
China discovered the problem through crisis. After decades of manufacturing-led growth that lifted 800 million people from poverty, China faced pressure to rebalance toward services and domestic consumption. The 2008 global financial crisis provided the catalyst: massive stimulus, doubled bank lending, infrastructure investment boom.
GDP growth remained steady at 7%. The service sector expanded to 50% of economy. By conventional metrics, the transition was working. But total debt doubled to nearly 3x GDP. Nearly 40% of bank loans went to property speculation. Domestic consumption remained stuck at 37% of GDP versus 60-70% in developed economies.
When the property bubble deflated, the diagnosis was confirmed. Evergrande defaulted in 2021 with over $300 billion in debt. The “service boom” had generated speculative froth, not sustainable prosperity.
Chinese policymakers paid attention to signals that conventional economic metrics weren’t designed to capture. Similar patterns were emerging across advanced economies, but going largely unnoticed.
Around 2000, U.S. life expectancy stopped improving for the first time since 1918. Between 1992 and 2017, deaths from suicide, drug overdoses, and alcohol abuse (what economists Anne Case and Angus Deaton call “deaths of despair”) more than tripled among middle-aged Americans without bachelor’s degrees. By 2021, over 176,000 Americans died from despair-related causes, making it the fifth leading cause of death in the U.S. Between 2019 and 2021, American men lost 1.8 years of life expectancy; women lost 1.2 years.
Case and Deaton identified the likely cause: “The collapse of the steady, decently paid manufacturing jobs that once gave meaning and purpose to working-class life.”
The pattern appears across advanced economies. Finnish men in stable service positions like police work saw 9% fertility decline between 2010-2019. Men in precarious ICT contract work saw 40% decline. Same education level, same sector, wildly different family formation patterns. In the Netherlands, administrative records show the highest-earning women now have 60% higher birth rates than the poorest, a reversal of historical patterns. When researchers measure income before childbearing decisions, higher income consistently predicts more children.
Homeownership rates for Americans under 35 dropped to 36.3% in late 2024, the lowest since 2019. For the first time on record, graduate unemployment now exceeds general unemployment. Young college graduates aged 23-27 experienced 4.59% unemployment in 2025, compared to 3.25% in 2019.
People make long-term decisions about children, education, and home-buying based on employment stability and economic security. Traditional metrics measured whether people had jobs. These indicators reveal whether those jobs support major life decisions. Countries with precarious employment show declining fertility, household formation, and human capital investment, regardless of GDP growth.
These are the signals that conventional measurement systems weren’t designed to detect.
The Competitive Foundations
Both Singapore and China independently converged on structural requirements for manufacturing competitiveness. Energy costs matter. Not as one input among many. As foundation.
Matt Parlmer (
) analyzed industrial electricity pricing data and found Southern China’s Pearl River Delta, the world’s electronics manufacturing hub, pays roughly $0.09/kWh for industrial electricity. California companies pay ~$0.24/kWh. Massachusetts pays ~$0.18/kWh. When electrons on one side of an ocean cost three times what electrons on the other side cost, you don’t need sophisticated economic modeling to predict where capacity gets built.Europe’s experience confirmed this logic. Between 2019 and 2023, UK industrial electricity prices grew 124%, Hungary 171%, Poland 137%, and France 93%. During the same period, U.S. industrial prices grew only 21%. When Europe’s industrial electricity costs doubled relative to the U.S., manufacturing weakened systematically.
The competitive gap is widening rapidly. Between 2017 and 2023, China quintupled its industrial robots per manufacturing worker while simultaneously growing its human manufacturing workforce. In 2023, China added more industrial robots than the rest of the world combined, installing roughly as many robots per year as the entire U.S. stock. On a per capita basis, China now deploys 470 industrial robots per 10,000 manufacturing workers, surpassing the United States’ 295.
But automation’s impact depends entirely on how it’s implemented. W. Edward Deming’s 11th point of quality management warned: “Eliminate work standards (quotas) on the factory floor. Substitute leadership. Eliminate management by objective. Eliminate management by numbers, numerical goals.”
The American approach to automation followed exactly what Deming warned against. Rather than designing systems for long-term capability building, management pursued quarterly cost reduction targets. Automation became a tool for hitting numerical goals: reduce headcount by X%, cut labor costs by Y%. When firms automated to hit cost targets rather than build capabilities, they got cost reduction without productivity gains.
China automated while maintaining manufacturing employment as a system goal. In 2017, China had 97 industrial robots per 10,000 manufacturing workers compared to 200 in the U.S. By 2023, China quintupled its robot density while growing its human manufacturing workforce. The U.S. robot growth over the same period, for its shrinking human workforce, was less than 0.5x.
Same technology. Opposite outcome. The difference isn’t the robots. It’s whether management views workers as essential to the system or variables to optimize.
As Parlmer observes: “The main reason people offshore to China nowadays is access to a dramatically more sophisticated manufacturing sector than anywhere else in the world.” Not labor costs. Manufacturing sophistication.
Why the Consensus Was Wrong
If this analysis is correct, why did most advanced economies miss it? The forces that made abandoning manufacturing appear rational were powerful and self-reinforcing.
Short-term metrics looked good. GDP growth continued. Unemployment stayed low. The stock market tripled during the 1980s. Clinton’s 1990s averaged 4% annual growth with balanced federal budgets by decade’s end. Services created jobs. Why question success?
Management by objectives dominated corporate decision-making. Quarterly earnings targets, numerical performance metrics, and cost-reduction goals shaped strategic choices. Manufacturing required patient capital and long-term capability building. Services offered immediate returns measured in financial metrics. When you manage by quarterly objectives rather than system performance, abandoning manufacturing for services makes perfect sense.
Financial sector profits were enormous. Finance, insurance, and real estate generated huge returns with less capital investment than manufacturing. Reagan’s deregulation of financial markets continued under Clinton, making financial services increasingly lucrative.
The political economy favored change. Manufacturing unions were powerful and expensive. The service sector offered flexibility. Reagan breaking the air traffic controllers’ strike signaled broader anti-union shift that continued across administrations.
Israel’s experience shows how financial interests reshape systems. When the country’s venture capital sector gained influence in the 2000s, they successfully lobbied to weaken the IP retention requirements that had driven manufacturing scale-ups. The VCs wanted state R&D subsidies without conditions that would slow exits or reduce returns. The employment multipliers that had generated 2:1 spillovers during the IP retention era largely evaporated.
When Senator Marco Rubio raised similar questions during U.S. Innovation and Competition Act debates (what prevents Chinese acquisition of IP developed with U.S. funding?) the bill passed without proposed restrictions. Senator Bernie Sanders proposed amendments requiring recipient firms to give government partial equity stakes and prohibiting stock buybacks and outsourcing. The amendments failed.
The pattern repeats across political systems: Financial interests favor public risk-taking and private reward-taking. Manufacturing interests favor requirements that ensure domestic spillovers. When financial interests dominate policy, systems optimize for exits rather than employment.
Development theory provided intellectual justification. Every economics textbook endorsed the service transition as natural progression. The academic establishment, with few exceptions, dismissed concerns about manufacturing decline as protectionist nostalgia.
The hidden costs weren’t visible in real time because conventional metrics weren’t designed to detect them. Employment quality deterioration showed up in demographics but not GDP. Multiplier effects aren’t measured in national accounts. Energy cost foundations seemed like minor inputs. Automation benefits depended on implementation approaches nobody was tracking systematically.
But why could Singapore and China see what others missed?
Singapore’s small size forced obsessive empiricism. When you’re 280 square miles, you can’t hide from consequences as much as a country large as the United States. The Ministry research that discovered multiplier asymmetries came from institutional necessity to optimize limited resources. Mistakes show up immediately.
China’s crisis experience made the service transition failure visceral. Property bubble collapse affected millions directly. Leadership with long time horizons could see demographic warning signs early. When social stability is threatened, you pay attention in ways that countries coasting on legacy advantages don’t.
The Counterarguments
The mainstream economic response to this analysis is sophisticated and deserves engagement. Economists at Peterson Institute and Bruegel aren’t naive. Their calculations are technically correct within their framework. Eliminating the U.S. manufacturing trade deficit would marginally increase manufacturing’s employment share. Industrial subsidies haven’t delivered promised job growth in the timeframes politicians claim.
The dispute isn’t about arithmetic. It’s about what economic systems are meant to optimize. IMF economists are correct that services productivity growth matches manufacturing in aggregate statistics. Finance sector productivity soared throughout the 2000s. GDP per capita rose steadily even as manufacturing employment declined.
The question is whether aggregate productivity captures what matters for social stability.
Finance sector productivity soared while deaths of despair tripled. GDP grew while homeownership for under-35s collapsed. Fertility plummeted among precarious workers while remaining stable among secure ones. Graduate unemployment exceeded general unemployment for the first time on record.
The mainstream response would be: these problems reflect other failures (housing policy, healthcare costs, education debt, social safety net inadequacy). Fix those, and service-dominated economies work fine. Switzerland, Luxembourg, and Singapore itself thrive with majority service economies.
This argument has merit. But it assumes employment structure and social policy are separable. The evidence suggests they interact. Services can grow GDP while undermining social stability. Gig work generates economic activity while making family formation decisions nearly impossible. The question isn’t whether services create economic activity. It’s whether that activity generates the employment stability that supports long-term decisions about children, homes, and education.
The demographic patterns suggest that at current implementation levels in most advanced economies, service-dominated structures correlate with social instability regardless of aggregate growth. This could reflect policy choices rather than inherent limitations, but those policy choices have proven remarkably difficult to change across different political systems.
The Reconstitution Challenge
Understanding this analysis doesn’t mean copying it works. Local institutional context determines which policies actually succeed.
Two Mexican states pursuing identical industries over 40 years illustrate the problem. Jalisco and Querétaro both developed electronics, ICT, automotive, and aerospace industries from the 1980s through 2010s. Same global opportunities. Same federal support. Radically different results.
Jalisco’s business-guided approach let private sector design policies: tax exemptions, deregulation, land grants. The state hosted over 30 corporate R&D centers and diversified from 3 electronics products to over 30. But less than half the ICT firms survived past 2015. Researchers documented “scarce spillovers” to local firms. Explosive proliferation without sustainability.
Querétaro’s state-guided approach created lasting institutions: specialized training programs, quality infrastructure networks. The state achieved world-class efficiency in automotive and aerospace. OEMs transferred entire production lines from Japan and Canada citing superior production efficiency. But product diversification remained limited.
These patterns persisted through multiple political transitions. Early wins locked in approaches that later became constraints. Countries can’t simply copy policies from others. Local institutional styles, built over decades, determine which policies actually work.
The U.S. lost global semiconductor manufacturing share from 37% in mid-1990s to 12% by 2020. The pattern mirrors Asimov’s Empire exactly. Peripheral regions first lose the ability to maintain their systems. Then they forget the underlying principles. Finally, they treat remaining technology as something mysterious, importing what they can no longer make.
The U.S. semiconductor industry followed similar stages. First, manufacturing moved offshore while design stayed domestic. America would innovate, Asia would produce. Then advanced packaging and testing migrated because they required proximity to manufacturing. Now cutting-edge process development happens where the fabs are, because you can’t separate R&D from production at the technological frontier.
The knowledge atrophied because the institutional systems that created and transmitted it (the fabs, the supplier networks, the engineering communities, the daily problem-solving on production floors) dissolved. You can’t rebuild capacity by funding research alone.
Rebuilding requires reconstituting the entire system: manufacturing facilities, supplier ecosystems, technical communities, institutional knowledge embedded in thousands of daily decisions. Israel discovered this when IP retention requirements forced Given Imaging to build domestic manufacturing capability that didn’t exist. China is building it through patient, systematic investment at scale.
What Policy Actually Requires
Current U.S. policy reveals the difficulty. The CHIPS and Science Act allocated $52.7 billion to rebuild semiconductor manufacturing: $30.6 billion in grant awards to 19 companies across multiple states. This represented the first serious industrial policy in decades.
Then President Trump called the program “a horrible, horrible thing” and urged Congress to eliminate it. The Commerce Department laid off 40 workers from the CHIPS program. Manufacturing employment has fallen 42,000 jobs since his tariff announcement in April 2025.
Tariffs without industrial policy don’t rebuild what decades hollowed out, assuming that’s the goal. They raise costs for manufacturers that remain, since over half of U.S. imports are business inputs: industrial supplies, capital goods, materials. Tax Foundation analysis documents how tariffs on manufacturing inputs hurt American manufacturers who depend on global supply chains.
The U.S. needs energy infrastructure that makes industrial electricity competitive with China, not three times more expensive. It needs patient capital that abandons quarterly objectives for decade-long capability building. It needs functional labor relations that treat workers as system components rather than costs to minimize. It needs political institutions that can sustain multi-administration commitments.
Brazil’s agricultural research corporation Embrapa provides evidence that successful industrial policy can work against conventional wisdom. Between 1973 and 2010, Embrapa increased Brazilian agricultural productivity by 110%, generating $17 in benefits for every dollar spent. The key: geographic dispersion, not concentration. Embrapa established 40+ research centers across Brazil’s six biomes, with researchers in the Amazon studying Amazon problems.
Had Embrapa concentrated all resources in Brasília headquarters, productivity gains would have been just 70% instead of 110%. The benefit-cost ratio would drop from 17 to 11. This matters because manufacturing, like agriculture, increasingly requires solutions adapted to local conditions: energy infrastructure, workforce capabilities, supplier networks, regulatory environments.
For the United States, this suggests regional strategies rather than national mandates. Texas’s energy costs, Washington’s hydropower, remaining technical expertise in specific sectors provide foundations. But even regional strategies require institutional transformations that take years: patient capital, functional labor relations, infrastructure investment measured in decades.
China’s Made in China 2025 robotics program (recent analysis) reveals both achievements and constraints. The program mobilized 12 pilot cities and 604 companies, generating nearly 400,000 patents. China’s domestic industrial robot market share rose from marginal to 52.45% by 2023.
Yet upstream component producers showed no statistically significant improvement in core technologies. Precision reducers, controllers, and sensors remain beyond reach despite seven years of targeted support. Core component development requires 5-10 year cycles versus 1-2 years for applications, materials science breakthroughs rather than incremental improvements, and precision manufacturing capabilities developed over generations.
Chinese policymakers understand these limits. The question for the United States is whether it can match even this level of sustained commitment.
The Experiment’s Results
The evidence is clear. Singapore’s 14.5x asymmetry. Israel’s 2:1 employment spillovers under IP retention requirements. Over 176,000 annual deaths of despair. Collapsing homeownership among the young. Plummeting fertility concentrated among precarious workers. China quintupling robot density while growing human manufacturing employment as the U.S. does neither.
Singapore’s manufacturing sector grew 4.3% in 2024, with S$31 billion committed through 2025. China could control half of all industrial robots globally by late 2025. American industrial electricity prices in population centers remain three times China’s rates, despite superior natural advantages.
The United States has spent four decades testing whether advanced economies can thrive on services alone, whether employment quality doesn’t matter as long as GDP grows, whether you can break the link between manufacturing and prosperity without social consequences. The results are written in life expectancy tables, birth rates, and homeownership statistics.
In Foundation and Empire, the sequel to Asimov’s original work, the Foundation faces what appears to be an overwhelming external threat: a mutant warlord called the Mule with powers psychohistory couldn’t predict. Everyone focuses on this external shock that broke Seldon’s Plan (despite the previous Seldon Crises). But the real threat isn’t revealed until later. The First Foundation had maintained surface-level technological superiority (advanced weapons, superior equipment, higher productivity) while losing the deeper institutional knowledge that made innovation possible. They operated sophisticated technology without understanding underlying principles. When tested by external pressure, this hidden weakness became catastrophic.
China isn’t America’s Mule, the unpredictable external threat. China is revealing what four decades of policy choices already broke. The U.S. maintains technological leadership through globally dispersed supply chains it no longer controls. Advanced chips designed in California, manufactured in Taiwan using Dutch lithography equipment and Japanese materials. Like the First Foundation’s weapons built from imported components they couldn’t manufacture themselves.
The institutional transformations necessary to rebuild can’t happen overnight. They require sustained commitment across multiple administrations, patient capital that abandons quarterly thinking, functional labor relations, and energy infrastructure investment measured in decades. The question isn’t whether transformation is difficult. The question is whether the alternative is sustainable.
As Case and Deaton wrote: “Destroy work and, in the end, working-class life cannot survive.”
Seldon’s warning at his trial echoes across the centuries: “The fall of Trantor cannot be stopped by any conceivable effort. It can be hastened easily, however.” The fall he predicted came from “a rising bureaucracy, a receding initiative, a freezing of caste, a damming of curiosity.” From the center focusing on administration and finance while the periphery lost the knowledge to produce.
The United States still has technical advantages, remaining manufacturing expertise, natural resource wealth that could support competitive energy costs. The institutional capacity to exploit these advantages is dissolving in real time. How long before recognition comes? In Asimov’s Foundation, it came too late. Reconstituting what had been lost required centuries.