Li Xiang & Lu Ming on AI, Involution, Housing Prices and Consumption

Summary

• Lu Ming argues that AI will accelerate the concentration of people in large cities, and that long-term demand for cities and housing will continue to depend on where jobs are going. As machines replace standardized, process-driven, codifiable manufacturing jobs, workers will move toward services that depend more on population density, scale and foot traffic. Innovation also requires frequent, cross-disciplinary, serendipitous interaction, drawing companies further into city centers. “The AI era won’t reduce agglomeration; it will accelerate it” (“人工智能时代不是减少集聚，它是加速集聚的”).
• High housing prices need to be separated into the byproduct of prosperity and the cost of constrained supply; falling prices also cut both ways. When rising incomes and population growth push prices higher, that has very different policy implications from displacement caused by restricted land supply and a housing shortage. Lu believes that, for households with sufficient income, a 20%-30% decline in home prices—alongside lower rents for homes, shops and offices—could make the city attractive again to residents and businesses. The trade-off is near-term pressure on land-sale revenues, and this does not mean he believes the market has bottomed.
• Consumption potential depends not only on income, but also on the availability of services and whether people have time to consume them. In Lu’s view, the decline in retail sales in Beijing and Shanghai from January to August also reflects the pull-forward of trade-in purchases, where online sales are booked, and the undercounting of services; it cannot be equated directly with a broad contraction in consumption. He stresses that average weekly working hours have reached 48.6, while service consumption takes time. “More holidays, limits on working hours and less overtime” are also conditions for boosting domestic demand.
• Local governments competing for investment does not automatically maximize national welfare; economies of scale require a unified market and the right fiscal incentives. Lu assesses county-level competition through “long-term, holistic and multidimensional development.” Strategies suited to an early stage of limited capital and small industrial scale can turn into duplicative construction, subsidy races and local debt in the era of cars, large aircraft and AI. He sees broader consumption taxes, collection later in the supply chain and a larger local-government share as ways to shift the focus from production to attracting people and consumption, while stressing that reform cannot happen overnight.
• Price wars in industries such as autos squeeze margins, delay payments and drive excessive overtime throughout the supply chain. Lu notes that China now produces more than 30 million cars a year, with some of the competition effectively taking place between Chinese companies. If subsidies keep loss-making firms alive while costs are pushed onto suppliers, export growth may not benefit companies and workers alike. “Domestic consumers’ welfare is being sacrificed to subsidize foreign consumers” is his sharp summary of this production model. Solutions include exit mechanisms, labor protections, and enforcement of environmental and intellectual-property rules.
• The key to industrial policy is how and under what conditions it is implemented; one successful local investment cannot simply become a national template. Lu supports government coordination of early-stage investment, land, finance and talent, but opposes picking technology winners too early and having every locality build the same industries regardless of its advantages. In discussing Hefei, he says its Yangtze River Delta location, USTC talent base, BOE’s growth stage and the timing of investment all need to be part of the explanation. Local funds’ “reinvestment in the locality” requirements, he argues, can also fragment the market.
• AI is changing not only research tools, but also what is scarce, what people need to be capable of, how companies are organized and what social protection must cover. Lu groups memory, rote learning, calculation and derivation as “machine functions,” and argues that universities should devote more space to general education, practice and understanding human nature and demand. On AI provision, he believes the US is “most likely” to follow a model in which companies invest and charge for access, while China is moving toward broader access; that model still has to address wasted compute and subsidies accelerating job displacement.
• Where to live should depend on a person’s career, preferences, social ties and stage of life; the trend toward agglomeration does not produce one answer for everyone. Lu compares city-center housing to “a front-row concert ticket,” reminding residents to understand what they pay for convenience and a richer urban life. For people planning to buy in their hometown and return when their health deteriorates, he cautions that medical care, caregivers and familiar social networks may make the city where they have lived for years more valuable in old age. Attachment to one’s hometown and property investment should be considered separately.

Deep dive

1. “Escaping” and “returning” are both happening, but big cities still draw people with opportunity and peers

• Li Xiang opens by asking whether people have largely stopped talking about “escaping Beijing, Shanghai and Guangzhou.” Lu Ming disagrees that the discussion has disappeared. Slower growth, pressure on services, fewer job openings and barriers to public services for migrants can still make life in big cities feel too difficult.
• Lu says those who return home face more than an income problem: smaller cities offer fewer jobs and a less varied life, and it can be harder to find colleagues, collaborators or like-minded friends. “People may be living in different worlds, to the point that it’s hard even to have a conversation,” so some start wanting to “escape back to Beijing, Shanghai and Guangzhou.”
• At the city level, population concentrates through constant turnover: some stay, some leave and others move in. In boom times, rising home prices bring “urban problems” into sharper focus; under pressure from slower growth, Lu is more concerned with whether core cities and surrounding metropolitan areas can create jobs again and revive the economy.

2. Whether high housing prices are an outcome or a cause determines what they mean for the economy

• Lu first treats housing prices as an outcome: where the economy is growing well, incomes are higher and more people live there, housing demand is naturally stronger. Higher prices in high-income regions do not, by themselves, mean development has gone wrong.
• A separate dynamic comes from quotas for construction land. Lu explains that even when a city wants to convert agricultural land to residential or industrial use, it is constrained by quotas. Past efforts to control the population of large cities, which treated congestion as equivalent to urban sprawl, reinforced those restrictions.
• Restricting land has not stopped people from moving in, but insufficient housing supply has driven up costs. At that point, prices stop being a byproduct of prosperity and become a barrier to entry: “It raises the cost of living and pushes people out.”

3. Falling housing prices hurt land-sale revenues but may also lower the bar for entering a big city

• Asked by Li about whether housing-price controls after 2011 kept prices in check—or even drove them down—in Beijing and Shanghai, Lu calls the effect “double-edged.” Local-government revenue is closely tied to land sales, so falling prices weigh on land-sale revenues. But for households with sufficient income, a 20%-30% decline may mean they can now afford what was previously out of reach.
• Lower rents for homes, shops and offices may also make residents and businesses feel they can afford to stay in Shanghai again. Lu says the trend of lower costs becoming an attraction “may well be re-emerging” over the past six months or so. He adds a caveat: opportunities in a new cycle would come if the economy bottoms and rebounds.

4. The decline in Beijing and Shanghai retail sales needs to be unpacked by base effects, booking location and consumption mix

• Li links housing-price movements to the year-on-year decline in retail sales in Beijing and Shanghai from January to August. Lu acknowledges that demand for appliances and renovations is tied to housing, but does not see it as the main explanation. The first things to consider are last year’s trade-in base and the “intertemporal substitution effect”: if people bought a car early, a new subsidy will not make them buy another one this year.
• The platform economy has changed where sales are recorded. Lu gives the example of a Shanghai resident buying goods online: the sale may not be counted in Shanghai, so measuring consumption by sales location is affected by platforms and settlement arrangements.
• As Lu describes it, most retail sales statistics cover goods, while the service component includes only some catering and accommodation; many other services are excluded. First-tier cities have higher incomes and more young people, and consumption there skews more toward services, so the measure may not fully capture the shift in what people actually spend on.
• He still sees problems specific to the cities themselves: industrial transition and policies that have long been unfriendly to young people can weaken the consumer base. The number of young people in Beijing has been falling persistently; Shanghai is somewhat better but faces similar pressure. Slower growth in the young population is not good for consumption.

5. Before debating a city’s population ceiling, align the definition of its metropolitan area

• Lu cautions that Chinese prefecture-level cities and municipalities may cover 10,000-20,000 square kilometers; Shanghai itself covers more than 6,000. He says such administrative areas would often be closer to a county or metropolitan area abroad. Comparing Beijing’s 20 million people or Shanghai’s 25 million with the population of an overseas “city” mixes different definitions.
• He uses the Greater Tokyo Area—about 12,000 square kilometers and 37 million people—as a reference, saying its area is roughly equivalent to Shanghai plus Suzhou. If the circle is drawn around central Shanghai, it would reach Kunshan and Taicang, not Suzhou’s urban core; on that definition, he puts the population at about 31 million.
• Expanding capacity is therefore not just a matter of packing more people inside existing administrative borders. It means developing metropolitan areas as connected regions and allocating land to meet the needs of industry, infrastructure, schools, hospitals, nursing homes and housing. More housing supply also means more people can stay.
• Capacity can change as infrastructure improves. Shanghai once faced “water-quality-driven water scarcity”; after the Qingcaosha Reservoir was built, Lu says the city gained tap-water supply capacity for about 8 million to 10 million more people. “A city’s capacity to support population is a variable, not a constant.”

6. A rising services share means continued concentration in big cities and even their urban cores

• Lu cites estimates that about 50% of South Korea’s population is concentrated around Seoul, about one-third of Japan’s in the Greater Tokyo Area, and about one-third of the UK’s around London. Japan offers a more intuitive comparison: with a population of about 120 million and an area close to that of the Yangtze River Delta, one-third of its population lives in an area equivalent to Shanghai plus Suzhou.
• In his view, this concentration reflects the convenience of work and daily life, not a policy advocated by economists. “People don’t concentrate because economists tell them to; they vote with their feet.” Agglomeration economies continue to work even where governments try to limit the population of central cities.
• The larger the services share, the more important in-person interaction, learning and offline consumption such as concerts, movies and sports become. A larger market can support a wider, more varied and more fashionable range of services. That is why Lu argues in Centripetal Cities: “People concentrate in expensive places, not cheap ones.”

7. Big cities rarely lose their status; technology and political systems are important exceptions

• Asked by Li whether megacities can decline, Lu says “very rarely,” but starts with technological change. Boston led in the early history of the US; later, as water transport and international trade expanded, New York gained prominence thanks to the Hudson River’s outlet to the sea and its links between inland shipping and the port.
• War, natural disasters and the planned-economy practice of locating industry by administrative decision can also reshape a city’s fortunes. Lu points to industrial cities such as Shiyan: a city built around a major factory may contract after marketization. Setting those forces aside, he sees the location and spatial distribution of big cities as broadly stable.

8. Regions losing population should adapt to the trend; Jingdezhen is not a template for everyone

• Lu stresses that reform and opening up profoundly reshaped the division of labor between cities. As long as the economy is open and international trade relies mainly on sea transport, the coasts of the East and South China Seas have an advantage. Manufacturing declines and population outflows in some regions may have deep historical and geographic causes that cannot be reversed by will alone.
• For places far from the coast, major rivers and big cities, he proposes more concentrated development, centered on urban districts, while seeking sectors where they retain a comparative advantage, such as agriculture, tourism and natural resources. During research in the northwest, he saw specific opportunities in solar power in deserts and Gobi areas, and in raising marine fish in high-salinity salt lakes.
• He believes population decline may be a long-term trend in some places, even “unavoidable.” National support can still improve the environment and the quality of life in county towns, making low housing costs and a slower pace attractive; that is different from requiring every place to return to population growth.
• Jingdezhen’s opportunity came from its porcelain-industry base: shifting from mass production to many small workshops, then combining that with talent development and an influx of young people from China and abroad to build a new industry. Lu emphasizes that these conditions “are not common.” Local governments and residents need to find their own opportunities; outside scholars cannot simply prescribe a copy-and-paste model.

9. Innovation can become more concentrated without necessarily widening per-capita living-standard gaps

• Lu credits universities with helping cities develop: four years of study can build emotional ties, create internship opportunities and raise the chances that graduates stay to work locally—but only if there are companies suited to their fields. Li adds that companies also move in because universities and talent are concentrated, creating a mutually reinforcing cycle.
• In response to Li’s question about whether technology widens inequality between cities, Lu distinguishes between the short and long term. AI may widen regional gaps in the short term, and innovation is more concentrated than economic activity, which in turn is more concentrated than population. That does not mean every kind of gap is widening.
• He distinguishes totals from per-capita measures: gaps in population and total economic output may widen even as gaps in per-capita GDP, real income after living costs and quality of life narrow. Big-city wages are higher, but so are housing and other costs, commuting times and stress; in a smaller city, being able to go home for lunch is a non-monetary difference. His research finds that China’s per-capita gaps are narrowing, though they remain large.

10. Moving to a big city is not a universal recommendation; career, preferences and relationships determine the choice

• Li brings up Paul Graham’s advice that young people should move to big cities. Lu declines to offer a one-size-fits-all answer: “I don’t make recommendations.” The broad pattern of population flows cannot replace an individual assessment of their own circumstances.
• Start with career. Industries such as AI are better suited to big cities, while some creative and arts workers prefer quieter places. Lu makes the trade-off concrete: living in a small apartment downtown and working at an AI company, or returning to a smaller city to provide digital services to the government, bring different benefits and costs.
• Personality and social ties matter too. Whether someone can accept an expensive, small apartment and a long commute, whether they want a rich urban life, and whether they have friends and family nearby all affect the choice. “If you love money, stay in a big city” is just his colloquial shorthand; he adds that many people stay simply because they find big cities fun.

11. AI innovation needs frequent, cross-disciplinary and serendipitous exchange, drawing companies back downtown

• Lu observes that companies clustered in Silicon Valley during the internet era, while this wave of AI firms is moving further into downtown San Francisco. He describes modern innovation as “high-frequency,” “cross-disciplinary” and “random”: someone in AI might talk every day with people working in philosophy, ethics or design, or meet a potential collaborator by chance at an event.
• Companies’ return to downtown also supports consumer spending. Employees grab coffee, go for drinks after work or see a play, attracting more service businesses. Agglomeration therefore affects not only PhDs and engineers, but also whether service workers find jobs in a big city or return home to open a shop.

12. Downtown housing is like a front-row ticket; residents should recognize what they have already paid for

• Lu’s analogy is that housing in a big city is like a concert ticket, and downtown housing is a “front-row seat.” Living close to performances, exhibitions and other activities costs extra; the convenience of the location is part of the home’s price.
• When he encourages students to go out more, he lists “three payments” they have already made: the urban-convenience premium in rent or the purchase price, free museums and libraries supported by taxes, and the income and quality-of-life sacrifices made to stay in a competitive city. After paying all that, staying home all the time is like buying a front-row ticket “and then falling asleep.”
• A student later told him that after hearing this, they took their child out every weekend to shop and visit museums. Lu also likes the idea of a quiet life with a yard and flowers, but his career as a university professor and his interest in exhibitions and concerts mean he enjoys the convenience of a big city.
• Choices can also change with life stage. Lu points to Americans who work in New York when young and retire to “sunbelt” destinations such as Miami, drawing a parallel with Hainan in China: “You don’t have to live in the same place your whole life.”

13. The advantages and governance challenges of being a “large country” both belong in economics research

• Lu explains that he has kept writing his “large country” series because the size of a country has not received enough attention. Large language models and platform businesses depend both on having many language users and on applications reaching a large enough market quickly. The scale of Chinese-language users is one factor in his understanding of China’s AI development.
• When globalization was surging, people put more faith in free trade, cross-border investment and coordination through international institutions, making the size of a country seem less important. As geopolitical competition has risen, the influence of large countries has become more prominent again. He also mentions Canada’s proposal for “middle powers” to coordinate in response to great-power competition.
• Governance in a large country cannot simply be modeled on the US federal system. Lu uses local debt as an example: in the US, a local government can shut down, furlough public employees or even go bankrupt. Under China’s unitary system, the responsibilities of central and local governments—and relations between localities—are structured differently.
• Population flows make the issue concrete. If fiscal transfers are still allocated according to household registration, people may work in coastal regions while education, housing and other resources remain in the places they left. The result is inadequate public services in one place and empty schools in another: a “spatial mismatch of resources” that calls for coordinated land, household-registration and fiscal reform.

14. AI replaces manufacturing jobs, then deepens concentration through service-sector employment

• Lu focuses on the secondary and tertiary sectors, which together account for more than 93% of China’s GDP. Many manufacturing jobs are “standardized, process-driven and codifiable”; 24-hour dark factories and unmanned plants show that robots and AI can already do much of this work.
• Many service jobs still require people to meet face to face. As manufacturing employment shrinks, workers will move toward services that depend on population size, density and foot traffic. Lu’s example is today’s delivery riders and ride-hailing drivers, who may have worked in factories in their previous jobs.
• This is not just a technological forecast. If people conclude that AI will make agglomeration disappear, they may decide there is no need to keep improving public services for migrants; if concentration is set to accelerate, the case for policy action becomes stronger. The same applies to homebuyers: believing people will leave big cities or continue to move in leads to very different decisions.

15. The intensity of local competition is linked to control over land and tax revenue

• Lu says races to the bottom exist in the US, Japan and Europe too, but their scale is not directly comparable with China’s. In China, the central government sets quotas for farmland and construction land, while local governments decide whether land is used for housing, industry or commerce and even how tall buildings can be. Land, fiscal and household-registration rules together give local governments greater scope to intervene in the economy.
• Li counters with Amazon’s HQ2 bidding process, asking why China does not seem to have similarly public contests for headquarters. Lu says a company usually has only one headquarters, so the focus should broaden to regional headquarters, branches and subsidiaries that can settle their accounts independently, rather than fixating on the single headquarters location.
• In his view, settlement and tax revenue often accrue to the headquarters or place of registration. Local governments therefore especially welcome independent legal entities and settlement centers. A branch that merely conducts business may still leave tax revenue with the headquarters; a local settlement entity makes it easier to keep taxes, output and jobs in the area.

16. Local growth from county-level competition does not automatically add up to national welfare

• Li puts Zhang Wuchang’s emphasis on county-level competition alongside Lu’s call for a unified national market. Lu proposes a test: “If every locality tries to maximize local output, will those outputs automatically add up to maximum national output?” His answer is no.
• He uses “long-term, holistic and multidimensional development” to unpack the conflict. Borrowing to invest can lift near-term growth but leave debt that cannot be repaid; every locality building a car plant can boost local investment but create duplication and excess capacity nationwide.
• In China’s early development, capital was scarce, so competing for investment helped build the capital stock. It was not necessarily a problem for different places in Zhejiang to produce shoes and pearls. Today, economies of scale are stronger in autos, large aircraft and AI, making concentration in a smaller number of regions more efficient. Yet localities may all compete to develop AI, big data, cloud computing, the Internet of Things and biomedicine, making the conflict sharper.
• Lu does not describe the earlier period as all upside. Market fragmentation has always existed; the benefits of competition were simply overstated while its costs were overlooked. Compared with neighboring counties blocking one another’s goods in the 1980s, market integration has improved. But changes in the stage of development make the remaining problems more urgent to address.

17. Pro-business competition should mean a better business environment, not unconditional subsidies

• Asked by Li whether local competition is at least good for companies and investors, Lu again says “not necessarily.” A petrochemical project in a part of Hunan had an industrial base and was planned with environmental conditions in mind, so the investment could be beneficial. Bringing in a centrally owned company just to boost growth without those conditions could instead mean environmental damage and long-term costs.
• Lu describes a cycle: in places with good conditions, companies grow through the market and do not want interference; local governments have so many businesses that they cannot oversee them all. Places with weaker conditions rely more on investment promotion and intervention, and seek help from companies, which can worsen the business environment. Outside investors may also be uncomfortable with a system where everything requires personal connections.
• He rejects “investment doesn’t cross Shanhaiguan” as a way to describe the Northeast, saying it stigmatizes an entire region, and specifically notes Liaoning’s recent efforts to improve the business environment. What is worth competing for over the long term is a welcoming environment for private firms, fair competition and the rule of law: “Improving the business environment is the best way to attract investment.”
• But making the argument alone is not enough. When taxes are based on production and settled at a company’s place of registration, and performance reviews emphasize GDP and tax revenue, local governments still have incentives to compete for projects. It is especially unsustainable to keep judging regions with large population outflows mainly on growth. Views of official performance, the tax system and evaluation criteria all need to change.

18. Auto-sector involution runs down the supply chain; exports and output do not mean everyone makes money

• Li cites figures that, in the first half of 2020, total auto-industry profits were a little over RMB20B, with BYD accounting for half. He says that figure may have been similar to CATL’s profit of a little over RMB20B in the first half, and asks whether the industry’s involution is rooted in past local-government competition.
• Lu says the causes are complex, but localities competing to attract auto projects are one part of the story. He notes that China now produces more than 30 million cars a year. With a limited domestic market, companies turn to exports and price cuts; if local governments keep supporting loss-making firms, competition and capacity expansion may drag on, and subsidies may ultimately be wasted.
• Another part is downstream companies passing price pressure upstream: automakers squeeze tire suppliers, who in turn squeeze their own suppliers. Output, exports and sales can all look strong while profits throughout the chain remain thin or nonexistent. With serious payment delays, companies end up in a bind: “If you don’t do the work, you get no business; if you do it, you don’t make money.”
• Cost-cutting also takes the form of layoffs and overtime. Lu cites average weekly working hours of 48.6 at companies in the previous year, saying excessive hours violate labor law to a significant degree and crowd out time for service consumption. “Put bluntly, domestic consumers’ welfare is being sacrificed to subsidize foreign consumers.”

19. Li asks who would dare raise prices first; Lu turns to the conditions for competition and enforcement

• Li’s counterargument is direct: automakers are already competing so fiercely—who would dare raise prices and share the profit with suppliers? Lu acknowledges that if a foreign company sells for 100 yuan and a Chinese company for 99, then raising costs and needing to charge 101 would indeed cost it competitiveness.
• But first, he says, it matters whether the competition is international or between Chinese companies. He offers a different scenario: foreign companies sell for 100 yuan, while a Chinese company sells for 50 only because another Chinese company charges 49. Under those conditions, he believes a Chinese company could charge 60 or 70 yuan without losing international competitiveness.
• That is why he stresses enforcement of labor rights, environmental rules and intellectual-property protections, and raising the cost of violations. In his research, small companies have complained that “innovation gets copied, and the company’s core team gets poached.” In his view, the country cannot demand innovation while tolerating IP infringement and lower compliance standards.
• Subsidies should not indefinitely delay the exit of loss-making firms. Lu argues that local governments rescuing companies on the verge of bankruptcy prolongs destructive competition and excess capacity, and invites overseas questions about fair competition. Lead firms should also ensure their suppliers and employees benefit, rather than focusing only on their own profits.

20. Consumption-tax reform could change local governments’ incentives to compete for production

• Lu explains that VAT is closely tied to production, while consumption tax covers a narrower base and is often collected at the production stage, making it mainly a central-government tax. Cigarettes are his example: consumers ultimately bear the tax, but it is collected at the factory, so local governments do not get the same incentive to promote consumption.
• The reform he describes would broaden the consumption-tax base, move collection to the consumption end and increase the local-government share. That would give local governments more reason to promote consumption. Since consumption depends on people, public services, quality of life and attracting residents would be more likely to become priorities.
• Lu stresses that major tax reform “cannot happen overnight,” but says the issue is urgent. In the near term, the response also requires central coordination, fair-competition reviews, strict enforcement and a change in how officials are evaluated. In a domestic market without regional tariffs, local governments should not protect local production through subsidies, cheap land or tolerance of illegal conduct; they need to consider “the national interest as a whole.”

21. Learn from the US, Japan and Southeast Asia—taking their strengths while avoiding their old traps

• What Lu values in the US is the involvement of private high-tech companies in major projects and the ability of its research and education systems to attract talent from around the world. But he points to the California high-speed-rail project’s planning and litigation costs to show how the system can constrain cross-regional coordination and infrastructure construction.
• In Japan, he admires the capacity to introduce, absorb and adapt technology, as well as corporate social responsibility, compliance and respect for local laws and cultures when investing abroad. The lesson to avoid is continuing to subsidize loss-making firms and preventing inefficient companies from exiting.
• He also suggests learning from social organizations in some Southeast Asian countries. Even with lower per-capita incomes, their social participation on poverty, equality for ethnic minorities and women’s rights offers lessons for China. Answers should not be sought only in the highest-income countries.
• Li asks whether these strengths are closely linked to liberal traditions of governance. Lu acknowledges the role of history but says countries should not be bound by tradition. Borrowing to build infrastructure may have worked in the past; today, regions losing population may be left with eight- or ten-lane roads and vacant housing. As development stages change, every country needs reform.

22. When the technology path is unclear, industrial-policy support should not be too narrowly targeted

• Lu sees Japan’s bet on hydrogen as the product of both government and business decisions, and says that, in retrospect, it “wasn’t that successful.” His point is not to stop supporting new energy, but not to equate new energy prematurely with a single option as if the answer for the next few years were already known.
• AI faces the same problem. There is no settled answer on whether embodied intelligence, physical robots or large language models are the right path. Even large companies may disagree internally, sometimes prompting teams to leave. Lu sees great risk in a government declaring that it has already chosen the only direction.
• He therefore argues for preserving diverse competition and the possibility of failure, making government support broader and relying more on rewards after the fact than on selecting all the companies and technologies in advance. One adjustment he proposes: “Reward success; let failure be淘汰.”

23. Industrial policy must be considered separately at the central and local levels; the real debate is over implementation

• Lu’s position is clear: “There has to be industrial policy.” Early-stage industries may lack clear market prospects and investment, and may require coordination across finance, land, taxes, talent development, regions and industries—work that start-ups cannot do on their own.
• In China, however, central and local industrial policies need to be discussed separately. Lu sees no problem with the central government encouraging high tech, AI and biomedicine. The greater concern is that, after a single national call, localities all launch the same projects without adapting to their circumstances, leading to subsidies, duplication, pockets of excess capacity and trouble repaying debt.
• The debate should therefore move from whether there should be policy to “how it should be implemented.” The central government needs to improve the institutional environment through the rule of law, fair-competition reviews and tax reform. Local governments need policies suited to their own conditions, rather than reducing industrial policy to investment promotion and leaving “a pile of messes” behind.

24. Hefei’s success needs to be attributed to location, talent, the company’s stage and the timing of investment

• Asked about “learning from the Hefei model,” Lu first cautions that some investment-promotion tactics used in the past are precisely the kind no longer allowed under today’s fair-competition reviews. He also mentions a recent central-government policy prohibiting district- and county-level governments from setting up investment funds, and asks: if government investment guarantees success, why shouldn’t every locality be encouraged to do it?
• He is especially concerned about funds’ “reinvestment in the locality” requirements: the government provides capital, and the company must invest locally, set up a company, hire workers and generate tax revenue there. But companies should choose their locations based on costs and benefits; those restrictions can fragment the market. His team’s research, he says, suggests that these issues should be included in fair-competition reviews.
• In explaining Hefei, Lu says the Yangtze River Delta location, USTC’s talent base and BOE’s high-growth stage—when it happened to need a location—must all be considered. Did success come from those conditions and the timing, or simply from government investment? Li agrees: “We should do the attribution carefully.”

25. Demand for services is already there; attitudes and restrictions on places to consume may be holding it back

• Asked what worries him most, Lu says it is not a lack of opportunity, but that opportunity is blocked by attitudes and traditional rules. Services have been growing faster than manufacturing for years, not just in the first half of this year. Goods prices sometimes fall while service prices rise, which in his view shows that real demand for many services exists.
• Supply constraints take concrete forms: restrictions on activities on rivers, skateboarding and outdoor restaurant seating; requirements to clear out within half an hour after concerts; and some subway services ending at 10:30 p.m., all of which can cut off follow-on spending. Developing services is not just about adding businesses; it also depends on everyday management rules.
• He rejects the idea that personal services are simply low-end. Healthcare, education, eldercare, sanitation, childcare and dining are all foundations of quality of life. High-income consumers have needs too: insufficient golf-course supply, for example, may send spending overseas. Meeting that demand can also create domestic jobs.
• Lu does not deny that households may lack money, and agrees that a return to growth, a stronger capital market and a recovery in real estate could boost income and wealth. His point is that another part has not been discussed enough: open up service supply, reduce overtime and add holidays so people have time to turn demand into spending.

26. A near-term response to aging also means moving existing workers to higher-income regions

• Lu supports raising the retirement age, encouraging childbirth and improving workers’ skills, but notes that children born today will not enter the workforce for 16 years. In the meantime, higher birth rates cannot be expected to increase labor supply quickly; other measures are needed.
• His solution is free movement within China. Someone who can earn 2,000 yuan in their hometown might earn 4,000 or 5,000 yuan elsewhere. That would sharply raise the individual’s income, make better use of the national workforce and strengthen the resources available for eldercare. That is why he places particular importance on spatial planning, city clusters and the urbanization rate.
• The obstacle remains in the destination regions. Lu says developed areas face greater pressure from aging and low birth rates, yet may treat migrants as a burden on public services and remain constrained by fiscal and land rules. He points out that Shanghai, at around 6,000 square kilometers, still has one-third of its land under cultivation, and asks whether places with high housing prices and high locational value need to retain such a large share of farmland.

27. In the AI era, education is shifting from how many years to what students should learn

• Lu calls memory, rote learning, calculation and derivation “machine functions”—capabilities that machines can also perform. He is clear that this does not mean primary and secondary schools should stop teaching memory and recall, since children’s brains still need training. What needs to be reconsidered is whether universities should continue to devote so much time to these tasks.
• He wants universities to offer more cross-disciplinary electives, general education and practical experience, developing aesthetic judgment, values, policy understanding, and insight into human trends, human nature and market demand. Science and engineering students in particular need to do more than complete assignments; they should have more exposure to companies and society. Faculty and curricula also need to change.
• Lu says that as high school becomes more universal and declining birth rates make university entry easier, graduation rates are no longer the only priority. Citing discussions in the Northeast about “everyone going to university,” he brings the question back to content: “What people should really care about is not the university graduation rate, but what students learn.”

28. AI changes not only the methods of economics, but also scarcity, decision-making and organizations themselves

• Li asks whether supply-demand relationships could change if AI creates an enormous supply of goods and services. Lu says academia has discussed using AI for modeling and statistics, but much less often how AI changes the subject being studied. In some areas, AI will greatly increase supply, making resources that were once scarce no longer scarce—and making other things scarce instead.
• The decision-making subject is changing too. AI can provide information and participate in decisions, potentially amplifying emotions and creating hallucinations and false expectations. At the agent stage, it will no longer just execute instructions but act as a representative; organizations will face “multiple principal-agent relationships,” requiring new institutional arrangements.
• Companies may become smaller and flatter, and work may no longer involve an employer. Lu notes that an OPC may have a few employees, or in some cases just one person. Once there is no traditional employment relationship, how social insurance and protection systems should cover these workers becomes a new question for economics.

29. Broad-access and paid AI models each have costs; Lu is offering a view of the likely models, not a settled conclusion

• Lu believes that the US model—in which large companies invest and then need to recoup their investment—will “most likely” lead to companies charging for access. If people need money to make full use of AI, this will create divisions between individuals and between companies. He sees that as an important consequence the model will have to address.
• On China, he qualifies his view as based on the information he has received so far, and describes the country’s current public, broad-access direction as “socialism in the AI era.” Local subsidies that help one-person companies afford AI are his example of how this approach might work.
• He does not discuss only the benefits of broad access. Could cheaper compute be wasted—for example, used to write novels for personal amusement on subsidized capacity—and how should that be regulated? Could companies use subsidized compute to replace workers and worsen employment pressure? In his view, both questions remain to be studied and answered.

30. More mature research methods can also become a burden when responding to new realities

• Lu values mathematical modeling, mathematical statistics and causal analysis in economics, and believes they have helped the discipline mature. But in the face of rapid real-world change, established frameworks can also become constraints. He repeatedly clarifies that he is not rejecting models; his point is that “economics should not use its methods to cut its questions to size.”
• New problems may lack theory, literature or data, and may arise from institutional features specific to China that are difficult to model for now. Lu says that is no reason not to study them: “We should not refuse to research something because the methods do not fit or it cannot get published in a journal.” That cannot be an excuse for failing to respond to society and government.
• He traces the internationalization of Chinese economics through the founding of Peking University’s China Center for Economic Research in 1994 and the introduction of modern economics teaching at Wuhan University. Closing the international gap is an achievement, but promotion criteria that increasingly emphasize publication in top international journals may marginalize important Chinese questions that lack standardized data and models.
• Lu criticizes himself and his peers, as well as younger scholars, for their limited public engagement. He worries that peer evaluation has become “you evaluate me, I evaluate you,” with less concern for external needs, turning the discipline into “a game played behind closed doors.” If society decides the discipline is no longer needed, economics itself will lose its vitality.

31. As innovation shifts to companies, universities have to explain their value again

• Li notes that frontier generative-AI research is moving from university labs to companies. Lu explains that this wave of innovation requires capital of hundreds of millions of yuan and the ability to use large companies’ compute and application scenarios to validate results quickly. Universities often lack both the money and the settings needed to test ideas.
• He observes that companies in both the US and China are starting to hire high school graduates directly. Students may weigh four years at university for a degree against four years at a major company, where they may learn more. Lu’s warning to universities is blunt: if they keep teaching “things that don’t matter,” they can “just wait to be replaced by big companies.”
• Universities are already responding. Some leading US institutions are combining the humanities with data science and AI. At the Bund Summit, Lu heard a university vice-president emphasize externally funded, industry-linked projects that keep teams connected to business practice; that made him more confident in his own team’s ability to adapt.
• Lu argues that externally funded projects should not automatically be dismissed as money-making or business. Basic and applied research are now closely connected. He is especially critical of the weak response from social sciences and economics in China: interviews, engagement with real-world issues and government projects may not count toward evaluation and can be treated as a waste of time—or prompt accusations that someone is “trying to become an internet celebrity.”

32. Whether to join an AI company means weighing frontier opportunities separately from personal responsibilities

• Asked whether economics research might also move into companies, Lu says he cannot predict. But if he imagined being 30, with no current team or disciplinary responsibilities, he would “accept without hesitation” an offer from an AI company. In reality, now past 50, he has to consider his research institute, team and discipline at Shanghai Jiao Tong University, so he would hesitate.
• His support for joining companies is not just talk: his team has substantial collaborations with platforms and AI companies, and a PhD who graduated that year is interviewing with a research institute at a major Chinese company. Platform companies care about consumption on their platforms, but academic research may not answer the questions companies actually have. That is the gap he wants to narrow.
• Lu admires Paul Krugman and Edward Glaeser, author of Triumph of the City, because both take real-world problems seriously. He particularly rejects the idea that Krugman’s media columns undermine his academic work: explaining research to the public is a way to serve society, and does not erase the originality of his award-winning work.

33. Policy involvement and independence are compatible; research must leave room for different conclusions

• Lu believes economists as a group should be more involved in practice and national development, but that does not mean everyone should work on the same kind of applied question. Mathematical modeling and research into general principles that apply to humanity also matter. His criticism is that the field as a whole is not responding enough to important questions within its own remit.
• His example is monetary transmission: monetary policy has been eased, yet it does not appear to have adequately stimulated consumption, credit or investment. The country wants to understand the mechanisms and get advice. If even a core economics question like this lacks enough research, Lu says, the discipline has a problem.
• Independence means not “deciding on the answer before doing the research” and then tailoring the work to fit. He recalls that many of his positions 10 years ago differed from the policies of the time, but later became a basis for reform. He has advised central and local governments and says he is a member of the expert group for the country’s 15th Five-Year Plan; that does not, in his view, compromise academic distance.
• That distance also shows up in new research: much rental housing is built in suburbs, even though many people seeking it work in service jobs concentrated downtown. The distance can leave demand weak and units vacant. Lu argues for more housing in city centers, showing that close engagement with policy can still lead to conclusions that differ from current practice.

34. Separate emotional ties to home from buying property there; people may need city services more as they age

• Lu cautions against treating a temporary shock to big cities and workers returning home as proof that population concentration has permanently reversed. He has seen people rent in a big city, buy a home in a county town and build a house in the countryside at the same time. That arrangement may rest on an expectation of eventually returning home.
• In the Pearl River Delta, he met a woman who went to Guangzhou at 17 and is now around his age. She said she liked Guangzhou but had also built a house back home, planning to return “when I can’t get around anymore.” Lu frames his view as a probability: if 99 out of 100 people think that way, more than half may never return—especially because, when their health deteriorates, they may need hospitals, caregivers and familiar social networks.
• He says this does not prove that her choice is wrong. But vacant homes and falling property values in places losing population may hurt household welfare. Feeling attached to one’s hometown and returning to see family and friends are separate questions from whether to buy property there as an investment or return when one’s health worsens.

35. General principles and China-specific institutions can form two layers of the same research

• Asked how to balance enduring research with a fast-changing reality, Lu puts the answer in a combination of two layers. Manufacturing employment declines as incomes rise and demand for services increases; that is a general pattern. Public services are not yet fully open to migrants and land allocation is regulated; those are China-specific features. The former explains the pattern, while the latter points toward specific reforms such as floor-area ratios.
• He acknowledges that English-language journals prefer general principles. Explain China’s institutions in detail and readers may not know them; explain them too briefly and the analysis becomes unclear, which can hurt publication prospects. But whether to give up is a personal choice: “I live in China, and I’m a locally trained PhD. That is no reason for me not to study China’s problems.”

36. AI infrastructure and data coordination require national coordination, but administrative power has limits

• Asked about Elon Musk’s suggestion that the US should learn from China’s coordinated investment in AI infrastructure and energy, Lu says the idea makes sense. Large-scale infrastructure creates “positive externalities”: companies may be unable to recover the cost through market pricing, and projects require coordination across industries and regions. He cites China’s ability to coordinate solar power in the central and western regions with compute and storage centers.
• Coordinating data is another shared challenge. Platform data is walled off by company and industry, while government data remains unconnected. The difficulty increases from sharing across departments in a city, to cities in the same province, to nationwide integration. Bringing in private-sector data also raises the question of compensating the companies and individuals that provide it.
• Technical frictions are real: different departments use different storage formats and time variables, and records may not line up for the same person or company. A willingness to share does not mean the data can be combined and used directly.
• Institutional frictions matter too: department heads may not want to share, and a big-data office at the same administrative level may lack the authority to coordinate other departments. Lu concludes that power divided by department and region can work well for internal issues but become an obstacle across boundaries, requiring coordination at a higher level. China still has to work to turn its advantages in market integration and central coordination into better outcomes.