For two decades the dominant Western story about Chinese manufacturing was a story about the middle of the smile curve. Foreign brands kept design, software and marketing. Taiwanese intermediaries ran electronics manufacturing services and, later, pure play foundries. The People’s Republic of China supplied land packages, customs facilitation for processing trade, and a hukou segmented migrant labour pool large enough to staff launch economics. The arrangement produced employment, export growth and spectacular volumes. It also produced thin assembly rents and thick brand and intellectual property rents elsewhere. That distribution was never a stable end state. It was a developmental lease.
The lease is now being renegotiated in public. BYD sells more battery electric vehicles globally than Tesla on recent tallies. CATL and BYD together hold more than half of world EV battery installations. Huawei has clawed back the top smartphone slot in mainland China after five years under US entity list pressure. Solar manufacturing is so concentrated in China that overcapacity has become a trade complaint rather than a development success story. Western tariffs, export controls and industrial subsidies are reactions to a rival claiming higher value chain positions, not to a permanent workshop content to assemble other people’s products forever.
The analytical task is to describe that claim without collapsing into either Beijing’s self congratulation or Washington’s moral shorthand. Hukou discrimination, dormitory labour regimes and processing trade platforms are real political economy. They are not chattel slavery, and treating “slave state” rhetoric as literal analysis obscures the upgrading project that Chinese policy and firms have pursued since at least the mid 2000s Medium and Long Term Plan for science and technology.
The bargain: scale rented, rents reserved
Reform and opening after 1978, and the first special economic zones from 1980, created controlled spatial experiments for foreign direct investment. Shenzhen and its peers were not only cheap labour enclaves. They were institutional learning devices: land lease fiscalisation, bonded processing, local cadre competition for GDP and jobs, and gradual absorption of management practice. By 1992, processing exports were already around 46 percent of China’s exports, according to CEPII work associated with Françoise Lemoine and colleagues. The share rose toward 55 percent by 1996 and stayed above half through the late 1990s. By 2000, foreign affiliates accounted for roughly 70 percent of processed exports and about three quarters of processing imports.
WTO accession in December 2001 locked in tariff cuts and predictability that deepened global value chain integration. Through the mid 2000s the classic modular electronics triangle hardened. US and other Western brands retained the high rent ends of the smile curve. Taiwanese EMS and ODM firms, notably Foxconn after its 1988 Shenzhen opening and its later Zhengzhou scale up, organised labour intensive final assembly. TSMC’s foundry model separated fabrication from branded design. The PRC supplied demography, infrastructure and local state co authorship of megafactory campuses. Dedrick, Linden and Kraemer famously showed Apple capturing the bulk of iPhone gross margin while Chinese assembly labour was a small absolute slice per unit. Foxconn parent profits accrued in Taiwan. Chinese localities got employment, taxes and land related revenue. The brands got the capitalised intangibles.
National Bureau of Statistics migrant monitoring still shows the labour base underneath that bargain. Migrant workers numbered 297.53 million in 2023 and 299.73 million in 2024. Manufacturing employed roughly 27.5 to 27.9 percent of them, on the order of 82 to 84 million people. Average monthly manufacturing migrant income rose from ¥4,780 in 2023 to ¥4,978 in 2024. Those are wages in a segmented labour regime, not a permanent comparative advantage theorem. Hon Hai’s FY2024 gross margin near 6.25 percent remains the EMS reminder: volume is not the same as rent capture.
Why the lease was unstable
Three mechanisms eroded the “forever workshop” assumption. The first was learning by doing inside dense supplier ecosystems. Process knowledge, engineer cadres and functional upgrading from pure import and assembly processing toward more domestic sourcing are measurable in trade accounts. Stanford SCCEI summaries of Cai, Wang and Wei’s work on The World Economy find domestic value added in Chinese exports rising from 66 percent in 2007 to 76 percent in 2020, while the processing share of exports fell from about 41 percent to about 23 percent. MIC2025 linked sectors saw larger DVA gains: ICT roughly plus 13 percentage points, machinery plus 12. By 2020 private firms were about 55 percent of exports with DVA near 81 percent. Policy did not invent every percentage point. It did not need to. The direction of travel was already away from pure assembly anonymity.
The second mechanism was explicit indigenous innovation policy. The 2006 Medium and Long Term Plan targeted R&D intensity at or above 2.5 percent of GDP, foreign technology dependence below 30 percent, and a larger science contribution to growth by 2020. Made in China 2025, issued by the State Council in May 2015, diagnosed Chinese manufacturing as “large but not strong” and set self sufficiency ambitions for core components and materials, with sector roadmaps that Western summaries often put in the 70 to 90 percent domestic content range by 2025 to 2030. Those targets were aspirational and political, not audited outcomes. MERICS and Rhodium both note that after 2018 the MIC2025 brand was de emphasised under foreign pressure while substantive industrial policy continued under other labels. Dual circulation, articulated in 2020, cast domestic circulation as mainstay and international circulation as complement: a hedge against tech decoupling and a programme for capturing more value at home.
The third mechanism was domestic market scale plus patient capital, public and private. Joint venture and market access leverage, talent recruitment, open source absorption and outbound acquisition all featured in Western complaints. The forced technology transfer debate remains live. USTR Section 301 reports from 2018 and the 2024 four year review treat JV requirements, opaque licensing and cyber theft as systematic coercion. Academic and legal analyses stress that explicit WTO illegal conditions were often harder to document than informal approval leverage, and that spillover is not proof of force. The Foreign Investment Law of 2020 added voluntary cooperation language. The essay need not settle the morality play. It needs only to note that upgrading combined learning, subsidies, scale, contested coercion and entrepreneurship rather than a single cause.
UNCTAD’s later reassessment of Chinese special economic zones stresses continuity as well as change: early FDI was heavily Hong Kong labour intensive light industry before electronics clusters thickened. The political economy lesson is temporal. When China was capital scarce and brand poor, renting land and hukou segmented labour to foreign smile curve winners looked mutually beneficial. The same learning goals that justified the zones, technology and management absorption, made indefinite specialisation in the thin middle irrational for Beijing and for Chinese firms that had acquired process capability. Comparative advantage is a snapshot. Industrial policy treats it as a moving target.
Evidence of ascent: jagged, not mythical
Clean tech and consumer electronics show the rent claim most clearly. China passenger NEV retail data for 2025 compiled by CPCA and reported via CnEVPost put BYD at 3.48 million units and 27.2 percent share, down from 34.1 percent in 2024 but still dominant; Tesla China fifth at 4.9 percent; Xiaomi Auto at 411,800 units and 3.2 percent. BYD’s global BEV sales of 2.26 million in 2025 exceeded Tesla’s 1.64 million on the same reporting. CATL held 43.42 percent of China’s EV battery market by volume (333.6 GWh) and BYD 21.58 percent (165.8 GWh) per CABIA. SNE Research global figures for 2025 put CATL at 39.2 percent and BYD at 16.4 percent of installations, combined above 55 percent, on 1,187 GWh worldwide, up 31.7 percent year on year. These are not assembly subcontract statistics. They are chemistry, scale, customer lock in and brand rents.
Smartphones tell a related story under sanctions. Omdia data for mainland China in 2025 show Huawei returning to first place with 46.8 million units and about 17 percent share after five years, ahead of vivo, Apple, Xiaomi and OPPO in a 282.3 million unit market that shrank one percent. Huawei’s trajectory from telecom gear into handsets, then into operating system and chip localisation after the May 2019 Entity List designation, is a sovereignty project as much as a consumer brand recovery. Ascend AI chips fabricated with SMIC advanced DUV processes without EUV access show catch up that is real, costly and capacity constrained. Early yield reports near 20 percent for some 910C vintages, widely circulated in 2024 trade press, counsel against claiming parity with Nvidia and TSMC stacks.
Solar and steel illustrate ascent’s dark twin. Industry and CSIS linked tallies for 2024 still cite Chinese shares on the order of 93 percent of polysilicon, 97 percent of wafers, 92 percent of cells and 86 percent of modules. Price wars and utilisation problems followed. worldsteel figures put Chinese crude steel near 1,019 million tonnes in 2023, about half of world output. Scale destroyed foreign competitors’ margins and often Chinese producers’ profits at the same time. Overcapacity discourse in Washington, Brussels and the OECD is partly protection of incumbents and partly a correct observation that subsidy driven capacity can flood markets faster than demand can absorb it.
Standards and system exports matter alongside branded goods. The Jakarta Bandung high speed railway entered commercial service in October 2023 with CRRC EMUs rated to 350 km/h. The China Laos railway has operated since December 2021. Both packages export rolling stock, construction, finance and operating norms, not only steel. DJI’s consumer drone category ownership, commonly estimated above 70 percent in industry tallies, and Shein or Temu’s assault on Western retail marketing and logistics rents, show different routes to the same end: exit from OEM anonymity. Shein revenue ballparks near $38 billion for 2024 should be treated as directional secondary estimates, not audited filings.
Champion structure matters for who captures rents. CATL, BYD, Huawei and DJI are private or mixed ownership stories as much as state stories. CRRC and heavy rail remain more obviously SOE centred. Credit, land, procurement and standards still tilt the field even where equity is private. Collapsing the ascent into “only SOEs” or “only theft” erases domestic competition that is often brutal: price wars in solar and EVs are not the behaviour of a single planner allocating quiet monopolies. They are the behaviour of scale seekers burning margin to win market share and learning curves, then meeting tariff walls abroad.
Pushback as confirmation
If China were still only a low wage assembly platform, the policy response would look different. US Section 301 tariffs from June 2018 began with an additional 25 percent on roughly $50 billion of industrial and technology goods aimed at MIC2025 linked sectors. The 2024 four year review raised strategic rates further: electric vehicles to 100 percent, solar cells to 50 percent, semiconductors to 50 percent on a phased schedule, among other lines. The CHIPS and Science Act, signed 9 August 2022, authorised about $52.7 billion for US semiconductor manufacturing, research and workforce. Entity list and allied tool controls try to deny chokepoint inputs, especially EUV and advanced AI accelerators.
The European Union imposed definitive anti subsidy duties on Chinese battery electric vehicles from 30 October 2024 for five years: BYD plus 17.0 percent, Geely plus 18.8 percent, SAIC plus 35.3 percent, other cooperating exporters plus 20.7 percent, Tesla Shanghai an individual plus 7.8 percent, all on top of the standard 10 percent car duty. Autor, Dorn and Hanson documented concentrated US labour market losses from the earlier China import surge, helping assemble the political coalition for tariffs. Geoeconomic containment of a peer industrial rival and reaction to real subsidy and non market practices can both be true. Pushback proves the thesis that higher rents are being contested. It does not by itself prove that every Chinese industrial policy instrument is illegitimate or that every Western instrument is pure market restoration.
Modular capitalism’s next phase
The Foxconn and TSMC story remains the hinge. Modular product architectures allowed brands, foundries and EMS firms to specialise across borders. US design and IP, Taiwanese process intermediaries, and PRC land, labour and local state scale formed a triangle. China was co author of that scale, not a passive backdrop: Zhengzhou style campuses were fiscal and logistical projects as much as factory sheds. Once co author, Chinese capital and firms predictably sought to internalise hinge functions. Luxshare takes Apple assembly share from Foxconn inside China. Huawei and Xiaomi contest Apple on brand. SMIC contests TSMC at trailing nodes under control pressure. CATL contests Panasonic and LG Chem on cells. BYD contests legacy OEMs on vehicles. China plus one assembly shifts to India and Vietnam rearrange final screwdriving more easily than they rearrange component ecosystems or leading edge fabs. Modular capitalism made China the indispensable middle. Indigenous innovation policy and private champions are the bid to own more of the ends.
Counters that still hold
Ascent is sectorally jagged. Electronics EMS margins remain thin. Garments, furniture and toys still employ millions. Processing trade shrank as a share; it did not vanish. Chokepoint dependence persists on ASML tools, Synopsys and Cadence EDA, TSMC leading edge capacity, Nvidia software ecosystems, and high end machine tools. “Made by China” is not full tech sovereignty. Demographics are tightening: ageing and a fading migrant surplus raise wage and fiscal pressure. Local government financing vehicle debt and the property downturn constrain the domestic demand leg of dual circulation, as World Bank China updates have repeatedly flagged. USCC and Rhodium assessments of MIC2025 typically score EVs, batteries and solar as strong and leading edge logic, civil aviation and some robotics as weaker where frontier science and tooling dominate.
A further constraint is software and tooling depth rather than headline factory footage. Even where Chinese fabs report advanced DUV nodes, EDA suites, IP blocks, advanced packaging know how and full stack AI software ecosystems remain contested bottlenecks. Phones can move final assembly to India faster than leading edge capacity can be replicated. China plus one therefore re sorts the visible labour intensive stage while leaving many invisible rents, and many chokepoints, where they were. That asymmetry is why export controls target tools and accelerators rather than dormitory headcount.
Overcapacity is the political price of scale ambition. Flooding world markets invites the tariffs and duties that then reinforce dual circulation rhetoric. The Western “slave state” trope fails here as analysis. Hukou segmentation and harsh dormitory discipline structured a developmental labour regime that was always intended, in Chinese policy documents, as a phase. Mistaking that phase for destiny was the Western error. Mistaking every Western complaint for bad faith would be the symmetric Chinese error. The accurate description is rent politics along the value chain: who designs, who owns the standard, who brands, who captures the intangible, and who is left with the thin middle.
China was never going to remain only a low wage assembly workshop making things for everyone else. Learning, industrial policy, domestic scale and firm level competition pushed Chinese capital toward brands, intellectual property, design, standards and higher value chain rents. Some of that claim has already succeeded. Some is blocked at the tool and software frontier. Some destroys profits even as it destroys rivals. The bargain that rented Chinese scale to foreign smile curve winners has not been cancelled. It has been revised, under tariff fire and export control pressure, into a contest over who gets the action.