Chapter 9: Industrial Relocation, Cluster Governance and Regional Differentiation

The container ship left Shekou before dawn, bound for Haiphong. Its manifest listed 340 cubic metres of polyurethane foam cushions, cut and sewn in a Longjiang workshop that had run on the same lane for eleven years. The owner had not closed the Foshan factory. He had opened a second one 900 kilometres south-west of the Pearl River Delta and was now running both. Foshan still handled the European orders that demanded FSC-certified timber and a documented carbon footprint. The Vietnamese plant handled volume orders for North America, where buyers cared mostly about price and delivery date. Two factories, one firm, two regulatory environments. The balance sheet showed a split, not an exit.

The standard account of carbon constraints and industrial geography runs like this: tighten environmental regulation in one jurisdiction and dirty production migrates to another with looser rules. Emissions fall where the rules bite, rise where they do not, and the global total barely moves. The account is not wrong. It is incomplete in a way that matters for how we think about clusters and about Chinese manufacturing. Carbon-driven relocation is rarely a flight from regulation to deregulation. More often it is a re-sorting of activities inside a firm's own network—a spatial division of labour in which the clean, documented, high-value steps stay in the regulated cluster and the carbon-intensive, low-margin steps move outward. The Longjiang owner did not move his business. He moved part of it and kept the part that paid for the brand relationship.

Three questions organise this chapter. Why do firms relocate under carbon constraints? How do clusters govern the transition when some firms stay and some leave? And why are the regional outcomes inside China so uneven? The argument, stated up front and then tested against cases: relocation is one of the mechanisms of green governance, not a failure of it. Clusters that build governance capability keep the upgrading activities and shed the compliance-heavy ones. Clusters that do not lose both. Whether that is good or bad depends on where you sit, and on what happens to the workers and suppliers left behind.

9.1 Drivers of Industrial Relocation: Cost, Carbon and Policy

The leakage literature often blurs a distinction worth keeping. There are two analytically separate relocation decisions. One is the classical cost-driven move: wages rise, land prices rise, environmental compliance adds a line item, and a labour-intensive process becomes cheaper somewhere else. The other is the carbon-driven move: a specific instrument—a carbon price, a border adjustment, a product standard—changes the relative cost of emissions across jurisdictions, and production follows the price signal. In practice the two are entangled. A furniture firm weighing Vietnam against Foshan is weighing wages, port costs, tariff exposure and the cost of complying with the EU Deforestation Regulation all at once. Teasing out the carbon component is hard, and most survey evidence cannot do it cleanly.

The carbon component is real and growing. Dechezleprêtre and Sato, surveying the environmental regulation and competitiveness literature, find that regulation's effects on relocation are small but detectable, concentrated in pollution-intensive sectors, and sensitive to the stringency differential between jurisdictions. Carbon does not by itself empty a cluster. At the margin it tilts decisions. For a firm already close to the threshold where a move makes sense, a carbon cost of two or three percent of total cost can be the difference.

The leakage framework makes the mechanism easy to see. When emission controls tighten in region A, a firm can adopt cleaner technology in A or relocate the dirty process to region B where controls are weaker. Babiker's estimate that relocation can raise aggregate emissions by up to 30 percent in some configurations is the extreme case; Felder and Rutherford show how stringent regulation in one region can suppress global demand and price for a product, shifting consumption—and emissions—elsewhere. Against this, Di Maria and van der Werf point to an induced-technology effect: regulation changes the incentive to innovate, and innovation can offset the leakage. The net sign is an empirical question, not a theoretical one.

For our purposes the important point is that the leakage framework treats the firm as a black box that either stays or goes. Real firms do neither. They reconfigure. The furniture case shows the reconfiguration most clearly.

A mid-sized Longjiang firm producing outdoor furniture for European buyers faces, as of the mid-2020s, a stack of requirements. The EU Timber Regulation and its successor, the EU Deforestation Regulation (EUDR), require geolocation data for the plot of origin of every timber input. The Carbon Border Adjustment Mechanism (CBAM), in its transitional phase, covers cement, iron and steel, aluminium, fertilisers, electricity and hydrogen, and casts a long shadow over any product with a steel or aluminium component. On top of these sit buyer-specific sustainability scorecards from retailers who have made public net-zero commitments. None of these is a carbon price on furniture as such. Together they function as one, because they raise the fixed cost of documentation and the variable cost of non-compliance.

The firm's response is a spatial split. Timber sourcing, design, finishing and the customer-facing documentation function stay in Foshan, close to the port, the testing laboratories and the buyers' sourcing offices. Cutting, sanding and assembly—the steps with the highest labour content and the least brand visibility—move to a leased shed in Binh Duong or Haiphong. The Vietnamese operation is not a pollution haven in the classic sense. It is a labour-cost hedge with a compliance discount attached. Foshan becomes more capital-intensive and more knowledge-intensive, which is the upgrading trajectory described in earlier chapters.

Is this leakage? In emissions terms, partly. The Vietnamese plant may draw on a grid with a higher emissions factor than Guangdong's, and the transport leg adds emissions the Foshan-only configuration did not have. In upgrading terms it is something else: a retention of the high-value activities in the regulated cluster and an export of the low-value ones. The cluster does not lose the firm. It loses the headcount.

Humphrey and Schmitz distinguish process upgrading (doing the same thing more efficiently), product upgrading (moving to more sophisticated products) and functional upgrading (taking on new functions in the chain, such as design or branding). What the Longjiang firm is doing is a spatial variant of functional upgrading: keeping the functions that generate rents and relocating the functions that generate only margins. The relocation is how the firm finances its compliance with green governance.

9.1.1 The cost calculus, itemised

Numbers help, even illustrative ones, because the public debate tends to treat relocation as a binary moral choice rather than a cost comparison. Table 9-1 sets out the main cost lines for a hypothetical mid-sized furniture exporter weighing a full move to Vietnam against a split configuration, holding output constant.

Table 9-1: Illustrative annual cost comparison for a furniture exporter (indexed, Foshan-only = 100)

The table is constructed, not measured; treat the magnitudes as illustrative. The structure of the result survives reasonable changes in the parameters. A full move captures the labour and compliance savings but pays for them in coordination costs, logistics and the loss of cluster externalities—the pattern-makers, the testing labs, the freight consolidators, the buyers' sourcing offices. The split configuration captures most of the labour saving while keeping the externalities, at the price of running two organisations. For many firms the split wins. That is why the observed pattern is bifurcation rather than exit.

The compliance line deserves comment. It falls in Vietnam not because Vietnamese environmental rules are absent—they are not—but because the documentation burden attached to European market access is lower when the exporting entity is not the one making the sustainability claims. The Vietnamese plant supplies the Foshan trading company, which holds the certifications and faces the auditors. Regulatory stringency has not been arbitraged away. It has been concentrated in the entity that can bear it.

9.1.2 When carbon dominates

Some products have high embodied emissions and thin margins, and there the carbon signal dominates the cost calculus. Cement is the textbook case, and the one Drake uses: with carbon pricing in the European Union, some cement manufacturers may move production to North Africa. Cement is a homogeneous commodity, transport costs are non-trivial but bounded, and the emissions intensity of production is high and hard to abate in the short run. A carbon price of EUR 80 per tonne applied to roughly 0.6 tonnes of CO2 per tonne of cement is a material share of the product price. The firm that can produce outside the price and ship in has an advantage.

Drake's analysis of border adjustments adds a twist that matters for policy. A symmetric border adjustment—a charge on imports from unregulated regions equal to the carbon price on domestic production—does not fully eliminate leakage. But it can induce the firm in the unregulated region to adopt cleaner technology than the firm in the regulated region, if the former has a cost advantage to begin with. Total emissions fall. The instrument that looks like protectionism turns out, under specified conditions, to be an environmental improvement. The same policy can be read as leakage prevention or as green industrial policy, and the reading depends on parameters that are rarely observed.

For Chinese manufacturing the steel and aluminium case matters more than cement, because steel and aluminium feed into the export products—appliances, furniture frames, machinery—that dominate the tradeable sector. CBAM's transitional phase, which began in October 2023, requires importers of steel and aluminium into the EU to report embedded emissions. The reporting obligation is already reshaping supplier selection, before any financial obligation attaches. A Chinese steel supplier that cannot produce verified emissions data is a liability to its downstream customers, whatever its price. The relocation response is not to move the steel mill. It is to move the downstream fabrication closer to the mill, or to switch to a mill that can document.

9.1.3 Policy as a relocation driver in its own right

Policy operates through channels that the cost calculus does not always capture. The 2016–2017 central environmental inspections are the clearest example. For firms in pollution-intensive sectors, compliance costs rose faster than in most competing jurisdictions. That was the intended effect, and it was also a relocation incentive; the two cannot be separated.

Predictability matters as much as stringency. A firm can adapt to a strict rule that is consistently applied. It struggles to adapt to a rule that is strict in one province and lax in the next, or strict this quarter and relaxed the next. The central inspections were effective precisely because they were unpredictable, and their unpredictability was itself a relocation driver: firms that could not forecast their compliance burden hedged by moving some capacity offshore. Predictability, not stringency, is often the binding constraint.

Then there is the emerging architecture of carbon markets and its interaction with trade policy. China's national emissions trading system, launched in 2021 and initially covering the power sector, is expanding to other sectors. The EU's CBAM will, from 2026, impose a financial obligation on imports. The interaction between a domestic carbon price and a foreign border adjustment is governed by the rules on deducting carbon paid at home, and those rules are still being negotiated. A firm that faces a domestic carbon price but cannot demonstrate it to a foreign importer pays twice. The relocation response is to site production where the carbon accounting is legible to the destination market—which may mean staying in China and investing in measurement, or moving to a jurisdiction with a mutual recognition arrangement.

The implication for the cluster is counter-intuitive. The firms most likely to relocate are not the dirtiest. They are the ones with the weakest measurement capability—the firms that cannot prove what they emit, and therefore cannot claim any deduction, any credit or any buyer preference. The relocation decision is driven less by the level of emissions than by the ability to account for them. Section 9.2 returns to this measurement gap, because it is where cluster governance earns its keep.

9.2 Cluster Governance for Green Upgrading

A cluster is not a firm. It is a population of firms, suppliers, institutions and informal arrangements that share a location and, to some degree, a fate. The governance question is how that population coordinates on a transition no single member can complete alone. The previous chapter argued that voluntary firm action is insufficient because of fixed-cost barriers, information asymmetries and free-rider problems. This section examines the institutional machinery—formal and informal—that clusters use to overcome those problems, and why some clusters succeed where others fail.

The cluster literature has long relied on a definitional contrast. Marshallian industrial districts, as Becattini described them, coordinate through what he called the socio-economic notion of the district: shared norms, trust, labour mobility and a dense web of small-firm interaction. Coordination is informal and largely spontaneous. The cluster-policy literature that emerged in the 1990s and 2000s—Altenburg and Meyer-Stamer on Latin America, Boekholt on the arm's-length public sector—emphasised deliberate institutional design: technology centres, export promotion agencies, supplier development programmes. Coordination through social embeddedness, against coordination through formal organisation.

Green upgrading under carbon constraints fits neither category cleanly. It needs both. The measurement infrastructure—shared laboratories, emissions factor databases, certification support—is a public good no small firm will provide and informal norms cannot generate. It requires formal institutional action. The diffusion of new practices through the cluster—adoption of a lower-carbon finishing process, a switch to a certified timber supplier—runs through the informal channels: labour mobility, supplier gossip, the pattern-maker who moves from one firm to another. Neither channel alone is sufficient.

9.2.1 What cluster governance actually does

In 2019 a provincial inspection in the Yangtze River Delta found that individual firms in a 120-firm electronics and precision-components park were failing to meet volatile organic compound (VOC) emission limits. The park operator, a state-owned development company, responded by building a shared solvent recovery facility. It collects spent solvent from tenant firms, distils it, and returns recovered solvent at a discount. The operator bore the capital cost; the operating cost is covered by the recovery margin and a per-tonne gate fee. A single firm generating 50 tonnes of spent solvent a year cannot justify a recovery plant. The 120 firms together generate 6,000 tonnes. The operator solved the collective action problem by acting as coordinating investor and recovers its cost through the gate fee. VOC emissions fell measurably, and tenant firms could now credibly claim compliance with the solvent-handling requirements their European and Japanese customers impose.

That facility is one function among several. A cluster of 200 furniture firms cannot each afford a gas chromatograph or a full-time carbon accountant; a shared testing laboratory funded by a consortium of firms or a local government spreads the fixed cost. Guangdong's furniture clusters have used a variant of this model, with provincial testing centres offering subsidised emissions and formaldehyde testing to small firms. The subsidy is the governance intervention; the laboratory is the technology.

Standards translation is a second function, and it is chronically underfunded. ISO 14067 for product carbon footprints, the GHG Protocol, the EUDR's geolocation requirements—all are written for large firms with dedicated compliance departments. Translating them into a form a 60-person workshop can use is a collective action problem. Industry associations do some of this work; local standards bodies do more. The clusters that do it well have a dedicated translation function, often housed in a trade association or a technology park.

Supplier development follows. A large firm facing a carbon requirement from its buyers pushes the requirement down to its suppliers, who are often smaller and less capable. The large firm can provide help, or the cluster can provide it collectively. The collective route avoids the hold-up problem: a supplier that invests in compliance for one buyer can serve others, and the cluster-level investment is not captured by a single firm.

Infrastructure is the most visible function. The Hawassa Industrial Park in Ethiopia, built by a Chinese contractor in nine months and devoted to apparel and textiles, includes a zero-emission water treatment plant. Park-level infrastructure is what allows individual tenant firms to meet the sustainability requirements of global brands such as Calvin Klein and Tommy Hilfiger without each building its own plant. The Chinese firms that built and operated the park also transferred know-how to the Ethiopian factories.

The fifth function is adjustment support, and it is the one most often skipped. When a cluster restructures, some firms exit and some workers are displaced. The governance question is whether the cluster manages that adjustment or leaves it to the market. Clusters with retraining programmes, supplier finance and orderly exit mechanisms retain more of their capability through the transition. Clusters without them lose firms and skills together.

Table 9-2 maps these functions against the formal and informal channels through which they are delivered.

Table 9-2: Cluster governance functions and delivery channels

The table makes a point the leakage literature tends to miss. The binding constraint on green upgrading is not the stringency of the regulation. It is the presence or absence of these five functions. A cluster with strong governance can absorb a stringent regulation and turn it into an upgrading opportunity. A cluster without them will lose firms to relocation regardless of how lax its rules are, because the firms cannot comply with the rules of their export markets anyway.

9.2.2 Why the park model does not generalise

The solvent recovery facility worked, and it is worth asking why similar projects fail elsewhere. Three conditions had to hold. The park operator needed the balance sheet to make the upfront investment and the authority to impose the gate fee on tenants; not all park operators have either. The tenant base had to be dense enough in the relevant waste stream—a park with 120 firms in mixed industries may lack critical mass in any one stream. And the regulatory signal had to be credible enough that tenants would pay the gate fee rather than lobby against it. The provincial inspection made the status quo untenable. Without it, the collective action problem would have persisted.

Where any of the three conditions is missing, the park model stalls. This is the general lesson of formal cluster governance: an actor with authority and a balance sheet, a demand base dense enough to justify the investment, and a regulatory signal credible enough to make the status quo costly.

9.2.3 Informal institutions: norms, trust and the limits of voluntarism

The informal channel is harder to document and easier to romanticise. The Marshallian district runs on trust, and trust is a real economic asset. But trust has limits when the transition requires capital investment and technical knowledge the trusting parties do not possess.

The honest version of the informal story is this. Informal institutions are good at diffusing practices that are cheap to adopt and easy to observe. A pattern-maker who moves from a firm that has switched to a water-based adhesive to one that has not will carry the practice with him. A supplier who has learned to document timber origin for one buyer will offer the same documentation to others. That diffusion is real, and it is faster than any formal programme.

Informal institutions are bad at financing fixed costs. No amount of trust will build a solvent recovery plant or a testing laboratory. They are also bad at coordinating on standards, because standards require a decision about which of several possible specifications to adopt, and trust does not resolve distributional conflicts over specification. A cluster that relies only on informal coordination will diffuse the cheap practices and fail to make the expensive investments. The informal channel diffuses what the formal channel has built; the formal channel builds what the informal channel cannot finance. Clusters with both make the transition. Clusters with one stall at the point where the cheap practices are exhausted.

9.2.4 The free-rider problem in cluster-level green infrastructure

The collective action problem deserves a closer look, because it is the analytical core of why cluster governance matters. Consider a cluster of firms that would each benefit from a shared emissions measurement service. The service costs a fixed amount per year. Each firm's willingness to pay is less than the total cost, but the sum of willingness to pay exceeds it. The service is socially efficient and not privately provided.

The standard solutions are familiar, and each has a failure mode. A public subsidy depends on fiscal capacity and is vulnerable to political cycles. A mandatory membership fee requires an association with the authority to collect, which many clusters lack. A lead firm that internalises the benefit by requiring its suppliers to use the service works only where a lead firm exists and has the leverage to impose the requirement—the case in buyer-driven chains such as apparel and furniture, less so in fragmented supplier industries.

The Chinese experience adds a fourth solution: the local government as coordinating investor, with cost recovery through land value capture or park fees. This works where the local government has fiscal capacity and land value appreciation is large enough to fund the infrastructure. It works less well in lagging regions where land values are flat and fiscal capacity is thin. The regional differentiation examined in Section 9.3 is, in part, a differentiation in the capacity to solve this collective action problem.

9.3 Regional Differentiation and Policy Responses

The previous two sections treated "the cluster" as a unit. It is not. China's manufacturing clusters are spread across regions with vastly different fiscal capacity, industrial composition and institutional depth, and the carbon transition is widening the gaps rather than narrowing them.

The coastal clusters—the Pearl River Delta, the Yangtze River Delta, the Bohai Rim—concentrate the export-oriented, buyer-facing industries and the institutional infrastructure (testing labs, certification bodies, trade associations) that green governance requires. The inland clusters—in Henan, Sichuan, Chongqing, Anhui—concentrate the labour-intensive, cost-sensitive industries that relocated there from the coast in the 2010s to escape rising coastal wages. The two groups face the carbon transition from opposite starting points.

Coastal clusters have the governance capability. They also have the highest compliance costs, because they are the most exposed to European and North American market requirements. Their firms are the ones that must produce EUDR geolocation data and CBAM embedded-emissions reports. The compliance burden falls hardest on the clusters best able to bear it, which is why the coastal clusters are not collapsing under the carbon transition. They are upgrading.

Inland clusters have the cost advantage and the governance deficit. Their firms are less exposed to the most stringent export-market requirements, because they serve domestic and lower-stringency export markets. But they are exposed to China's domestic environmental regulation, which has tightened, and to the carbon reporting requirements spreading through domestic supply chains as lead firms impose them on suppliers. They have less capacity to meet those requirements. The result is divergence: coastal clusters upgrade; inland clusters either attract the relocated low-margin activities or lose them to Southeast Asia.

The divergence is not simply coastal versus inland. Guangdong has both the Shenzhen-Dongguan electronics cluster, with deep institutional capacity, and the northern Guangdong furniture and ceramics clusters, with much thinner capacity. Zhejiang has both the Hangzhou digital cluster and the Wenzhou footwear and apparel clusters. The relevant variable is not provincial GDP per capita but the density of the five governance functions described above.

9.3.1 The relocation cascade and its regional incidence

The relocation decisions described in Section 9.1 do not distribute evenly across regions. They cascade. A coastal firm that moves its labour-intensive steps to Vietnam frees up capacity in its coastal cluster for higher-value activities. The suppliers that served the moved steps lose orders. Some relocate to follow the firm; others move inland to serve the domestic market; others exit. The cascade runs from the coastal cluster to the inland cluster to Southeast Asia, and a cluster's position in the cascade determines its trajectory.

The furniture sector shows the cascade most clearly. The Longjiang and Lecong clusters in Foshan are the core. They have the testing labs, the export trading companies, the design studios and the buyers' sourcing offices. They are where the EUDR and CBAM compliance functions sit. The suppliers that cut and sand the timber are more dispersed: some in Foshan, some in northern Guangdong, some in Jiangxi and Guangxi. As coastal firms move assembly to Vietnam, the inland suppliers that served assembly lose orders. Some follow the assembly to Vietnam, a difficult move for a small Chinese supplier. Some consolidate into the domestic market. Some close.

The regional incidence of the cascade is therefore not a simple transfer of activity from coast to inland. It is a transfer from coast to offshore, with the inland suppliers caught in between. The inland cluster that hoped to inherit the coastal firms' assembly work finds that the work has gone to Vietnam instead. Its cost advantage is not enough to offset the governance deficit and the logistics penalty.

The carbon transition is thus a regional divergence mechanism. Coastal clusters have the governance capability to retain the upgrading activities and the compliance functions. Inland clusters do not, and they lose the activities to offshore locations that offer a better combination of cost and compliance. The inland cluster is not competing with the coastal cluster. It is competing with Vietnam, and it is losing on the dimension that matters most for the export market.

9.3.2 Fiscal capacity and the governance deficit

The governance deficit in the inland clusters is, at bottom, a fiscal problem. The five governance functions all cost money, and the money has to come from somewhere. Coastal clusters fund them through a combination of local government expenditure, industry association fees and lead-firm investment. Inland clusters have less of all three.

Local government expenditure is constrained by fiscal capacity, which is constrained by the tax base, which is constrained by industrial composition. A cluster of small furniture firms generates less tax revenue per hectare than a cluster of electronics firms, and it has less fiscal room to subsidise testing labs and retraining programmes. Industry associations in the inland clusters are weaker, with smaller memberships and less ability to collect fees. Lead firms are fewer, because the inland clusters are more fragmented.

The gap is self-reinforcing. The coastal cluster invests in governance, attracts higher-value activities and generates more tax revenue to invest in governance. The inland cluster cannot invest, loses activities and has less revenue. The carbon transition accelerates the gap because it raises the return to governance investment: a cluster with a testing lab can certify compliance and retain the export order; a cluster without one cannot.

Three regional policy levers bear on this, and the next chapter takes up the full policy architecture. The first is extending governance capability across regions. A coastal testing lab can serve inland firms remotely, and some do. The constraint is the cost of sample transport and the trust relationship; a certification body that has never visited an inland cluster is less likely to be used by its firms. The lever is to subsidise the extension of coastal governance infrastructure to inland clusters—not to build duplicate labs, but to extend the reach of existing ones.

The second is coordinating relocation. When a coastal firm moves assembly offshore, the inland suppliers that served it are stranded. A coordinated relocation—the firm moves assembly to an inland cluster rather than offshore, with governance support—could retain the activity. The obstacle is that the inland cluster's governance deficit makes the move less attractive than the offshore option. The lever is to bundle relocation with governance investment: the firm gets the cost advantage of the inland location plus the compliance support it needs.

The third is managing exit. Some inland clusters will not make the transition. Their firms will close, and their workers will need to move. An orderly exit—with retraining, supplier finance and redeployment support—retains more of the cluster's capability and shortens the adjustment. A chaotic exit destroys both. Fiscal capacity to manage the exit is, again, the binding constraint.

9.3.3 The regional dimension of the dual carbon goals

The dual carbon goals—peak emissions before 2030, carbon neutrality before 2060—are national targets, but their implementation is regional. Provincial emissions-reduction targets, local carbon market pilots and the energy-consumption dual control system all allocate the national target across regions, and the allocation is not uniform. Coastal provinces, with higher GDP per capita and more advanced industrial structures, have more capacity to meet stringent targets. Inland provinces, with energy-intensive industries and thinner fiscal capacity, have less.

The allocation has a relocation consequence that is easy to miss. A firm facing a stringent provincial target has an incentive to move production to a province with a looser target, even within China. The leakage the international literature describes can occur domestically. Provincial governments know this, and the allocation of targets is partly designed to prevent it—by setting targets ambitious enough to matter but not so ambitious that they drive industry out. The balance is delicate and not always achieved.

Domestic leakage has a further consequence for the cluster. A firm that relocates from a coastal province to an inland province to escape a stringent target takes its compliance capability with it, but not the cluster's governance infrastructure. The inland province gains the activity but not the capability. The coastal province loses the activity but keeps the infrastructure. The net effect on national emissions is ambiguous; the net effect on the cluster system is a further dispersion of capability away from the places that can use it.

9.4 What the Relocation Pattern Tells Us

Three claims can now be made with some confidence, and one caveat.

Carbon-driven relocation is real, and it is not primarily a flight from regulation to deregulation. It is a re-sorting of activities within firms and within value chains. The firms that relocate are not the dirtiest; they are the ones with the weakest measurement capability. The activities that relocate are not the highest-emitting; they are the lowest-value. The cluster that retains the measurement capability and the high-value activities upgrades. The cluster that does not loses both.

Cluster governance is the variable that determines which clusters upgrade and which do not. Shared measurement, standards translation, supplier development, infrastructure and adjustment support are the machinery of the transition. Clusters that have them can absorb stringent regulation. Clusters that do not will lose firms regardless of how lax their rules are, because the firms cannot comply with the rules of their export markets anyway.

Regional differentiation is widening, and the carbon transition is the mechanism. Coastal clusters have the governance capability and the fiscal capacity to invest in it. Inland clusters do not, and they are losing activities to offshore locations that offer a better combination of cost and compliance. The inland cluster is not competing with the coastal cluster. It is competing with Vietnam.

The caveat: the evidence is uneven. Firm-level relocation decisions are documented in case studies and surveys, but the aggregate magnitude of carbon-driven relocation is hard to measure, and the attribution to carbon versus other cost drivers is contested. The cluster governance functions are described in the literature, but their relative importance is not established. The regional divergence is visible in the aggregate data, but the causal mechanism is inferred rather than demonstrated. The reader who wants certainty will not find it here. The reader who wants a framework for interpreting the pattern will.

One further point runs against the intuition of the leakage literature, and it concerns the direction of the transition. The standard framework treats relocation as a loss: emissions move, the regulated region loses activity, and the global outcome is no better. The cluster perspective suggests a different reading. Relocation is how firms finance their compliance. The firm that moves its low-margin assembly offshore and keeps its high-margin design and documentation onshore is not evading regulation. It is generating the margin that pays for the regulation. The cluster that retains the design and documentation functions is not losing the firm. It is retaining the functions that generate the rents.

The policy implication follows. The goal should not be to prevent relocation but to shape it—to ensure that the functions generating rents and capability stay in the cluster, and that the functions that move are the ones the cluster can afford to lose. The instruments for shaping relocation are the cluster governance functions described in Section 9.2 and the regional policy levers described in Section 9.3. The next chapter sets out the full policy architecture that connects these instruments to the national targets and the international trade regime.

Consider, finally, the worker in the Longjiang workshop whose job moved to Haiphong. The firm kept its design studio, its testing lab and its buyer relationships in Foshan. It moved the sanding and assembly. The worker's skills are not obsolete—they are still needed, just not in Foshan. The cluster governance question, at its most concrete, is whether a retraining programme, a supplier finance scheme or a redeployment pathway connects that worker to the design studio, the testing lab or the logistics operation the cluster is building. The five functions are not abstractions. They are the difference between a transition that retains capability and one that destroys it.

The pattern is not uniform, and it is not stable. Coastal clusters are upgrading, and shedding headcount. Inland clusters are losing activities, and some are building governance capability from a low base. Offshore locations are gaining activities, and the compliance obligations that come with export-market access. The next five years will show which trajectory dominates. The workshops and testing labs and container manifests of the Pearl River Delta already show the direction: clusters with governance capability keep the upgrading activities, and clusters without it do not.