Chapter 9: Cluster Governance, Learning and Regional Upgrading under Carbon Constraints

The previous chapter followed the demand signal down the chain: a European buyer sets a product carbon footprint limit, and the requirement travels through sourcing offices, tier-one suppliers, and eventually to a coating line in a furniture factory in Houjie. That account treats the firm as the unit that must respond. But no firm responds alone. The factory in Houjie learned to measure coating consumption because a local supplier of water-based finishes had already set up a trial line two streets away, because the town's testing station had bought an ISO 17025-accredited emissions chamber, because a trade association had circulated a template for the buyer's questionnaire, and because a retired engineer from a state-owned coatings plant ran a consultancy that half the export factories in town had used at least once. None of those things appear on any firm's balance sheet. All of them shaped whether that factory could meet the buyer's limit at a cost it could survive.

This chapter takes the cluster as the unit of analysis and asks a simple question with a complicated answer: under carbon constraints, what makes some clusters able to convert a compliance shock into an upgrading step while others convert it into an exit? The answer runs through governance. Clusters are not automatic learning machines. They are institutional arrangements that can either lower the cost of collective green action or raise it, and the difference is visible in the specific rules, intermediaries, and shared assets that a locality has built up over decades. Where those arrangements are thin, a carbon constraint arrives as a pure cost. Where they are thick, it arrives as a coordination problem that the cluster has some capacity to solve.

I should flag the evidentiary limits at the outset. The cluster literature is rich in case studies and thin in controlled comparison. My own fieldwork in Guangdong and Zhejiang furniture and electronics clusters, conducted between 2019 and 2023, gives me forty-one usable firm interviews and thirteen interviews with association officials, testing laboratory managers, and local economic planners. That is enough to describe mechanisms and not enough to estimate their prevalence. Where I generalize, I am generalizing from a small number of cases and from the secondary literature, and I say so. The chapter is organized around three mechanisms: collective action and governance, learning and knowledge diffusion, and external linkages and regional upgrading. Each has a carbon-specific twist that the older cluster literature did not anticipate.

9.1 Cluster Governance and Collective Action

9.1.1 What the cluster actually provides

Start with a definition that will do work rather than decorate. A cluster, in the sense used here, is a geographically concentrated group of firms and supporting organizations in a related set of activities, bound together by traded and untraded interdependencies. The traded ones are straightforward: supplier relationships, subcontracting, labor pooling. The untraded ones are the interesting ones. They include shared knowledge about process, shared norms about quality and delivery, shared infrastructure such as testing labs and wastewater treatment, and shared expectations about how disputes get settled. Alfred Marshall described the phenomenon in 1890 as "industrial atmosphere," a phrase that has survived because nobody has improved on it. The atmosphere is real but it is not free. It is produced by institutions, and institutions decay.

The carbon constraint changes the economics of the untraded interdependencies in a specific way. Consider a furniture cluster facing a buyer requirement to disclose the formaldehyde and volatile organic compound content of finishes, and to document the energy source of drying operations. A single firm can meet this requirement by hiring a consultant and buying test equipment. That is expensive but feasible. Now consider that the requirement applies to the whole cluster because the buyer sources from fifteen factories in the same town. The fifteen factories each hire the same three consultants, each buy the same test equipment, each run the same trials, and each make the same mistakes. The duplication is pure waste. A cluster with a functioning association, a shared testing lab, and a norm of information exchange converts fifteen separate learning curves into one. A cluster without those things converts a solvable coordination problem into fifteen solvable but expensive private problems, and the marginal factories exit.

The constraint, then, is often a public-good problem in disguise. Cleaner process knowledge, emissions measurement methods, supplier qualification for green inputs, and the credibility of a cluster's environmental claims to buyers are all at least partially non-excludable within the locality. Standard economics predicts under-provision. The question for any given cluster is whether the local institutional set has found a way around the under-provision.

9.1.2 Formal institutions: associations, standards bodies, and local state

The formal layer of cluster governance in China has three components that matter for green upgrading: trade associations, local state agencies (particularly the economic and information technology bureaus and the environmental protection bureaus at town and district level), and the quasi-public service platforms that sit between them.

Trade associations in Chinese manufacturing clusters vary enormously in capacity. The strongest ones, typically in clusters with a long export history and a dominant product, employ full-time staff, run their own testing or design centers, and have a working relationship with the local bureau that amounts to co-governance. The Houjie furniture association, which I visited in 2021, is of this type. It had negotiated a group standard for low-emission coatings with the provincial market regulator, operated a shared spray booth that small factories could rent by the hour, and maintained a list of approved coating suppliers that its members used as a de facto qualification screen. None of this was mandated. It emerged because the association's leadership concluded, around 2017, that the export buyers were going to keep raising the bar and that the cluster's small factories would fail one by one unless someone organized the upgrade.

The weaker type is a letterhead association: an office, a part-time secretary, a website, and an annual dinner. Clusters of this type do not lack for firms or for export exposure. They lack the internal governance to act collectively, usually because the firms are too heterogeneous, too competitive with each other, or too dependent on a single large buyer who prefers to deal with suppliers individually. In these clusters, green requirements are mediated firm by firm, and the transaction costs are borne privately.

Local state agencies matter in a way that is easy to misread from outside. The environmental protection bureau is the enforcement arm, and its inspections are the proximate trigger for most compliance investment. But the economic and information technology bureau controls the industrial upgrading funds, the land allocations, and the "little giant" and "specialized and sophisticated" enterprise designations that carry subsidies and prestige. Green upgrading in a Chinese cluster is typically financed through a combination of environmental enforcement (the stick) and industrial policy (the carrot), and the two agencies do not always coordinate. I have sat in meetings where a factory manager described receiving a shutdown notice from one bureau and an upgrading subsidy application form from another in the same week. The cluster's effective governance capacity depends heavily on whether those two agencies have learned to sequence their instruments.

The quasi-public service platforms deserve more attention than they usually get. These are the product testing centers, the design institutes, the energy efficiency diagnostic centers, and the shared treatment facilities that are typically set up with a mix of local government capital and association management. They are the physical embodiment of the cluster's collective capability. A testing lab that can run a full VOC emissions panel on a coating sample in forty-eight hours at a price a small factory can afford changes the economics of compliance for every firm in the cluster. The same lab, if it is under-equipped or under-staffed, forces every firm to send samples to Shanghai or Guangzhou and wait two weeks, which for a factory on a tight delivery schedule is the difference between winning and losing an order.

9.1.3 Informal institutions: social capital and the limits of trust

The informal layer is harder to measure and does more work than the formal layer in most clusters I have seen. It consists of the norms of reciprocity, the reputation mechanisms, and the personal networks that allow a factory manager to call a competitor and ask how they solved a coating adhesion problem, or to borrow a technician for a day, or to warn a neighbor that an inspector is coming.

The standard reference here is Putnam's distinction between bonding and bridging social capital, and the cluster literature has used it well. Bonding capital—dense ties within a homogeneous group—helps with day-to-day coordination and mutual aid. Bridging capital—ties across different groups—helps with access to new information. Green upgrading requires both. A cluster rich in bonding capital but poor in bridging capital can coordinate a shared response to a known problem but struggles to learn about a solution that exists outside the locality. A cluster rich in bridging capital but poor in bonding capital can access new knowledge but cannot act on it collectively.

The carbon constraint has a specific effect on this balance. It raises the value of bridging capital, because the relevant knowledge—about carbon accounting methods, about low-carbon process routes, about the requirements of the EU's carbon border adjustment mechanism—is mostly external to the cluster. It also raises the cost of relying on bonding capital alone, because a cluster that only talks to itself will converge on a local optimum that may be far from the buyer's actual requirement. I watched this happen in a Zhejiang hardware cluster in 2022, where the dominant firms had collectively convinced themselves that the buyer's carbon footprint requirement was a negotiating tactic and would go away. It did not go away. The firms that had maintained ties to the buyer's technical staff, or to a university lab in Hangzhou, adapted in time. The firms that had relied on the local consensus did not.

There is a darker side to informal governance that the cluster literature often underplays. Dense social ties can enforce collusion as easily as cooperation. In clusters where a handful of large firms dominate the association and the local supply chain, the informal norms can serve to suppress information about compliance costs, to coordinate on the lowest common denominator of environmental performance, or to exclude new entrants who might raise standards. I have heard association officials in two different clusters describe, off the record, how a proposed group standard was watered down because the largest member threatened to withdraw. The threat was credible because the member's withdrawal would have taken most of the association's budget with it. This is governance failure, and it is endogenous to the cluster's structure, not an external shock.

9.1.4 Collective action problems specific to carbon

Three collective action problems recur across the clusters I have studied, and they are worth naming precisely because they are not the same as the classic pollution externality.

The first is the shared-infrastructure problem. Decarbonizing a cluster often requires investment in shared assets: a centralized heat supply, a common wastewater treatment plant with energy recovery, a shared photovoltaic installation on the industrial park's roofs, a common logistics platform to consolidate shipments and reduce empty running. These investments have high fixed costs, long payback periods, and benefits that are hard to exclude. The standard collective action problem applies with full force. What varies across clusters is whether there is an entity with the authority and the balance sheet to make the investment and the legitimacy to allocate the costs. In some clusters, the local government's industrial park development company plays this role. In others, the association does. In many, nobody does, and the investment does not happen.

The second is the verification problem. Buyers, regulators, and financiers all want assurance that a factory's environmental claims are true. Individual factories cannot credibly self-certify, and third-party certification is expensive. A cluster that has built a credible local verification capability—an accredited lab, a recognized audit protocol, a registry of verified firms—lowers the cost of assurance for every member. A cluster that has not forces each firm to buy assurance separately, and the smaller firms often cannot afford it. The verification problem is a public good with a twist: the credibility of the local verifier depends on its independence from the local firms, which means the cluster has to solve a second-order governance problem (how to credibly commit not to capture the verifier) before it can solve the first.

The third is the transition-cost problem. When a cluster shifts from one process route to another—say, from solvent-based to water-based coatings, or from coal-fired boilers to natural gas or electric heat—the transition imposes costs on specific firms and workers before the benefits arrive. The distribution of those costs is politically salient within the cluster. A cluster with mechanisms for sharing transition costs (a fund, a phased schedule, a retraining program) can move faster than one where the costs fall entirely on the early movers. The early movers are often the smaller, more flexible firms, and if they bear the full cost of the transition and then face competition from laggards who free-ride on the cluster's improved reputation, the incentive to move early disappears. This is a classic assurance game, and the clusters that have solved it have usually done so through a combination of a credible regulatory schedule (everyone knows the deadline) and a subsidy that lowers the cost of moving early.

9.1.5 A typology of cluster governance regimes

Table 9-1 lays out four governance regimes that I have observed, defined by the strength of the formal associational layer and the density of bridging ties to external knowledge sources. The typology is descriptive, not normative, and the boundaries are fuzzy in practice.

Table 9-1: Cluster governance regimes and green upgrading outcomes

The coordinated upgrading regime is the one the policy literature holds up as the ideal, and it does deliver. The cluster can set its own standards ahead of the regulator, invest in shared assets, and bring the marginal firms along. The risk is that the coordination becomes exclusive: the association and the large firms set a standard that the small firms cannot meet, and the small firms exit or become subcontractors to the large ones. That may be efficient, but it is not neutral for regional employment.

The isolated compliance regime is more common than the literature admits. The cluster has strong associations and capable local government, but its external ties are thin. It can execute a known compliance requirement very efficiently—the testing lab, the shared treatment plant, the group standard all work—but it struggles when the requirement changes in a direction the cluster has not anticipated. The Zhejiang hardware cluster I mentioned is of this type. Its firms were excellent at meeting the requirements they knew about and slow to see the ones they did not.

The fragmented adaptation regime is the mirror image. External ties are strong—often because a few large firms are deeply integrated into global value chains and have their own technical teams—but the associational layer is weak. Upgrading happens, but it happens firm by firm, and the cluster as a whole does not capture the learning. The gap between the leading firms and the rest widens. This is the regime I worry about most for regional inequality, because the leading firms' success is real and visible, and it masks the stagnation of the tail.

The dependent stagnation regime is the one that produces relocation. The cluster cannot coordinate and cannot access external knowledge, so when the carbon constraint tightens, the marginal firms exit and the mobile ones move to a location with lower compliance costs or better infrastructure. The cluster does not disappear overnight, but it loses its dynamic core.

The typology is useful because it directs attention to the two variables that policy can actually influence: associational capacity and external bridging. Neither is easy to change. Associational capacity is built over decades and depends on the structure of the local economy and the history of local government–business relations. External bridging can be built faster, through trade fairs, technical missions, university partnerships, and the deliberate placement of cluster firms in global buyer networks. But it is also easy to fake, and the fakes—memoranda of understanding, delegations that tour a foreign cluster and return with nothing—are common.

9.1.6 The local state as a governance actor

I have been treating the local state as one actor among several, but in the Chinese context it deserves separate treatment because its role is distinctive and because it is changing under the dual carbon goals.

The traditional model of local state involvement in clusters is developmental: the local government provides land, infrastructure, and subsidies, and in exchange expects employment and tax revenue. Environmental enforcement was, for a long time, subordinate to this developmental logic. The central government's environmental targets were mediated by local officials who had their own growth priorities, and the result was the familiar pattern of enforcement cycles: a campaign, a period of laxity, another campaign.

The dual carbon goals have changed this in ways that are still working through the system. The 2021 announcement of the 1+N policy framework, and the subsequent assignment of carbon peaking and carbon intensity targets to provincial and sub-provincial governments, has given local officials a new set of metrics on which they are evaluated. The metrics are not yet as hard as GDP growth was, and there is considerable variation in how seriously they are taken. But the direction is clear: local officials now have an incentive to demonstrate carbon performance, and that incentive is transmitted to the cluster through the industrial policy system.

The effect on cluster governance is ambiguous. On one hand, it gives the local state a reason to invest in the shared infrastructure and verification capacity that collective action requires. On the other hand, it can produce a new round of campaign-style enforcement, in which factories are ordered to meet targets on a timetable that does not match their investment cycles, and the associational layer is bypassed rather than strengthened. I have seen both. The clusters that have done well are the ones where the local bureau has used its planning authority to set a credible medium-term schedule and then let the association and the market work out the allocation. The clusters that have done badly are the ones where the bureau has treated the carbon target as a quota to be met by administrative order.

There is a further complication. The local state's fiscal capacity varies enormously, and the clusters that most need public investment in shared green infrastructure are often the ones with the least ability to fund it. A wealthy coastal district can build a district heating network or a shared testing center out of its own budget. A poorer inland cluster cannot, and depends on provincial or central transfers that are allocated by formula rather than by need. This is a structural constraint on cluster governance that no amount of local initiative can fully overcome, and it is one reason why the regional distribution of green upgrading is likely to be uneven.

9.1.7 What the cluster does not do

I want to be explicit about the limits of the cluster as a governance unit, because the literature has a tendency to overclaim.

Clusters do not solve the problem of firm-level investment in process change. A shared testing lab can tell a factory whether its coating meets the standard; it cannot pay for the new spray line. The investment decision remains with the firm, and the cluster's governance affects the cost and the risk of that decision, not its necessity.

Clusters do not substitute for the buyer's requirement. The buyer sets the standard, and the cluster responds. A cluster with excellent governance can respond faster and cheaper, but it cannot change what the buyer wants. This is why the external linkages discussed in the third section of this chapter matter as much as the internal governance.

Clusters do not automatically distribute the gains from upgrading. The coordinated upgrading regime can produce a cluster that is globally competitive and locally unequal, with a core of upgraded firms and a periphery of subcontractors who bear the compliance costs without capturing the price premium. Whether that is acceptable is a political question, not an economic one, but it is worth flagging because the policy discourse around clusters tends to assume that cluster success is regional success.

9.2 Learning, Innovation and Knowledge Diffusion

9.2.1 The learning mechanism the cluster literature describes

The classic account of cluster learning runs roughly as follows. Proximity lowers the cost of interaction, so firms in a cluster interact more than firms outside it. Frequent interaction, combined with labor mobility and the informal exchange of process knowledge, produces a local knowledge pool that no single firm owns. Firms draw from the pool and contribute to it. The pool grows over time, and the cluster's rate of innovation exceeds what the same firms would achieve in isolation. Marshall's industrial atmosphere, Becattini's industrial district, and the more recent work on regional innovation systems all describe variants of this mechanism.

The account is plausible and, in my experience, partly true. Factory managers in Houjie do know each other's processes in detail. Technicians move between firms. A problem that one firm solves becomes common knowledge within a season. But the account is also too smooth. It describes learning as a diffusion process, as if knowledge were a fluid that flows from where it is concentrated to where it is not. In practice, learning in clusters is lumpy, contested, and often blocked. The carbon constraint makes the blockages visible.

9.2.2 Three types of green knowledge and their different diffusion patterns

Green knowledge in a manufacturing cluster is not a single thing. It comes in at least three types, and each diffuses differently.

The first is codified technical knowledge: the specification of a low-VOC coating, the operating parameters of a new drying oven, the calculation method for a product carbon footprint. This type diffuses relatively easily, because it can be written down, taught, and purchased. Its diffusion is limited mainly by cost and by the absorptive capacity of the receiving firm. A cluster with a good testing lab and a training program can diffuse codified knowledge quickly. A cluster without them diffuses it slowly, through consultants and equipment vendors.

The second is tacit process knowledge: the feel of a coating line, the judgment about when a batch is ready, the diagnosis of a defect that the manual does not cover. This type diffuses through people, not documents. It moves with technicians who change employers, with the informal conversations at the association's monthly meeting, with the retired engineer who consults for half the town. Its diffusion is limited by the density of social ties and by the willingness of the knowledgeable to share. In clusters with high labor mobility and strong bonding capital, tacit knowledge moves fast. In clusters where the leading firms hoard their technicians and the association is a letterhead, it does not.

The third is organizational knowledge: how to set up a carbon accounting system, how to manage a supplier qualification process, how to integrate environmental data into production planning. This type is the hardest to diffuse, because it is embedded in routines and requires changes to how the firm is managed, not just what it does. It diffuses mainly through demonstration and imitation, and it requires a certain level of managerial absorptive capacity in the receiving firm. A cluster can have excellent codified and tacit knowledge diffusion and still fail to diffuse organizational knowledge, because the firms that need it most are the ones with the least capacity to absorb it.

The distinction matters for policy. A cluster that invests in a testing lab is investing in codified knowledge diffusion. A cluster that invests in apprenticeship programs and technician exchange is investing in tacit knowledge diffusion. A cluster that invests in management training and peer learning among factory managers is investing in organizational knowledge diffusion. The three are not substitutes, and the third is the one most often neglected.

9.2.3 Absorptive capacity and the two-tier cluster

The concept of absorptive capacity, introduced by Cohen and Levinthal in 1990, is central here. A firm's ability to learn from external knowledge depends on its prior stock of related knowledge and on its investment in research and development. Firms with low absorptive capacity cannot benefit from proximity to knowledge, because they cannot recognize, assimilate, or apply it. This produces a two-tier cluster: a group of firms with high absorptive capacity that learn from each other and from external sources, and a larger group with low absorptive capacity that is physically proximate but cognitively distant.

The carbon constraint widens the gap. The firms with high absorptive capacity can engage with the buyer's technical team, understand the carbon footprint methodology, and redesign their processes accordingly. The firms with low absorptive capacity receive the requirement as a document they cannot interpret and a cost they cannot avoid. The cluster's knowledge pool, which the classic account treats as a shared asset, is in practice accessible only to the upper tier.

I have seen this play out in the furniture cluster's response to the EU's deforestation regulation and the associated due diligence requirements on wood sourcing. The large exporters had staff who could read the regulation, trace their supply chains, and build the documentation system. The small factories relied on their buyers to tell them what to do, and when the buyers' requirements changed, they were caught flat-footed. The cluster's shared knowledge did not help them, because they lacked the absorptive capacity to use it.

The policy implication is uncomfortable. Investing in cluster-level knowledge infrastructure—labs, training centers, demonstration projects—helps the firms that are already capable of using them. It does little for the tail. Reaching the tail requires a different approach: hands-on diagnostic services, subsidized consultancy, and in some cases a public takeover of the compliance function for the smallest firms. The clusters that have done best at reaching the tail have used the association's staff as a roaming diagnostic team, visiting small factories and helping them identify the cheapest compliance path. That is labor-intensive and does not scale easily, but it is what the problem requires.

9.2.4 Learning by doing, learning by using, and the role of demonstration

Two additional learning mechanisms deserve attention because they are particularly important for green process change.

The first is learning by doing. A factory that installs a new low-emission coating line learns how to operate it only by operating it. The learning curve is real, and it is steep in the first months. The problem is that the learning is firm-specific and the early months are expensive. A cluster can lower the cost of learning by doing if it can arrange for the first firms to share their operating experience with the later ones, or if it can arrange for a shared demonstration facility where firms can experiment without committing to a full installation. The shared spray booth in Houjie is exactly this: a small factory can run its product through the booth, learn what the process requires, and then decide whether to invest. The booth is not a subsidy; it is a learning infrastructure.

The second is learning by using. A factory that adopts a new energy management system or a carbon accounting platform learns about its own operations in the process of using the system. The data the system produces reveals inefficiencies that were invisible before. This is a form of learning that the firm did not anticipate and that the cluster can amplify by creating a forum where firms compare their data and discuss what it shows. The Houjie association's quarterly energy benchmarking meeting, which began in 2021 with eleven member firms and now draws about thirty, has factory managers submit their energy consumption per cubic meter of finished product on a standard template and then discuss the outliers. The managers I interviewed in 2023 said the peer comparison was the main reason they kept attending; one described discovering that his drying kiln was running at 18 percent higher energy intensity than a competitor's identical kiln, a gap he traced to a misadjusted damper. That is learning by using, amplified by a cluster forum.

Demonstration projects are the policy instrument that corresponds to these mechanisms. A demonstration project installs a new technology in a real factory, documents the results, and opens the factory to visitors. Its effectiveness depends on three things: the demonstration must be in a firm that is similar enough to the target firms that they can imagine themselves doing it; the results must be documented in a form the target firms can use; and the visiting must be organized so that the target firms actually learn, not just tour. The last is the hardest. I have been on demonstration tours that were essentially public relations exercises, and on others where the factory manager spent two hours walking a group of peers through the problems he had encountered and how he had solved them. The difference was not the technology. It was the willingness of the demonstrator to talk about failure.

9.2.5 The role of universities and research institutes

The cluster literature has a lot to say about the role of universities in high-technology clusters and rather less about their role in traditional manufacturing clusters. The Chinese experience suggests that the relevant institutions are not always universities in the conventional sense. They include:

• Provincial research institutes that specialize in the cluster's core technology (coatings, textiles, ceramics, electronics).
• The technical universities that have opened branch campuses or technology transfer offices in industrial districts.
• The testing and certification bodies that combine a commercial service with a research capability.
• The equipment vendors that provide process knowledge along with their machines.

The last is often the most important and the least studied. A coating line vendor does not just sell equipment; it sells the process knowledge that makes the equipment work. The vendor's application engineers are, in effect, a mobile knowledge-diffusion mechanism. They visit multiple factories, see multiple problems, and carry solutions from one to another. The cluster's learning is partly the vendor's learning. This has implications for policy: a cluster that attracts the leading equipment vendors, for example by hosting their demonstration lines or their service centers, gains access to a knowledge channel that is otherwise hard to build.

9.2.6 The dark side of cluster learning

I have already mentioned the risk that dense ties produce collusion. There is a second risk that is specific to green upgrading: the risk of lock-in to a local standard that is not the global standard.

Clusters often develop their own technical conventions. A group standard, a local testing method, a shared understanding of what "low emission" means—these are useful because they reduce transaction costs within the cluster. But they can diverge from the standards that buyers and regulators use. When the divergence is small, it is a nuisance. When it is large, it is a trap: the cluster has invested in a standard that the market does not recognize, and the investment has to be written off.

The EU's product environmental footprint framework and the various carbon accounting standards (the GHG Protocol, ISO 14067, the PAS 2050) are the relevant global references. A cluster whose local standard is not aligned with them will find that its firms have to redo their measurements when they sell to European buyers. The clusters that have avoided this trap are the ones that brought the global standard into the local standard-setting process from the beginning, usually by involving the buyer's technical staff or an internationally accredited verifier in the process. That is a governance choice, and it is one that the association and the local bureau can make if they are paying attention.

9.3 Regional Upgrading and External Linkages

9.3.1 The cluster is not a closed system

The first two sections treated the cluster as if its governance and learning could be understood from the inside. That is a useful simplification and a dangerous one. Clusters are embedded in wider networks, and the direction and quality of their external linkages shape what they can learn and what they can upgrade to.

The global value chain literature, from Gereffi's early work through Humphrey and Schmitz's work on upgrading in clusters, has established the basic point: insertion into a global value chain exposes a cluster to the chain's lead firms, and the lead firms' governance determines the upgrading opportunities available. A cluster inserted into a buyer-driven chain as a contract manufacturer has different opportunities from a cluster inserted as a full-package supplier, and different again from a cluster that has its own brands. The carbon constraint adds a new dimension to this, because the lead firms' carbon requirements are now part of the governance.

Humphrey and Schmitz distinguished four types of upgrading: process upgrading (doing the same thing more efficiently), product upgrading (moving to more sophisticated products), functional upgrading (moving to higher-value functions such as design or branding), and inter-sectoral upgrading (moving into new sectors). The carbon constraint interacts with all four, but differently.

Process upgrading is the most direct: a factory that reduces its energy consumption or switches to a cleaner input is doing process upgrading, and the carbon requirement is the trigger. This is where most of the action in Chinese clusters has been so far, and it is where the cluster's shared infrastructure and knowledge diffusion matter most.

Product upgrading is more interesting. A carbon constraint can create a market for products that are certified low-carbon, and a cluster that can produce them at scale gains an advantage. The furniture cluster's move into water-based finishes is an example: the product is not fundamentally different, but the certification opens a market segment that is willing to pay a premium. The premium is the "green premium" discussed in Chapter 6, and its size determines whether product upgrading is worth the investment.

Functional upgrading is the hardest and the most consequential. It means moving from manufacturing to design, branding, or chain coordination. The carbon constraint can help or hinder. It can help if the cluster develops a capability in carbon accounting or life-cycle assessment that becomes a service it sells to other firms. It can hinder if the lead firm uses the carbon requirement as a reason to tighten its control over the supplier, for example by requiring the supplier to use the lead firm's own carbon platform and report into its system. The second is more common than the first, and it is a reminder that upgrading is not automatic and that the lead firm's interests shape the outcome.

Inter-sectoral upgrading is the least discussed and potentially the most important for regions facing the decline of a traditional industry. A cluster that has built capabilities in coatings, testing, and process control for furniture can redeploy those capabilities into other sectors—building materials, automotive interiors, electronics enclosures. The carbon constraint can accelerate this if the new sectors have tighter environmental requirements and are willing to pay for the cluster's accumulated knowledge. It can also block it if the cluster's assets are too specific to the old sector. The relevant concept is asset specificity, and it is the reason why some clusters diversify successfully and others do not.

9.3.2 The direction of linkages: upward, downward, and sideways

Three directions of external linkage matter for the argument.

Upward linkages are to buyers, lead firms, and final markets. They are the most studied and the most important for upgrading, because the buyer's requirements are the proximate driver of most green investment. The quality of the upward linkage matters: a cluster that sells through trading companies has a thinner linkage than one that sells directly to the buyer's technical team, and a cluster that sells to a buyer with a sophisticated carbon strategy has a thicker linkage than one that sells to a buyer who is only beginning to ask questions.

Downward linkages are to suppliers, subcontractors, and the local supply base. They matter for the carbon constraint because a large share of a product's footprint is in the supply chain. A cluster that can bring its suppliers along—by training them, by qualifying them, by helping them measure—can reduce its product footprint faster than one that cannot. This is the cluster-level analogue of the supply chain governance discussed in Chapter 8, and it is where the cluster's collective institutions have a comparative advantage. A single factory can pressure its suppliers, but an association can set a supplier standard that applies to all its members and that suppliers have to meet to stay in the cluster.

Sideways linkages are to other clusters, to research institutions, and to peers in other regions. They are the least studied and the most important for learning that is not available from the buyer. A cluster that only looks up the chain learns what the buyer wants; a cluster that also looks sideways learns what other clusters have done about similar problems. The Chinese experience with cluster-to-cluster learning is instructive. The provincial governments have organized exchange programs between clusters in different cities, and the associations have organized visits to counterparts in other provinces. The effectiveness varies, but the principle is sound: the knowledge a cluster needs is often in another cluster, not in a university or a buyer.

9.3.3 The role of the diaspora and the returnee

One external linkage that deserves specific mention is the role of people. The Chinese clusters that have upgraded fastest often have a diaspora of engineers and managers who worked in foreign firms, studied abroad, or managed export operations, and who returned to the cluster with knowledge and contacts. The returnee phenomenon is well documented in the high-technology clusters of Shenzhen and Suzhou; it is less documented in traditional manufacturing clusters, but it is real. In Houjie, several of the factories that moved earliest to water-based coatings were run by managers who had worked for foreign coatings companies or for European buyers. They knew the standard before it arrived, because they had seen it in their previous jobs.

The diaspora is a form of bridging social capital that is built through careers rather than through associations. Policy can support it—through returnee programs, through support for overseas study, through the maintenance of alumni networks—but it cannot manufacture it. The clusters that have it are lucky, and the clusters that do not have it have to build their external linkages through other means.

9.3.4 Relocation as an external linkage

The relocation dynamics discussed in Chapter 5 for the furniture industry have a cluster dimension that is worth spelling out. When a cluster's cost structure rises, some firms relocate to lower-cost regions or countries. The conventional view treats relocation as a failure of the cluster. The more accurate view is that relocation is a form of external linkage, and its effects on the originating cluster depend on what relocates and what stays.

If the relocating firms are the low-value-added, high-compliance-cost firms, and the firms that stay are the higher-value-added ones, relocation can be part of an upgrading trajectory. The cluster's employment falls, but its value added per worker rises. If the relocating firms are the dynamic ones and the firms that stay are the laggards, relocation is a hollowing-out. The difference depends on the cluster's governance and on the policy environment: a cluster that has invested in shared infrastructure and knowledge can retain the higher-value activities even as the lower-value ones move.

There is a further twist specific to carbon. Relocation can be a form of carbon leakage: the emissions move with the production, and the global total does not fall. The EU's carbon border adjustment mechanism is designed in part to address this by imposing a carbon cost on imports from countries with weaker carbon pricing. For Chinese clusters, the mechanism creates a new reason to keep production at home and decarbonize it: if the exported product carries a carbon cost at the border anyway, the cost advantage of relocating to a region with weaker carbon policy is reduced. Whether this actually slows relocation is an empirical question that the data cannot yet answer. My interviews in 2023 suggested that some furniture firms were reconsidering planned relocations to Southeast Asia because the buyers were asking about the carbon content of the relocated production and the answers were not favorable. That is anecdotal, and it may not survive a change in the policy environment.

9.3.5 Regional upgrading and the question of who captures the value

The final issue in this section is distributional. Regional upgrading is often described as if the region were a single actor that either upgrades or does not. In practice, upgrading redistributes value within the region. The firms that move up the chain capture more value; the firms that remain subcontractors capture less. The workers who have the skills for the new processes gain; the workers who do not lose. The local government's tax base may grow even as employment falls.

The carbon constraint intensifies these distributional effects because it raises the return to capability. A firm with the absorptive capacity to meet the buyer's requirement can turn compliance into a competitive advantage. A firm without it faces a cost it cannot pass on. The cluster's governance determines whether the gap between the two isbridged or widened. A cluster that invests in reaching the tail—through diagnostic services, subsidized training, shared facilities—can narrow the gap. A cluster that focuses only on the leading firms will widen it. The policy question is not whether the cluster upgrades but who upgrades with it.

9.3.6 The cluster as a node in a multi-scalar system

There is a temptation, in the cluster literature and in regional policy, to treat the cluster as the relevant unit and to stop there. The past two decades of research on global production networks suggest a more complicated picture. The cluster is a node in a system that runs from the town industrial park to the provincial planning commission to the national ministry to the international standard-setter and the global buyer. The cluster's governance capacity is shaped by all of those levels, and the carbon constraint is being imposed simultaneously at several of them.

The national level sets the dual carbon targets and the 1+N policy framework. The provincial level translates those into targets for prefectures and allocates the funds for industrial upgrading. The municipal and district level operates the industrial parks, the subsidy programs, and the enforcement. The cluster level contains the associations, the service platforms, and the informal networks. And the international level contains the buyers, the standard-setters, and the trade regime. A cluster that is well governed at its own level but poorly connected to the levels above it will struggle, because the resources and the authority it needs are not local.

This multi-scalar character explains a pattern that is otherwise puzzling: two clusters with similar internal governance can have very different green upgrading outcomes because one is in a province that has made carbon performance a priority and the other is not. The provincial political economy matters. Guangdong and Zhejiang, with their export-oriented economies and their exposure to buyer requirements, have moved faster than provinces whose clusters are oriented to the domestic market. That is not a difference in cluster governance narrowly defined; it is a difference in the wider system in which the cluster sits.

9.3.7 What the cluster cannot do about the border

The external linkage that has become most consequential since 2023 is the border. The EU's carbon border adjustment mechanism, phased in from October 2023 with reporting obligations and a financial obligation from 2026, imposes a carbon cost on imports of certain goods—initially cement, iron and steel, aluminium, fertilizers, electricity, and hydrogen. Furniture is not on the initial list. But the mechanism's logic, and the precedent it sets, matter for every export cluster. The EU has signaled that the scope may expand, and other jurisdictions (the UK, Canada, and the US Congress) have proposals of their own.

A cluster cannot negotiate with the border. It can only prepare. The preparation takes two forms. The first is measurement: a cluster whose firms can document their embedded emissions in a form the EU recognizes will face a lower compliance cost than one whose firms cannot. This is where the shared verification infrastructure discussed earlier earns its keep. The second is abatement: a cluster that has already reduced its emissions faces a smaller border adjustment than one that has not. The two are linked, because measurement is the precondition for abatement—you cannot reduce what you do not measure.

The clusters that are preparing are the ones whose associations and local governments have read the regulation and understood that it applies to their exports even if their product is not on the initial list, because their buyers will pass the requirement down. The clusters that are not preparing are the ones that have concluded, wrongly, that the border is a European problem and not a Chinese one.

9.3.8 The learning that matters is organizational, not technological

The binding constraint on green upgrading in most Chinese clusters is not the availability of technology. The equipment exists, the suppliers exist, the process knowledge exists. The binding constraint is organizational: whether the firms, the association, and the local bureaus can agree on who pays for the shared lab, who sets the group standard, and who absorbs the cost when the first movers' investment does not pay back on schedule. I have watched that coordination fail more often than it has succeeded, and the failure is rarely dramatic. It looks like a meeting postponed twice, a lab fee raised above what small firms will pay, a standard left unaligned with the buyer's method.

This is why the cluster matters. Organizational capability is not built by individual firms in isolation. It is built through the repeated interactions that a cluster makes possible: the shared testing lab, the association's working groups, the informal conversations between factory managers, the demonstration projects, the supplier qualification schemes. The Houjie spray booth is a case in point. It rents for 180 yuan an hour, roughly a quarter of what a commercial booth in Dongguan charges, because the association owns the equipment and the district covered the capital cost in 2019. In 2022 the booth logged just over 4,100 rental hours across 67 firms, most of them with fewer than 50 employees. Those hours are the mechanism. A small factory that cannot justify a 400,000-yuan booth can still test a water-based coating on its own product, discover that the finish fails a adhesion test at the third coat, and adjust before it commits capital.

The implication for the book's argument is direct. The green governance capability I described in Chapter 3 as an endogenous driver of upgrading is not only a firm-level attribute. It is also a cluster-level attribute, and the two are related. A firm's governance capability is partly built from the resources the cluster provides: the knowledge it can access, the verification it can rely on, the suppliers it can qualify, the peers it can learn from. A cluster's governance capability is partly built from the capabilities of its firms. The two co-evolve, and the co-evolution is path-dependent. A cluster that missed the first round of green upgrading finds the second round harder, because the capabilities it needed were not built.

9.3.9 Three clusters, three trajectories

To make the argument concrete, I want to compare three clusters I have studied, anonymized in the way that field research requires but described accurately in their essentials.

The first is the Houjie furniture cluster in Dongguan, Guangdong. It is the largest furniture export cluster in China, with several thousand firms and a deep supply base in coatings, hardware, and foam. Its governance is relatively strong: an active association, a testing center, a shared spray booth, and a local government that has used the industrial upgrading funds to support process change. Its external linkages are strong: direct relationships with European and American buyers, a diaspora of managers who worked for foreign firms, and a presence at the major trade fairs. Its trajectory under the carbon constraint has been process upgrading, with a growing number of firms certified to the buyer's low-emission standards, and the beginnings of product upgrading into low-carbon product lines. The cluster's weakness is its tail: the small factories that cannot afford the transition and that the association's diagnostic services reach only partially. Of the 67 firms that used the spray booth in 2022, 41 had fewer than 30 employees; the association's two roaming diagnosticians can visit perhaps 15 of them in a year.

The second is a hardware cluster in Zhejiang, which I will call Yueqiao. It is smaller and more specialized, with roughly 280 firms making a narrow range of stainless-steel fittings for plumbing and gas lines. Its governance is strong in the formal sense: an association with four full-time staff, a group standard for nickel leaching published in 2020, and a shared testing lab that opened in 2018 with 3.2 million yuan of district and provincial money. Its external linkages are thin. About 70 percent of its exports move through trading companies in Ningbo and Yiwu, and only a handful of firms have met the European buyers' technical staff face to face. Its trajectory has been isolated compliance. In March 2022 a German buyer sent a supplier questionnaire asking for product carbon footprint data per kilogram of fitting, calculated to ISO 14067. The association circulated the questionnaire, but no member had the methodology. The lab could test nickel leaching but could not run a life-cycle assessment. The cluster spent eight months and, by the association secretary's estimate, about 600,000 yuan on a consultant from Hangzhou before the first firm could answer the questionnaire credibly. The cluster's strength is its ability to execute a known requirement; its weakness is its ability to anticipate an unknown one.

The third is a cluster in an inland province that I visited in 2022, which I will call Xiping. It makes ceramic sanitaryware and has a long history and substantial export volume. Its governance is fragmented. The association is a letterhead with a part-time secretary. The testing lab, built in 2016 with a provincial grant, has a bomb calorimeter that has been out of calibration since 2020 because the county cannot fund the service contract. The environmental bureau and the industrial bureau report to different vice-mayors and have not jointly sequenced an enforcement and subsidy schedule. Its external linkages are dominated by a single large buyer that sources from six of the largest firms and deals with them individually. Its trajectory has been dependent stagnation: the six leading firms are upgrading with the buyer's technical support, while the tail waits. Between 2021 and 2023, at least fourteen firms in the cluster relocated assembly to Vietnam, according to the county's own industrial survey. The cluster is not collapsing, but its dynamic core is shrinking.

The comparison is not a controlled experiment. The three clusters differ in size, product, region, and history, and I cannot isolate the effect of governance from the effect of the other variables. But the pattern is consistent with the argument: the cluster with strong governance and strong external linkages is upgrading; the cluster with strong governance and weak external linkages is complying but not anticipating; the cluster with weak governance and concentrated external linkages is stagnating. The variables that the chapter has identified—associational capacity, bridging ties, verification infrastructure, the coordination of the local state—are visible in all three.

9.3.10 The policy implications, stated carefully

I am wary of drawing policy conclusions from a small number of cases, and I want to state the implications with the appropriate caveats. Three seem robust enough to state.

The cluster-level public goods—testing labs, demonstration facilities, shared treatment and energy infrastructure, diagnostic services for small firms—are the highest-return investments a local government can make in green upgrading. They are also the ones most likely to be underfunded, because they do not produce a visible ribbon-cutting and because their benefits are diffuse. The Houjie booth cost the district 400,000 yuan in capital and returns perhaps 740,000 yuan a year in rental revenue at the 2022 rate; the arithmetic is not the point. The point is that 67 firms got access to a capability none of them could have bought alone. The clusters that have invested in such assets are the ones that have made the transition look easy, which is precisely why the investment is easy to overlook.

External linkages matter as much as internal governance, and they can be built faster. Trade fairs, technical missions, buyer technical exchanges, and cluster-to-cluster learning programs are all ways to build bridging capital. The clusters that have done this well have treated the buyer's technical staff as a learning resource, not just a source of orders. Yueqiao's eight-month struggle with the ISO 14067 questionnaire is what a thin upward linkage costs: the cluster had the lab and the standard but not the method, and the method lived with the buyer.

The tail is the hard problem and the one that policy most often neglects. The leading firms can upgrade on their own. The marginal firms cannot, and if they are left to exit, the cluster's employment and its social fabric change. Reaching the tail requires labor-intensive, unglamorous work: diagnostic visits, subsidized consultancy, shared facilities that small firms can afford, and a regulatory schedule that gives them time to adjust. Xiping's fourteen relocations in two years are what happens when that work is not done.

9.3.11 What this chapter does not argue

I want to be explicit about the limits of the argument, because the cluster literature has a tendency to overclaim and I do not want to repeat it.

This chapter does not argue that clusters are necessary for green upgrading. Some firms upgrade without a cluster, and some clusters fail to upgrade. The cluster is a mechanism, not a guarantee.

It does not argue that cluster governance can substitute for firm-level investment or for the buyer's requirement. The cluster lowers the cost and the risk of upgrading; it does not eliminate the need for either.

It does not argue that the Chinese cluster model is generalizable. The Chinese clusters have specific features—the role of the local state, the density of the supply base, the scale of the export orientation—that are not present everywhere. The mechanisms I have described may operate differently in clusters in Vietnam, India, or Mexico, and I have not tested them there.

It does not argue that the trajectory of any particular cluster is determined. The three clusters I compared could change course. A change in the buyer's requirements, a new provincial policy, a new association leadership, or a large firm's decision to invest in the local supply base could shift any of them. The path dependence is real but it is not destiny.

The factory in Houjie that measures its coating consumption, which I introduced at the end of the previous chapter, is a case in point. Its measurement capability is its own, built up over years of trial and error. But the standard it measures against, the lab that verifies its results, the supplier that provides the low-emission coating, and the association that circulated the buyer's questionnaire all came from the cluster. The firm's strategy and the cluster's ecosystem are two halves of the same story.

There is one detail from that factory I have not been able to resolve. Its manager showed me, in October 2023, a spreadsheet tracking coating consumption per square meter across four product lines. Three of the lines had converged to within 6 percent of each other over eighteen months. The fourth, a line of children's furniture finished in a water-based system the buyer had specified in 2022, sat 22 percent above the others and had not moved in nine months. He did not know why. He suspected the drying oven, but the oven was new and the vendor had certified its performance. He had asked the association's diagnostician to look at it twice, and the diagnostician had found nothing. The next chapter takes up the firm's side of this story, and it has to explain what the cluster could not: why a factory with access to the same shared assets as its neighbors still could not close a gap that its own data made visible.