A foreign hardware company choosing where to build in China usually makes the decision twice. The first time is over two days, on a factory visit, on the strength of a plant tour, a sample that happened to be on the shelf, and a rent quote that looks well below what the company was quoted in Shenzhen. The second time is eighteen months later, when the tooling is in one city, the quality engineer is in another, the lease has two years to run, and the process the product needs turns out to be one the district no longer permits.

The comparison below is not a ranking. Shenzhen, Dongguan, Suzhou and Ningbo are not four versions of the same offer at four price points. They are four industrial systems with different specialisations, different labour markets, different regulators and different tolerances for the kind of process a product requires.

The most important conclusion is that the city decision is a decision about permissions and constraints, not about rent. Rent is the number every founder compares and the number that matters least once tooling, power, process permission, freight and engineering time are counted. A second conclusion follows: where the company is registered and where the product is built are separate decisions. Collapsing them into one, which is what happens when the first advisor registers the entity in whatever city is convenient, produces tax, incentive and permit outcomes that are awkward and expensive to undo.

What the location decision actually determines

A city choice fixes five variables that are expensive to change later.

It fixes the time to a first working article, because that is a function of how many of the components in the bill of materials have a supplier within a short drive. It fixes the time from design freeze to volume, because that is a function of how much queue there is at the tool shop and how quickly a change can be made, quoted and re-approved. It fixes the fully loaded unit cost at your volume, which is not the same as the quoted price. It fixes whether the district will permit the process the product needs, which is a binary question rather than a matter of degree. And it fixes how much leverage you hold when the relationship sours, because a cluster with fifty alternatives behaves differently from one with three.

It does not fix tax residence, customs treatment, intellectual property position or liability exposure. Those follow from the entity, the contract and the product, and they travel with you. A company that has chosen Dongguan for cost, drafted a contract governed by Shenzhen law and agreed to arbitrate in Hong Kong has made three decisions, and each one needs to be made on its own terms by the people competent to make it.

The regional shorthand remains a useful first cut. The Pearl River Delta, which in this context means Shenzhen and Dongguan, is the densest electronics cluster in the world. Factories there are organised around speed, small batches and frequent change, and the ecosystem assumes the customer will revise the design three times before the mould is cut. The Yangtze River Delta, meaning Suzhou and Ningbo, is larger, older and more institutional. Factories there are organised around process control, documentation and repeatability, and they assume the customer will arrive with a specification that has stopped moving.

Both assumptions are self-fulfilling. A PRD factory will quote a change in two days and a YRD factory will quote a change in two weeks, and each is telling you the truth about its own system. Neither is lying, and neither is better. The question is which system your product and your funding cycle can live inside.

One further distinction matters before any comparison. The PRD operates as a twin-city system rather than two competing cities. Design, engineering and commercial management sit in Shenzhen; moulds, tooling, injection moulding and volume assembly sit in Dongguan. The border between them is a forty-minute drive on a good day and a two-hour drive on a bad one. Companies that treat the two as alternatives usually discover that their chosen factory in Dongguan has its engineering office in Shenzhen and its tool shop in Chang’an, and that they have been comparing the same supply chain against itself.

The six criteria, defined so that they can be scored

Most city comparisons are made on adjectives: strong ecosystem, good talent, better protection. None of that can be scored, and a matrix built from adjectives returns whatever answer the person filling it in wanted. The six criteria below can each be tested within a week, with evidence that survives a second opinion.

Supplier depth is not the number of factories in the city. It is the number of qualified sub-tier suppliers within roughly ninety minutes of the assembler, in the specific processes the product touches. A device with an aluminium enclosure, an anodised finish, a custom overmoulded cable, a display module and a populated printed circuit board touches five separate supply chains, and each has its own geography. The constraint is usually surface treatment. Anodising, electroplating and painting are the processes districts regulate most tightly and that have migrated furthest from the city centres. A factory that can assemble the product may still truck the parts four hours each way for finishing, and that cost reappears in the quote as a line item labelled surface treatment that nobody questions.

The test is concrete. Mark every process in the bill of materials that is not assembly, and ask each candidate factory to name the supplier it would use for each one, with the address. Then check the addresses on a map. Factories that hesitate, or that name a supplier in another province without explaining why, are telling you something about the depth of their sub-tier relationships.

Talent splits into three pools that are hired differently and priced differently. Process engineers own yield and are the people who will actually solve the problem when the first article fails. Quality engineers own the customer audit and the corrective action report, and they are the people your customer’s auditor will interview. Technicians run the line and are the largest headcount. The number that matters is not the salary but the turnover, because a line that replaces a third of its workers each year will produce a different product in month nine than it produced in month one.

Ask for the names, tenure and previous employers of the four people who would be assigned to the programme. Ask what proportion of line workers returned after the Spring Festival in the most recent year. A factory that answers precisely is a factory that tracks the number, and a factory that tracks the number is usually a factory that can hold a process in place.

Logistics is four separate questions rather than one. Which port, and how long a container takes to reach it. Which airport, and whether it has freight capacity for the product’s weight and volume. Whether the factory holds foreign trade operator registration and can export in its own name. And whether cross-border trucking to Hong Kong is reliable, which matters for samples and small batches moving by air.

The third question catches more foreign buyers than the other three combined. A factory that sells only into the domestic market will quote an attractive price and then explain that the export must be handled by a trading company, which adds a margin and changes the documentation chain. The export VAT rebate is not free money; it accrues to whichever entity exports, and the price you negotiate should reflect who receives it. A quote that does not state whether it is tax-inclusive, and whether the rebate is retained by the factory, is not a comparable number.

Cost has to be compared at the same volume and the same specification, which almost never happens in practice. Rent is the visible number and, for most consumer electronics, a small fraction of unit cost. What moves the total is labour productivity, scrap rate, power reliability, freight, the cost of your own travel and the cost of the management time spent compensating for distance. A city that is 30 per cent cheaper on rent and 10 per cent worse on first-pass yield is not cheaper. It is simply cheaper on the line that the founder is looking at.

Incentives are granted at district level rather than city level, and they are negotiated rather than published. A typical package combines a rent-free or reduced-rent period in a designated park, a relocation or equipment subsidy, a share of locally retained tax for a defined period, and housing or education support for named employees. The value depends on what the company commits: registered capital, expected annual tax contribution, headcount and the date by which production begins. The agreement is documented in an investment agreement with the district, and it usually contains clawback provisions if the commitments are missed.

The mistake is to treat the headline as a discount on cost. For an export-heavy hardware company with thin margins and losses in the first two years, locally retained tax may be close to nil, and the incentive that looked like a subsidy turns out to be a rent holiday worth a few months of occupancy. Model the incentive against your own projections, not the district’s, and ask specifically what happens to the subsidy if the second funding round slips by a year.

IP and regulatory risk are two criteria that get merged and should not be. IP risk is about enforcement: whether a dispute would be heard in a court with technical experience, how long that takes, whether the defendant is a significant local employer, and whether the contract provides a route to arbitration outside the mainland. Regulatory risk is about permission: whether the district will allow the process, at that site, with that wastewater discharge, and under what category of environmental assessment.

The test for IP is to ask a litigator rather than a factory how a design patent or trade secret case against a company of the supplier’s size would run in that city, and what the realistic timeline and recovery look like. The test for permission is to ask the district ecology and environment bureau, in writing, whether the process falls inside the categories that require a full assessment or the categories restricted to designated industrial parks.

CriterionWhat it really measuresHow to test it in a weekThe false positive
Supplier depthNumber of qualified sub-tier suppliers within ninety minutes, by processAsk for named suppliers and addresses for every non-assembly process; plot themA large factory that outsources finishing to another province
TalentRetention and depth in three distinct poolsNames and tenure of assigned engineers; Spring Festival return rateImpressive headcount, high turnover, thin engineering bench
LogisticsPort, air, export licence, cross-border truckingAsk who exports and who claims the rebate; check forwarder ratesLow freight quote that excludes dangerous goods handling
CostFully loaded cost at your volume and specificationSame BOM, same volume, three quotes, all-inRent comparison at different volumes and specifications
IncentivesValue of what is actually bankable, against commitmentsRead the draft investment agreement, not the brochureHeadline subsidy tied to tax you will not generate
IP and regulatory riskEnforcement venue and process permissionLitigator’s view; written answer from the district bureauA factory’s own assurance that “there is no problem”

Building the matrix

A matrix is only useful if it changes the answer. The one below uses the six criteria, weights them by stage, and scores each city from one to five. The scores are illustrative and reflect the authors’ judgement of the four clusters; the weights are the part the reader should set.

For a consumer electronics product at pilot volume, speed and supplier depth dominate, because the design is still moving and the cost of a two-week change is greater than the cost of a higher unit price. For a regulated medical device, permission and enforcement dominate, because a design that cannot be certified is worthless and a design that can be copied with impunity is worse.

CriterionWeight, consumer electronicsWeight, regulated deviceShenzhenDongguanSuzhouNingbo
Supplier depth25205542
Speed and flexibility25554.52.53
Talent1525534.53
Cost15101325
Logistics101043.545
IP and regulatory103032.553

Weighted, the consumer electronics column puts Shenzhen first, Dongguan a close second and Suzhou third. The regulated device column inverts that: Suzhou moves to the top, Shenzhen falls to second and Dongguan drops well behind. Same four cities, same six criteria, different answer, because the constraint changed. That is the entire point of the exercise, and it is why a comparison published without weights is not a comparison.

Volume acts as a third weighting, and it is the one founders apply instinctively. Below roughly 5,000 units, Shenzhen’s speed and component availability usually outweigh its cost, because the programme is still in the phase where a design change costs more than a unit price. Between 5,000 and 100,000 units, Dongguan’s combination of tooling depth and lower overhead becomes the better fit for most products with mechanical content. Above 100,000 units, the labour and overhead savings in Suzhou or Ningbo start to exceed the logistics cost of being further from the design team, and a second production base becomes worth the management burden.

Those bands are heuristics rather than rules, and they break for products with unusual weight, unusual process requirements or unusual certification. A heavy product with a stable design can justify Ningbo at 3,000 units because freight dominates. A connected device with a radio cannot use Ningbo at any volume until the firmware and RF capability exists locally. The bands are a starting point for the weights, not a substitute for them.

Three disciplines keep the matrix honest.

Separate kill criteria from scored criteria. Some questions are not matters of degree. If the district will not permit electroplating at the site, no score on talent or logistics rescues the decision. Kill criteria should be tested first, before any scoring, because scoring a city that will not grant the permit wastes a month.

Require two sources per score. One source is an impression. Two sources, ideally one of them written and one of them from someone with no commercial interest in the outcome, is evidence. The sourcing agent who introduced the factory has an interest; the freight forwarder who handles the account does not.

Treat a close result as a decision about the supplier, not the city. If two cities score within a few points, the matrix has done its job by eliminating the rest. The remaining choice should be made on the specific factory, the specific team and the specific engineer who will run the programme. Cities do not build products; teams do.

What each city is actually good for

The descriptions below keep the cost profiles used in the published comparison and add the operating consequence that the profile implies.

Shenzhen

Shenzhen grew from a fishing village in the 1980s into a metropolis of roughly 17 million with a GDP above $400 billion, and it is the reference point for hardware prototyping worldwide. The density is the product. Huaqiangbei alone compresses component sourcing, sample buying and informal technical advice into a few city blocks, and the practical effect is that a design change can be sourced, bought and fitted within a day.

The cost profile is factory rent of roughly ¥35 to ¥65 per square metre per month, skilled workers at ¥6,000 to ¥9,000 and engineers at ¥12,000 to ¥20,000. Those numbers are the highest of the four cities and they are also the least important number in the Shenzhen case, because the premium buys iteration speed. A team that can run four design cycles in the time a competitor runs one is not paying 40 per cent more; it is buying a different product development schedule.

What Shenzhen does not do well is volume at low cost, and it does not do heavy or dirty process. Industrial land is scarce and expensive, which is why the factories that serve Shenzhen companies are physically in Dongguan. Electroplating, painting and other regulated processes have been pushed out of the city, and a Shenzhen address does not mean a Shenzhen factory floor.

The talent advantage is real and specific. Firmware engineers, industrial designers and mechanical engineers who have shipped consumer products are available in a way they are not in Ningbo. English proficiency is the highest of the four, which matters more than founders expect in the first six months, when every technical conversation is also a translation exercise.

Key areas: Nanshan for startups and software, Bao’an for electronics manufacturing, Longhua for Foxconn-linked logistics, Pingshan for lower-cost space.

Dongguan

Dongguan is a manufacturing city of roughly 10 million with more than 150,000 industrial enterprises, sitting between Shenzhen and Guangzhou. Its identity is tooling. Mould making, injection moulding, metal stamping and precision machining are the core competences, and Chang’an in particular has a concentration of mould shops that can cut a complex tool in two to four weeks where a Western supplier would quote eight to twelve.

Costs are materially lower: rent of ¥18 to ¥35 per square metre, skilled workers at ¥4,500 to ¥7,000, engineers at ¥8,000 to ¥15,000. For a product with significant mechanical content and a design that has stabilised, that gap is the difference between a viable gross margin and a marginal one.

The operating consequence of the cost profile is that Dongguan is where a product goes after it has been validated, not where it is invented. Engineering depth is thinner than Shenzhen, English is thinner still, and the startup ecosystem is close to absent. Quality varies more between factories than within a city, which makes the vetting step decisive rather than advisable. The regulatory environment is also less predictable than Suzhou’s, and enforcement varies by district.

The proximity advantage is easy to underestimate. A Shenzhen engineering team can be on a Dongguan line in the morning and back in the office by the afternoon, which means problems get seen rather than reported. Companies that run the two cities as one operation tend to have fewer nasty surprises at pilot than companies that treat the move as a relocation.

Key areas: Songshan Lake for the premium end, Chang’an for moulds, Humen for electronics assembly, Fenggang for proximity to Shenzhen.

Suzhou

Suzhou is a city of roughly 12 million people, twenty-five minutes from Shanghai by high-speed rail. Its distinguishing feature is institutional maturity. The Suzhou Industrial Park was built as a joint venture with Singapore and has spent three decades accumulating the things that make regulated manufacturing possible: clear procedures, functioning customs, a concentration of testing laboratories, and a local administration that is used to foreign companies asking procedural questions.

The cost profile reflects that: rent of ¥30 to ¥55 per square metre, skilled workers at ¥5,500 to ¥8,000, engineers at ¥10,000 to ¥18,000. In premium zones Suzhou is comparable to Shenzhen. The difference is what the money buys. It buys a cleanroom that can be certified, a supplier base that understands documentation, and an enforcement environment in which an IP case is more likely to be decided on its merits than on the identity of the defendant.

The weakness is speed and flexibility. Suzhou’s ecosystem is dominated by large established companies with long qualification cycles, and a startup with a design that changes weekly will find the pace frustrating. The electronics supply chain is also less dense than the PRD’s, which means component sourcing is slower even when the factory is better.

For medical devices, automotive electronics and anything that requires a certified process, the trade is worth making. For a wearable with a six-month runway, it usually is not.

Key areas: Suzhou Industrial Park, Suzhou New District for semiconductors and biomedicine, Kunshan for Taiwanese electronics and PCB work, Wujiang for textiles and optoelectronics.

Ningbo

Ningbo is a port city of roughly 9.5 million on the Zhejiang coast, and its manufacturing identity was built by private enterprise rather than foreign investment. The result is a cost-focused, highly entrepreneurial ecosystem that excels at mechanical and electrical products: moulds, castings, injection moulding, small motors, appliances and electrical components.

The cost profile is the lowest of the four: rent of ¥15 to ¥28 per square metre, skilled workers at ¥4,000 to ¥6,500, engineers at ¥7,000 to ¥13,000. The port is among the busiest in the world by cargo throughput, and for a heavy or bulky product the freight advantage compounds the manufacturing advantage.

The limitation is scope. Ningbo is not an electronics city. Consumer electronics supply chains are thin, firmware and RF talent is scarce, and the startup community is minimal. English proficiency is the lowest of the four. For a mature mechanical product with a stable design and a cost-driven buyer, that hardly matters. For a connected device with a radio, a display and an app, it matters a great deal.

Key areas: Beilun for the port and heavy industry, Yinzhou for mechanical and electrical work, Yuyao for moulds, Cixi for appliances and small motors.

Cost lineShenzhenDongguanSuzhouNingbo
Factory rent (¥/sqm/month)35 to 6518 to 3530 to 5515 to 28
Skilled worker (¥/month)6,000 to 9,0004,500 to 7,0005,500 to 8,0004,000 to 6,500
Engineer (¥/month)12,000 to 20,0008,000 to 15,00010,000 to 18,0007,000 to 13,000
Not in the rent quoteIndustrial land scarcity, so the floor may be in another citySurface treatment trucked out of districtPark management fees and cleanroom fit-outDistance from electronics sub-tier

The last row is the one that decides budgets. Rent is a line item; the things that are not in the rent quote are the reasons a business plan misses.

Beyond the four

Four other cities appear often enough in board discussions to warrant a view.

Shanghai has the best research and regulatory affairs capability in the country and the highest costs, above Shenzhen in most categories, with very little industrial land left. The common structure is headquarters, engineering and regulatory affairs in Shanghai, production in Suzhou. A company that tries to manufacture inside Shanghai is usually paying a premium for a constraint.

Guangzhou is the PRD’s commercial centre and the host of the Canton Fair, with strong automotive and machinery manufacturing and costs below Shenzhen. It is a reasonable location for a product in the automotive supply chain and a poor one for a consumer electronics startup, because the hardware startup ecosystem is in Shenzhen.

Hangzhou has built a credible smart manufacturing and internet of things cluster on the back of its software industry, with active local government support. It suits connected products where the software team is the centre of gravity, and it does not yet have the component density of Shenzhen.

Chengdu and Chongqing offer factory rent well below Shenzhen, with growing electronics assembly in Chengdu and automotive manufacturing in Chongqing. The trade-offs are a longer logistics chain to the coast, a thinner English-speaking talent pool and a supply chain that is deep in some processes and shallow in others. They are worth considering for a second base serving domestic demand, less so for a first base serving export customers.

The multi-city model, and what it costs to run

The most common structure among hardware companies that have scaled in China is not a choice between cities but a sequence through them.

The classic path runs prototyping in Shenzhen, tooling and pilot production in Dongguan, and volume in Dongguan or a second base. In the first six months the design iterates in Shenzhen with batches of ten to a hundred units, sourced from component suppliers within the city. In months six to nine the tooling is cut in Dongguan, the pilot runs at 500 to 5,000 units, and the process is proven on the line that will eventually run it. From month nine, volume production sits in Dongguan while engineering and quality remain in Shenzhen. Above roughly 100,000 units a year, a second base in Suzhou or Ningbo reduces cost and adds capacity, usually for a mature product whose design has stopped moving.

The model works because each city is doing the job it is organised for, and because the distance between the first two is short enough that a problem can be seen in person the same day. It fails when a company treats the sequence as a relocation and dismantles the engineering presence in Shenzhen once production moves, which removes the capability that made the move safe.

Running two sites costs more than the rent difference suggests. It requires a quality presence at each location, because nobody in Shenzhen can see a drift in Dongguan. It requires a documented quality system that both sites operate to the same standard, because otherwise the second site produces a different product under the same part number. It requires a logistics arrangement for moving material and finished goods between sites, and a single point of contact on the factory side so that the buyer is not mediating between two plant managers. And it requires travel, at least quarterly and preferably monthly during a ramp.

The companies that run it well share four habits. They start with one anchor partner and expand the relationship rather than the supplier list. They hire local quality engineers at each site rather than flying one person between them. They standardise documentation to the point where a first article inspection means the same thing in both plants. And they keep one commercial owner for the whole supply chain, so that price, capacity and quality are negotiated as one conversation rather than three.

The permission layer that foreign companies do not check

A factory lease in China is not a permission to manufacture. Permission comes from a set of approvals attached to the site and the process, and they are granted by different authorities on different timelines. A company that signs a lease on price and discovers afterwards that the process cannot be permitted has bought a building it cannot use.

Environmental assessment. A construction or process project requires an environmental impact assessment, filed with or approved by the district ecology and environment bureau, depending on the category the process falls into. Assembly is usually straightforward. Electroplating, anodising, painting, etching, cleaning with solvents and any process producing wastewater or volatile emissions fall into categories that require more, and in some districts are restricted to designated industrial parks or refused outright.

The practical consequence is that a product whose enclosure needs anodising may be assembled anywhere and finished in only a few places, and the finishing step may be the schedule constraint. Ask the question at the site level, before the lease, and get the answer in writing.

Fire acceptance. Factory buildings require fire safety acceptance before occupation and use, and the requirements differ for a warehouse, a production floor and a space storing lithium batteries or flammable materials. A landlord who cannot produce the building’s fire acceptance documentation is a landlord whose building may not be usable for your process.

Land and title. Foreign investors do not own land in China. They hold land use rights, granted by the local government for a fixed term, and most foreign manufacturers lease buildings from a landlord who holds those rights. The diligence question is whether the landlord holds a valid land use rights certificate and whether the building has a property ownership certificate. Unregistered buildings exist, and they create problems at the environmental and fire stages rather than at the lease stage.

Power. Capacity, not price, is the issue. A line that needs more connected load than the building was designed for requires an upgrade from the local grid company, which takes time and money. Interruptions are less frequent than they were, but factories in some districts still run generators during peak periods, and a product that requires uninterrupted process control needs to know this before it signs.

Discharge. A wastewater discharge permit is separate from the environmental assessment and is issued on the basis of the assessed volume and composition. A factory that holds a permit for a certain volume cannot simply increase it because a new customer’s process produces more effluent.

PermissionAuthorityWhat triggers itTypical timingWhat foreign companies miss
Environmental impact assessmentDistrict ecology and environment bureauAny construction or process project; category depends on processWeeks to months, category-dependentThat the category follows the process, not the product
Wastewater discharge permitDistrict ecology and environment bureauAny process producing effluentFollows the assessmentThat the permitted volume is a ceiling
Fire acceptanceLocal fire and rescue authorityOccupation and use of the building; storage of hazardous goodsBefore occupationThat battery storage changes the requirement
Land use and property certificatesLandlord, verified against the registerAny leaseAt lease signatureThat unregistered buildings exist and are leased
Power capacity upgradeLocal grid companyConnected load above the building’s designMonthsThat the tenant usually pays

Registration, incentives and where the entity sits

The entity question is separate from the factory question, and the two get conflated because both involve choosing a city.

A wholly foreign-owned enterprise is registered at a district-level administration for market regulation, and the registered address determines the tax bureau that administers the company, the social insurance base for its employees, and its eligibility for district-level incentives. A company that registers in Shenzhen because the founder’s advisor is in Shenzhen, and then builds in Dongguan, has a Shenzhen taxpayer with a Dongguan factory. That is workable, but it means the incentives available in Dongguan are not available to it, the local authority that hosts the factory has no tax relationship with the operating company, and any application that requires local tax records will be made in the wrong district.

The cleaner structure, where the scale justifies it, is a group with a holding or commercial entity in the city where the founders, investors and customers are, and a manufacturing subsidiary in the city where the plant is. The cost is a second set of books, a second set of statutory filings and an intercompany supply agreement that has to be priced on arm’s length terms for transfer pricing purposes. The benefit is that each city’s incentives, permits and tax administration attach to the entity that actually generates the activity.

There is a third point that catches companies without a local entity at all. A foreign company that places staff in a Chinese factory for extended periods, directs the work and pays them, may create a taxable presence and an employment law exposure that it did not intend. Secondment arrangements need to be documented, and the labour contract should sit with an entity that is entitled to sign it.

On customs, the question is which entity exports. If the factory holds foreign trade operator registration, it can export in its own name, claim the export VAT rebate and invoice you accordingly. If it does not, a trading company stands between you and the factory, takes a margin and receives the rebate. Neither structure is wrong, but the price is not comparable across the two, and the quality relationship is weaker in the second because the trading company, not the factory, is your counterparty.

Incentive agreements deserve the same scrutiny as the supply agreement. They are contracts with clawback provisions, they are usually conditional on registered capital being paid in within a period, and they are often drafted in a form the district has used for a decade. Read the termination and repayment clauses before signing, and check whether the commitment to headcount is measured at a point in time or on average, because a company that hires late in the year can fail a point-in-time test it would have passed on average.

Running the selection in ten weeks

The sequence below is the one that produces a defensible decision without consuming a quarter. It assumes a team of three: a founder or chief operating officer, an engineer who owns the bill of materials, and external advisers for the legal and tax questions.

WeekStepResponsibleOutputGate
1Define the process map: every non-assembly process, its category and its regulatory sensitivityEngineering leadProcess map with regulatory flagsAny process that is restricted narrows the city list before scoring
1Set weights and kill criteria; agree them in writingFounder and engineering leadWeighted matrix templateWeights agreed before any factory visit, so the visits do not set them
2Desk screen: districts, parks, available buildings, incentive policiesOperations lead, with a local sourcing adviserLong list of three districts per cityLong list cut to two cities per region
3Written regulatory enquiries: environmental category, fire, dischargeLocal counsel or environmental consultantWritten responses from district bureauxAny city that refuses the process is eliminated
4First factory visits, no price discussionFounder, engineering leadCapability notes, named sub-tier suppliersShortlist of two factories per city
5Quotation round at a fixed volume and specificationOperations leadComparable all-in quotes with tax status statedQuotes normalised before comparison
6Entity and tax structuring advice; incentive conversations at district levelCorporate counsel and tax adviserStructure memo; draft investment termsStructure and site are consistent
7Second visits: meet the owner, the process engineer and the quality managerFounder and engineering leadTeam assessment, culture readSupplier chosen, subject to contract
8Legal diligence on the shortlisted factory: licences, land, fire, litigationLocal counselDiligence reportNo unresolved title or permit defect
9Negotiate the NNN, the tooling terms and the supply agreementCounsel, with the founder on commercial termsExecuted or near-executed documentsContract terms match the matrix assumptions
10Lease, entity registration steps, pilot orderOperations lead and corporate counselSigned lease, filed registration, purchase orderPilot scheduled with a named engineer

Two things about the sequence are worth stating plainly. The regulatory enquiry in week three comes before the factory visits in week four, because there is no point visiting a factory that cannot host the process. And the entity work in week six runs in parallel with the commercial work rather than after it, because the incentive conversation and the site choice have to agree.

Where foreign companies get this wrong

Choosing on rent. The rent differential between Ningbo and Shenzhen is large in percentage terms and small in absolute unit cost for most electronics. Companies that optimise it discover in month nine that the scrap rate is three points higher, the engineer they need is in Shenzhen, and the cost of flying there twice a month exceeds the saving. The consequence is not bankruptcy; it is a slow, invisible margin leak that nobody owns because it is spread across travel, rework and delay.

Choosing on a single visit. A factory shows its best line, its best product and its best English speaker. The visit is a demonstration, not an audit. The consequence of deciding on it is that the buyer discovers the real capability after the tooling is paid for, at which point switching costs more than tolerating.

Treating English capability as engineering capability. The sales engineer who speaks fluent English is rarely the process engineer who will solve the yield problem. Companies that route all technical communication through the English speaker get a translation of an answer rather than the answer. The consequence is a specification that was understood by the wrong person.

Registering the entity wherever is convenient. The registered address determines the tax bureau, the incentive eligibility and the district relationship. A company that registers in the city where its lawyer sits, and builds elsewhere, forfeits the incentives attached to the plant and creates an administrative mismatch that surfaces at the first tax inspection. Fixing it means a migration, which means a new tax file, new bank accounts and a new social insurance relationship.

Running a multi-city supply chain without local presence. The Shenzhen-to-Dongguan model works because the distance is short. Extending it to a second base in Suzhou or Ningbo without a quality presence on site converts a cost saving into a quality risk, because nobody is close enough to see the drift. The consequence is that problems are discovered by the customer rather than by the buyer.

Cost and commercial impact

The cost differences between these cities are real and the direction is stable: Shenzhen is the most expensive, Dongguan and Suzhou sit in the middle in different ways, and Ningbo is the cheapest on most lines. What varies is what the difference buys.

For a product in development, the premium paid in Shenzhen buys schedule. For a product in volume with a stable design, the discount obtained in Ningbo or Dongguan buys margin. For a product that requires certification, the premium paid in Suzhou buys permission and a paper trail. These are not the same purchase, and a company that compares them as though they were is comparing prices for different things.

The costs that are easiest to underestimate are the ones that do not appear in any quote. The cost of moving tooling is a fraction of the cost of requalifying a product at a new factory, which includes new samples, new first-article inspection, new certification in some cases and a new customer audit. The cost of a delayed launch is usually larger than the entire manufacturing cost difference between the best and worst city. And the cost of management attention is the one nobody books, even though for a small team it is the scarcest input of all.

Practical checklist

The four cities are not interchangeable, and the differences that matter are not the ones that appear in a cost table. A company that decides on permissions, tests its assumptions in writing, and revisits the choice when the constraint changes will usually end up in the right place. A company that decides on a plant tour will usually end up somewhere defensible and expensive.

For the entity, lease and incentive documentation that follows the city decision, see incorporation and corporate structure and commercial contracts. The guides that continue this sequence are how to select and vet factories in China, the cost of manufacturing in China and negotiating with Chinese factories.

Frequently Asked Questions↑ Back to article

Which Chinese city is best for hardware startups?

There is no single best city — it depends on your product type, production volume, and stage. Shenzhen dominates for electronics prototyping and fast iteration. Dongguan offers the best balance of cost and manufacturing depth for mid-volume production. Suzhou excels at high-precision, regulated industries like medical devices. Ningbo is the most cost-effective for mechanical and electrical hardware at scale.

Is Shenzhen more expensive than other manufacturing cities?

Yes, Shenzhen is the most expensive manufacturing city in China. Factory rent in Shenzhen can be 30-60% higher than in Dongguan or Ningbo, and skilled labor costs are roughly 20-40% higher. However, the speed and density of its supply chain often offset the premium for early-stage hardware startups that value iteration velocity over unit cost.

What is Dongguan known for in manufacturing?

Dongguan is known as the 'Factory of the World' and specializes in mold making, injection molding, metal stamping, and precision machining. It has the highest factory density in the Pearl River Delta and serves as the production backbone for many Shenzhen-based companies. Dongguan is particularly strong in consumer electronics components, footwear, textiles, and mechanical assemblies.

Why is Suzhou popular for foreign manufacturing?

Suzhou is popular with foreign companies due to its well-developed industrial parks (especially the Suzhou Industrial Park), strong regulatory infrastructure, high-quality talent pool, and proximity to Shanghai. It has a mature ecosystem for high-end manufacturing, including medical devices, automotive electronics, and precision engineering. The city also offers clearer regulatory pathways and better intellectual property protection frameworks compared to many other Chinese manufacturing hubs.

How do I choose between manufacturing cities?

Start by evaluating your product type, production volume, and stage. If you are building complex electronics and need fast prototyping, start in Shenzhen. If you need cost-effective mid-volume production with strong tooling support, consider Dongguan. For high-precision, regulated products requiring scale, Suzhou is often the best fit. For cost-sensitive mechanical or electrical products at high volume, Ningbo offers the strongest value. Many startups use a multi-city strategy — prototyping in Shenzhen and scaling in Dongguan or Suzhou.

What is the difference between the PRD and YRD manufacturing regions?

The Pearl River Delta (PRD) in the south, including Shenzhen and Dongguan, is the world's densest electronics cluster characterised by extraordinary speed, supply chain density, and entrepreneurial energy, with factories accustomed to rapid change and small batch sizes. The Yangtze River Delta (YRD) in the east, including Suzhou and Ningbo, is China's largest and most advanced region with a stronger focus on high-end manufacturing, precision engineering, and regulated industries, tending to be more formal and institutionalised.

What are Ningbo's strengths as a manufacturing hub?

Ningbo delivers the lowest costs for mechanical and electrical hardware, with world-class port access for export logistics. It excels at cost-competitive mechanical and electrical production at high volume, making it the most cost-effective choice for products where unit cost is the primary driver. While Ningbo has lower engineering depth compared to Shenzhen or Suzhou, its combination of low cost and excellent logistics makes it ideal for scaling mature products.

Geason Law Offices AI Industry Research Group | Reviewed by He Qianggao (Lawyer License No. 14403201310092337, 10+ years of practice)