Two quotations arrive for the same server chassis. The Shenzhen supplier is cheaper on the piece price, holds the tooling already, and can turn an engineering change around in a week. The Vietnamese supplier quotes higher, needs a longer ramp, and has to import the thermal parts and the connectors. The customer buying the finished rack has asked for a bill of materials that does not read as China-origin, and the tariff exposure on China-origin goods is large enough to move the landed cost by more than the piece-price gap. Neither quotation answers the question that actually decides the programme, which is whether the Vietnamese plant can build the product to specification and whether anyone can prove, to a customs officer’s satisfaction, that the finished goods are Vietnamese.
That question is legal before it is commercial. Tariff exposure is set by origin rules, not by the address of the factory. Labour cost is set by statute as much as by the market. A factory cannot be built until land, environmental, fire and construction approvals are in place, and those approvals run on the authority’s timetable rather than the customer’s. The most expensive mistakes in a China plus one programme are made in the eighteen months before the first shipment, in documents that nobody drafted.
The conclusion this guide reaches is narrow and, in practice, unwelcome. There is no correct country. There is a correct allocation of process steps between the two, and a documentation set that makes the allocation defensible. Companies that move a whole product because a consultant produced a cost model tend to return to China within two years, or to discover that their Vietnam-origin claim does not survive a verification visit. Companies that move the steps Vietnam can perform competitively, and paper the link between the two sites, get the tariff benefit and keep the engineering depth.
The decision is an allocation of steps, not a choice of country
A server, a switch, a sensor module and a robot joint decompose differently, but each breaks into steps with distinct capital intensity, skill requirements and regulatory treatment. Design and design-for-manufacture support. Tooling. Precision component manufacture. PCB fabrication. PCBA. Sub-assembly. Final assembly. Test and burn-in. Packing. Export clearance. After-sales and returns. Each step has a location, an owner and a document that evidences it.
Once the product is written down as a sequence, the country question dissolves into a series of smaller questions, each answerable on evidence rather than intuition.
| Process step | China | Vietnam | What decides the location | Document that evidences the step |
|---|---|---|---|---|
| Design and DFM support | Deep process engineering inside the supplier base | Thin, concentrated in a few park-based teams | Whether the supplier can propose manufacturability changes without breaking the design intent | Design release package, DFM report, engineering change note |
| Tooling and moulds | Dense tool shop clusters, short iteration | Limited; much tooling imported from China, Korea or Taiwan | Tool steel supply, heat treatment, press tonnage availability | Tooling purchase order, tooling schedule, bailment and title clause |
| Precision machining and magnets | Deep, including robotics-grade parts | Growing, with capability gaps at tight tolerance | Tolerance, material and volume | Supplier declaration, material certificates, dimensional report |
| PCB fabrication | Full range, including high layer counts | Available, with limits at the high end | Layer count, substrate, impedance control | Fabrication drawing, coupon test report |
| PCBA | Mature, high line count | Available at scale in the northern parks | Component mix, traceability, line capacity | Process flow, first-article approval, inspection records |
| Final assembly and test | Available, higher labour cost | The usual relocation target | Labour content, test equipment, tariff position | Work instruction, test log, serial number record |
| Export clearance | Chinese exporter of record | Vietnamese exporter, or an export processing enterprise | The origin claim and any licence | Customs declaration, certificate of origin, export licence |
| After-sales and returns | Near the factory | Near the factory or near the market | Reverse logistics cost and duty on re-import | RMA procedure, warranty terms, re-import classification |
The table carries the central point. A programme described internally as “moving to Vietnam” that leaves tooling, PCB fabrication and precision parts in China has not moved. It has added a final-assembly step at the end of a cross-border supply line. That structure can be entirely rational, and many hardware companies end up there. It is also the structure that attracts the closest customs scrutiny, because the transformation performed in Vietnam is the thinnest part of the chain.
The decision matrix
Seven dimensions decide most programmes. They do not carry equal weight, and the weighting changes with the product, the destination market and the customer’s own sourcing rules. Cost enters once, as landed cost, because a piece-price comparison that ignores duty, freight and the working capital tied up in a longer inbound pipeline is not a cost comparison at all.
| Dimension | China | Vietnam | The question that decides it | Working weight range |
|---|---|---|---|---|
| Landed cost | Lower piece price, higher duty exposure | Higher piece price, lower duty on qualifying goods | What does a delivered unit cost after duty, freight, inventory and scrap? | 20–30 |
| Supplier depth | Full component, tooling and process ecosystem | Assembly strength with thin supporting industries | Can every critical input be second-sourced within a week’s travel? | 15–30 |
| Tariff and origin | Section 301 duties plus the 2025 additional tariffs | Lower base rate, punitive rate for transshipment | Can the Vietnam step be shown to be a substantial transformation? | 10–30 |
| IP enforcement | Specialised courts, criminal trade secret route, customs recordation | Registration and civil enforcement, slower and with lower damages | Is the value in the design, the process or the brand? | 10–20 |
| Entity and permits | WFOE plus project, environmental, fire and construction approvals | IRC and ERC plus land, environmental, fire and construction approvals | How long can the programme wait for a permit? | 5–15 |
| Export control | Comprehensive dual-use regime with licensing | Lighter dual-use regime | Would Chinese technology, equipment or staff cross the border? | 5–15 |
| Labour | Higher wages and employer contributions, stricter termination rules | Lower wages, rising, with a statutory 48-hour week and overtime caps | Does the product carry enough labour content to make the wage gap material? | 5–15 |
The ranges do not sum to a hundred by accident. They are fixed by the product and the market, and the first discipline is to agree them in writing before anyone scores a country. A team that argues about weights after the scores are on the page is arguing about the conclusion it has already reached.
Score each dimension from one to five, where five is favourable to the buyer, multiply by the weight, and divide the total by five to express the result on a hundred-point scale. The scores are judgements, so record who scored, on what evidence and on what date. A matrix that cannot be defended in an investment committee, or shown to a customer’s supplier audit, is a spreadsheet rather than a decision.
Three worked examples
A rack-scale server or switch sold mainly into the United States. The bill of materials is documented, most components are catalogue parts, and the customer cares about origin more than about engineering support.
| Dimension | Weight | China | Vietnam | What drives the score |
|---|---|---|---|---|
| Landed cost | 30 | 2 | 4 | Duty overwhelms the piece-price advantage in China |
| Supplier depth | 20 | 5 | 3 | Chassis, thermal and connector supply is local in China |
| Tariff and origin | 20 | 1 | 4 | A genuine assembly and test step in Vietnam is provable |
| IP enforcement | 10 | 4 | 3 | Design value is modest; process copying is the risk |
| Entity and permits | 10 | 4 | 3 | Both need a full permit set; the northern parks move faster |
| Export control | 5 | 3 | 4 | Chips and equipment are screened on either route |
| Labour | 5 | 3 | 5 | Assembly labour content is high |
| Weighted score | 100 | 58 | 73 |
A robot joint or precision sensor sold into Europe. Machining tolerance, magnet supply and process engineering decide whether the part can be built at all.
| Dimension | Weight | China | Vietnam | What drives the score |
|---|---|---|---|---|
| Landed cost | 20 | 4 | 3 | No Section 301 exposure; Vietnam’s imported content raises cost |
| Supplier depth | 30 | 5 | 2 | Precision machining and magnets are concentrated in China |
| Tariff and origin | 10 | 4 | 4 | Neither route carries a decisive duty advantage |
| IP enforcement | 15 | 4 | 3 | The process recipe is the asset |
| Entity and permits | 10 | 4 | 3 | Similar sequences; China’s approvals are more predictable at scale |
| Export control | 10 | 4 | 3 | Technology transfer to Vietnam triggers Chinese rules |
| Labour | 5 | 3 | 5 | Labour is a small share of cost |
| Weighted score | 100 | 85 | 58 |
An edge device whose firmware carries the product’s value, sold into the United States.
| Dimension | Weight | China | Vietnam | What drives the score |
|---|---|---|---|---|
| Landed cost | 25 | 3 | 3 | Duty offsets the piece-price gap |
| Supplier depth | 15 | 5 | 3 | Most inputs are catalogue parts |
| Tariff and origin | 25 | 1 | 4 | Origin is the commercial issue |
| IP enforcement | 15 | 4 | 3 | Firmware and tooling are the assets |
| Entity and permits | 5 | 4 | 3 | A small plant with a short permit list |
| Export control | 10 | 3 | 4 | Embedded accelerators attract screening |
| Labour | 5 | 3 | 5 | Assembly is a minor cost line |
| Weighted score | 100 | 60 | 69 |
The third example is close enough that the weighting decides it, which is the useful part. Move ten points from supplier depth to tariff and origin, as a customer with a hard origin requirement would, and the same device scores 52 in China and 71 in Vietnam. Apply the identical shift to the robot joint and China still scores 83 against Vietnam’s 62, because supplier depth carries more of that decision than any duty rate can. A matrix that returns the same answer under every weighting is measuring the author’s preference rather than the programme.
That is also why the matrix belongs in the board pack and the investor data room. It shows which risks the company has chosen to carry, and it dates the choice. When a customer asks in eighteen months why production sits where it does, the answer is a scored document rather than a recollection.
Two patterns recur. A company that selects Vietnam purely on the tariff line, without scoring supplier depth or origin provability, has decided to accept whatever the origin analysis produces. A company that stays in China purely on supplier depth, without scoring tariff exposure, has decided to absorb a duty its customer may refuse to pay. Both are decisions, but only one of them is written down.
Supplier depth and what the bill of materials hides
China’s advantage is density, and density is an engineering asset rather than a cost line. The Greater Bay Area, Suzhou, Kunshan, Ningbo and Dongguan cluster PCB fabrication, connectors, precision machining, tooling, injection moulding and contract assembly within a few hours’ drive. A design change that needs a new electrode, a different connector or a revised heat spreader can be quoted and sampled inside a week. That responsiveness is difficult to price and almost impossible to replicate in a park where the nearest competent supplier is an import.
Vietnam’s advantage is position. Industrial parks in Bac Ninh, Bac Giang and Hai Phong in the north, and around Ho Chi Minh City, Binh Duong and Dong Nai in the south, host electronics assembly at scale and give access to CPTPP, EVFTA and RCEP markets. Supporting industries are thinner, and much of the bill of materials still arrives from China, Korea, Taiwan and Japan. Power reliability, inland logistics and the supply of skilled technicians vary by park, and a park that suits a footwear tenant may not suit a line that needs stable power for reflow and test.
The bill of materials is where these differences become visible, and where most cost models fail. A BOM that lists two hundred lines, of which forty are mechanical and thirty are connectors, tells you nothing until each line is tagged with three facts: whether it can be sourced locally in each country, which tier supplies it, and whether it is single-sourced anywhere in the chain.
Four questions turn a BOM into a site decision.
First, which lines are single-sourced globally, irrespective of country? A connector, a fan or a power module made by one supplier in one place will not become local because the assembly moves. Those lines carry lead-time and allocation risk that no site choice removes.
Second, which lines are single-sourced within the proposed country? A Vietnamese supplier that is the only local source of a machined housing is a concentration risk dressed as localisation.
Third, which lines are local in name only? A Vietnamese trading company that imports a Chinese part and invoices it locally does not create local content, and the underlying origin of the part remains Chinese for the purposes of any origin analysis and any certificate of origin.
Fourth, which lines would need requalification at a new supplier, and how long does that take in your industry? Qualification is not paperwork. It is sampling, testing, reliability work and, for some customers, their own approval process.
The contractual tools that protect supplier depth are unglamorous and effective. An approved supplier list annexed to the quality agreement, with a consent right over additions. A subcontracting clause requiring written approval before any process step is outsourced, with a named list of approved subcontractors. A requirement that the supplier hold a complete technical data package, so that the product can be rebuilt elsewhere. A second-source obligation or right for critical lines, funded through duplicate tooling where necessary. And a transition assistance clause that obliges the supplier to cooperate with a move, which is the clause nobody negotiates until it is too late to matter.
Tariffs, origin and what a customs officer will ask for
US tariffs are the reason most companies look at Vietnam. China-origin goods face Section 301 duties and, since 2025, additional tariffs under emergency and reciprocal frameworks. Vietnam-origin goods generally face lower rates, but the 2025 US-Vietnam framework sets a substantially higher rate for goods treated as transshipped. Rates are adjusted frequently. Any landed-cost model built on a rate captured more than a quarter ago should be rebuilt with a customs broker before a lease is signed.
Origin is a legal test, not a label on a carton. US customs applies a substantial transformation analysis, asking whether the operations in Vietnam change the name, character or use of the imported inputs. Assembling imported kits with a screwdriver rarely meets that standard. A programme that imports a complete sub-assembly, adds a fan and a cable, and ships the result as Vietnamese is not a manufacturing relocation. It is a transshipment structure with a higher duty rate attached.
The evidence file is the whole game. A defensible Vietnam origin rests on records that a verification officer can follow without help from the company.
| Origin question | Evidence | Who should hold it | Common gap |
|---|---|---|---|
| What went into the product | Multi-level bill of materials with part numbers, quantities and supplier names | Buyer and Vietnam entity | BOM stops at tier one; tier-two origin unknown |
| What the inputs cost | Costed BOM, purchase invoices, freight allocation | Vietnam entity’s finance team | Costs recorded at group level only |
| What was actually done in Vietnam | Routing, process sheets, work instructions, machine and line logs | Vietnam plant | Routing describes steps the plant does not perform |
| How long it took and who did it | Payroll, attendance, headcount by line | Vietnam HR | Headcount inconsistent with claimed labour content |
| Where the inputs came from | Supplier declarations, certificates of origin, import entries | Vietnam entity’s logistics team | Declarations unsigned or generic |
| What was scrapped and reworked | Scrap reports, rework records, yield data | Quality function | No records, so the transformation looks theoretical |
Vietnam’s own trade agreements add a second layer with different rules. CPTPP, EVFTA, RCEP and the ASEAN arrangements each carry product-specific rules of origin and their own certificates. A good can be Vietnamese for US purposes and fail the EVFTA rule, or qualify under RCEP and not under CPTPP. Companies that treat “made in Vietnam” as a single status discover this at the port, when the buyer’s customs agent asks for a certificate the supplier cannot issue.
Three practical disciplines follow. Keep the origin analysis per destination market, not per product. Obtain a customs ruling where the transformation is arguable, rather than relying on the supplier’s assurance. And treat the origin file as a live document that is updated when a supplier, a process or a part changes, because the most common verification failure is a file that describes the product as it was two years ago.
The downside of getting this wrong is not a reclassification. It is retroactive duty, penalties, and in serious cases a finding that affects every shipment in the period under review. Those numbers routinely exceed the tariff saving that motivated the move.
Labour: the cost line and the compliance line
Both countries are regulated labour markets. Neither is a low-compliance environment, and a company that budgets for wages alone will be surprised in both.
China’s manufacturing wages are higher, and the employer burden is significant. Social insurance and housing fund contributions commonly add around a third or more on top of salary, depending on the city, and the contribution base is policed more closely than it once was. The Labour Contract Law requires written contracts, limits termination and provides severance. Overtime is capped. Dispatched labour is restricted to temporary, auxiliary or substitute roles and to a minority of headcount. The compensating advantage is engineering and technician depth: process engineers who can diagnose a yield problem, toolmakers who can rework a mould, and line leaders who have run the product before.
Vietnam’s wages are lower, with regional minimum wages set by government decree, but they are rising and the regional bands move. The Labour Code sets a 48-hour standard week, caps overtime at 40 hours per month and 200 hours per year with limited exceptions, and requires social, health and unemployment insurance contributions plus a union fee. Employers need internal labour rules, a grassroots union, and proper severance or job-loss allowances. Foreign workers need work permits, and the law expects local hiring to be prioritised.
| Item | China | Vietnam |
|---|---|---|
| Wage level | Higher, with wide regional variation | Lower, with decree-based regional minimums |
| Employer contributions | Social insurance and housing fund, commonly around a third or more on top of salary | Social, health and unemployment insurance plus union fee |
| Standard hours | Statutory standard with capped overtime | 48-hour week, overtime capped at 40 hours a month and 200 a year |
| Termination | Restricted, with severance and procedural risk | Statutory grounds, severance and job-loss allowances |
| Flexible labour | Dispatch limited to temporary, auxiliary or substitute roles and a minority of headcount | Fixed-term and seasonal arrangements within Labour Code limits |
| Foreign staff | Work permit and visa regime | Work permit, with local hiring priority |
| Practical strength | Process engineering and technician depth | Cost and availability of assembly labour |
Two operational points matter more than the table. The first is that productivity and turnover decide the real labour cost, and turnover in a new park is high because every competitor is hiring from the same pool. Budget for training, supervision and a ramp curve that is longer than the cost model assumes. The second is that employment terms carry IP consequences. Employee inventions, design work and process know-how are governed by statute and reinforced by contract, so assignment and confidentiality clauses in employment contracts, and reward policies for inventors, need to be in place in both countries before the first engineer starts.
Entity, land and permits
In China, a foreign manufacturer sets up a wholly foreign-owned enterprise with a business scope that includes manufacturing, then completes project filing, an environmental impact assessment, a pollutant discharge permit, fire acceptance and construction approvals. Industrial land is held under a land use right, commonly for fifty years, or leased from a park or a landlord. Local governments compete on incentives, but energy-intensive projects face stricter review, and the power allocation for a line with heavy test equipment is a real constraint rather than a formality.
In Vietnam, a foreign investor obtains an Investment Registration Certificate and an Enterprise Registration Certificate, usually through the provincial authority or the industrial park. Land is leased in an industrial park, typically for up to fifty years, and a land use right certificate follows. The project needs an environmental impact assessment or an environmental protection plan, a construction permit, fire approval and a discharge permit. Export Processing Enterprise status can suspend import duty on inputs used for exported goods, which is material for an assembly model that imports most of its bill of materials.
The sequence is broadly similar in both countries, and so are the bottlenecks.
| Step | Authority | Typical timing | What foreign companies miss |
|---|---|---|---|
| Entity incorporation | China: market regulation administration. Vietnam: provincial authority or park board | Weeks, if documents are clean | Business scope or investment objective drafted too narrowly for the actual process |
| Investment or project filing | China: project filing with the local development and reform authority. Vietnam: IRC | Weeks to months | The filed scope omits coating, plating or test steps added later |
| Land or lease | China: land use right. Vietnam: park lease and land use right certificate | Months, park-dependent | Power capacity and permitted use confirmed verbally, not in the lease |
| Environmental assessment | China: ecology and environment authority. Vietnam: provincial environment authority | Months | Assessment written for the wrong process scale; amendment needed at expansion |
| Construction permit | Local construction authority | Months | Design changes after permit issue trigger re-approval |
| Fire approval | Fire authority | Weeks to months | Warehouse layout and chemical storage fail at inspection |
| Discharge permit | Environmental authority | Weeks after construction | Wastewater and exhaust treatment sized for the pilot, not the volume line |
| Export Processing Enterprise status | Vietnam: customs and park authority | Weeks, with conditions | Duty suspension assumed without meeting the record-keeping conditions |
In both countries the lease, the environmental permit and the fire approval are the practical bottlenecks. Confirm permitted uses, floor loading, power capacity and wastewater limits before signing, and read the lease for what happens if the park’s own permits are amended. A factory that cannot discharge its process water, or cannot get fire approval for the racking layout, is a building rather than a plant.
Logistics, power and the cross-border supply line
A dual-site model creates a supply line that did not exist before, and the line needs its own documentation. Components leave China, enter Vietnam, are consumed in production, and the finished goods leave Vietnam for the customer. Each leg has a customs formalities set: export declaration in China, import declaration in Vietnam, and where Export Processing Enterprise status applies, the accounting and reporting that keeps the duty suspension valid. If the Vietnam entity is not an EPE, inputs are dutiable on import, and the duty becomes part of the cost base unless a drawback or similar mechanism applies.
Inland logistics decide how much inventory the model consumes. A northern Vietnam park is a truck journey from the Greater Bay Area, which is workable but not frictionless: border crossing times vary, and a component that misses a crossing stops a line. A southern park adds sea or air freight and several days. The cost model should carry the working capital of the extra pipeline, not just the freight rate.
Power reliability deserves the same treatment. Test and burn-in equipment draws heavily, and a park with an unreliable supply imposes generator cost, diesel storage and, in some cases, a compliance obligation on emissions. Ask for the park’s supply arrangement, the redundancy and the track record of outages before committing to a line layout.
The documents that make the cross-border line work are unglamorous: a supply agreement between the two group entities with delivery terms that match the customs treatment, a packing and labelling standard that satisfies both customs regimes, a bill of materials version that matches what physically crossed the border, and a reconciliation routine that ties the import entries to the production records. Where those four do not agree, an origin claim collapses.
IP enforcement across two jurisdictions
China offers patents, trademarks, trade secrets and criminal enforcement, with specialised intellectual property courts and customs recordation. For a hardware company, the practical value is speed: an injunction against a competitor selling a copied enclosure, or a criminal referral in a trade secret case where an engineer took drawings to a new employer. Customs recordation stops infringing goods at the border, which matters when the copy is made for export.
Vietnam’s IP Law provides registration and civil enforcement, but procedures are slower and damages lower. That does not make Vietnam lawless. It changes the economics: enforcement is a cost centre rather than a deterrent, and the return on a registration is realised through the ability to stop a shipment or support a negotiation, not through a damages award.
The operational consequences are the same in both countries, and they are contractual more than litigious.
Register before disclosure. A design disclosed at a trade fair, or in a supplier’s quotation pack, may lose novelty in jurisdictions where novelty is absolute. File the design and the mark in both countries before the first drawing leaves the building.
Record the rights with customs in both countries. A recorded right converts a border officer into an enforcement agent at no marginal cost.
Split the know-how. The Vietnam plant does not need the full process recipe for a component made in China. Send what the step requires, keep the rest with your own engineers, and treat the split as a design decision rather than an afterthought.
Own the tooling in writing. Tooling is the most litigated asset in Chinese manufacturing relationships, and the same disputes appear in Vietnam when a supplier holds a mould it paid a toolmaker to build. Title, bailment, marking and removal rights belong in the contract, with a schedule that identifies each tool.
Control secondment. Engineers who move between sites carry know-how in their heads. Secondment agreements need confidentiality, IP assignment, restrictions on what may be taken and a defined return obligation.
Enforcement across the border is not automatic. A judgment obtained in one country is not self-executing in the other, and the practical route for a group dispute is arbitration in a neutral seat, with the award enforced under the New York Convention, to which both China and Vietnam are parties. That choice should be made when the intra-group contracts are drafted, not when the dispute starts.
Export controls follow the technology, not the finished good
China controls exports under the Export Control Law and the dual-use regulations in force since December 2024, with licensing for listed items and end-user certificates. Critical minerals and some advanced technologies are restricted, and enforcement has become more active. The United States controls advanced computing chips, semiconductor equipment and certain technologies under the Export Administration Regulations, including entity-list and foreign direct product rules. AI hardware sits between both regimes.
A Vietnam factory does not remove either regime. China’s controls apply to the export from China, so a structure in which the Vietnam entity receives controlled Chinese technology, equipment or components, or in which Chinese staff transfer controlled know-how, can require a licence even though the finished product is assembled outside China. The same is true of the technology licence that permits the Vietnam plant to build to the Chinese design: China’s technology import and export rules may apply to that licence, and the royalty needs transfer pricing support.
Vietnam maintains a dual-use control list and licensing for certain goods, but its regime is lighter than China’s. That does not remove exposure to US, EU or UN sanctions, or to US controls that reach items made abroad with controlled technology or equipment.
The contractual response is to make classification somebody’s job. Require the supplier, and the group entity transferring technology, to state the classification of each item and each technology package, confirm licence status, provide end-use and end-user declarations, and notify any change. Include a right to suspend or terminate if a licence is refused. Screen customers, end uses and destinations on both sides. And map the technology flow before the structure is built, because a licence requirement discovered after the equipment has been shipped is a stranded asset.
Documenting the dual-site model
A China plus one structure is a group of contracts, not a corporate chart. The chart shows two entities; the contracts decide who owns what, who bears which risk, and what happens when one site fails.
| Document | Parties | What it fixes | What fails without it |
|---|---|---|---|
| Master supply agreement | Vietnam entity and China entity, or each entity and the customer | Price, volume, delivery, quality, liability | No enforceable allocation of risk between sites |
| Toll manufacturing agreement | Buyer and the processing entity | Who owns materials, who bears scrap and yield loss | Inputs and finished goods treated as the supplier’s stock |
| Technology licence | China licensor and Vietnam licensee | Scope, field of use, sublicensing, royalties, improvements | Technology transferred without title, and without export clearance |
| Technical data package licence | Design owner and each manufacturing site | Drawings, specifications, process parameters, updates | Each site builds to a different revision |
| Tooling bailment and ownership agreement | Buyer, supplier and, where relevant, the toolmaker | Title, marking, maintenance, insurance, removal | Tooling stranded at a factory that claims a lien |
| Quality agreement | Buyer and each site | Specifications, sampling, audits, traceability, warranty handling | Two sites, two quality standards, one unhappy customer |
| Secondment agreement | Sending entity and receiving entity | Payroll, tax, confidentiality, IP assignment, duration | Know-how moves without contractual cover |
| Intra-group services agreement | Group entities | Engineering, procurement, IT and finance support, cost allocation | Transfer pricing exposure and unrecorded related-party benefits |
| Origin support and audit cooperation agreement | Group entities | Who prepares and holds origin evidence, audit rights | Origin claim that cannot be defended at verification |
| Business continuity and allocation agreement | Group entities and customer | Capacity allocation during shortage, step-in rights, last-time buy | Customer learns about the single point of failure at the worst moment |
| Exit and requalification agreement | Buyer and each site | Transition assistance, data return, requalification support | Product cannot be moved without redesign |
Three drafting points decide whether the stack holds together.
The first is precedence. The supply agreement, the technical annex, the quality agreement and the purchase order will conflict. A clause that ranks them, and requires written change orders before any specification or process change, prevents the argument about which document governs a defect.
The second is the dispute route. Intra-group contracts should share one arbitration seat and one set of rules, so that a dispute involving both sites is heard once. Customer-facing contracts can use a different route, but the two should be checked against each other, because a customer claim that is arbitrated in one forum and an indemnity claim that is arbitrated in another creates duplication and inconsistent findings.
The third is evidence ownership. Origin files, quality records, tooling schedules and technology licences should each have a named custodian in the contract. In a verification or a dispute, the party that holds the records wins the argument, and in a two-site structure the records are usually split.
The sequence for standing up the second site
| Step | Responsible party | Output | Document |
|---|---|---|---|
| Fix the process split | Operations director with engineering | Step-by-step allocation between sites | Manufacturing plan |
| Model landed cost and origin | Finance with customs broker | Duty-inclusive cost per destination | Landed-cost model |
| Test the transformation step | Process engineering with customs adviser | Written origin analysis, ruling request where needed | Origin memorandum |
| Incorporate and file | Local counsel and corporate team | IRC and ERC, or WFOE registration | Constitutional documents |
| Secure land, power and permits | Project team with park and local counsel | Lease, environmental approval, fire approval, discharge permit | Permit set |
| Qualify the site | Quality with engineering | First-article approval, process capability data | Quality agreement and approval record |
| Paper the group structure | Legal with tax | Signed intra-group contract stack | Contract set |
| Prove the line | Operations and finance | Production records that match the origin file | Origin evidence file |
Where companies get this wrong
They treat assembly as transformation. The most common failure is a Vietnam plant that imports a near-complete product, adds a small number of parts and ships it as Vietnamese. The customs consequence is a transshipment finding, a higher rate applied retroactively, and a customer who discovers the problem when the goods are held. The fix is to design the process so that a genuine transformation happens in Vietnam, and to be able to prove it with routing, cost and labour records.
They move tooling without title. A buyer who has paid for a mould, never marked it and never recorded it in a schedule discovers at the end of the relationship that the supplier regards it as its own, or asserts a lien for unpaid processing fees. The tooling is then a negotiating chip in a dispute it was never meant to be part of. Title, bailment, marking and a removal plan belong in the contract before the first shot.
They transfer technology without a licence. A Chinese licensor grants its Vietnam affiliate the right to build to the design, and nobody checks whether the technology package is controlled or whether the licence must be registered. The consequence ranges from an unenforceable royalty to an export control breach. The check belongs at the design stage, when the technology package is being assembled.
They assume Vietnamese labour law is more flexible. Vietnam has lower wages and a structured Labour Code with a 48-hour week, overtime caps, insurance contributions, union requirements and severance. A ramp plan built on unlimited overtime fails at the first labour inspection, and a termination handled as if the employment relationship were at will produces a dispute that costs more than the severance saved.
They let the contracting entity drift from the operating entity. The customer contracts with the Chinese company, the invoice comes from a Hong Kong trading company, the goods ship from Vietnam, and the warranty is honoured by nobody in particular. When a defect appears, each entity points at another. Align the contract, the invoice, the exporter of record and the warranty provider, and say in the contract which entity is responsible for what.
They discover the permit sequence after signing the lease. A park that can host a warehouse cannot necessarily host a line with coating, plating or wastewater. The environmental assessment is written for the process the company described, and a later change of process requires an amendment that takes months. Confirm the process, the discharge and the power load with the authority before the lease is signed.
Cost and commercial impact
The costs of a dual-site model are mostly one-time and mostly invisible in a piece-price comparison.
Duplicate tooling is the largest single line for many programmes, because a second site needs its own moulds, fixtures and test jigs, even where the design is shared. Qualification adds engineering time, samples, reliability testing and, for regulated products, re-certification for the new manufacturing location. Certification is location-sensitive: a CE declaration or an FCC approval that names the factory may need to be reissued, and the technical file must reflect the new site.
Working capital rises with the pipeline. Components in transit, safety stock at the Vietnam plant and finished goods on the water all consume cash, and the longer the inbound leg, the larger the buffer. Engineering travel, dual quality systems and duplicated management add operating cost that a piece-price model does not capture.
Against those costs sit the tariff saving, the access to CPTPP, EVFTA and RCEP treatment, and the commercial value of a non-China source that some customers now require. Where Export Processing Enterprise status applies, the duty suspension on inputs is a real cash benefit, but it comes with record-keeping obligations that need a system rather than a spreadsheet.
The honest framing is that a dual-site model is a hedge with a running cost. It buys optionality and tariff diversification, and it charges duplication, complexity and a permanent compliance overhead. Companies that treat it as a cost reduction usually find the saving is smaller than modelled. Companies that treat it as risk management usually find the cost is bearable and the strategic value is real.
Practical checklist
- Write the manufacturing plan as a sequence of process steps, with a location and an owner for each.
- Tag every bill-of-materials line with local availability, tier and single-source status in both countries.
- Model landed cost per destination market using current tariff rates, confirmed with a broker.
- Draft the origin analysis before committing to the site, and obtain a ruling where the transformation is arguable.
- Confirm the transformation step is real: routing, cost, labour and scrap records must support it.
- Map technology, equipment and staff flows to Vietnam, and check whether a Chinese export licence is required.
- Confirm land use, power capacity, discharge limits and fire requirements with the authority before signing the lease.
- Cost labour including statutory contributions, overtime limits, training and the ramp curve.
- Register designs and marks in both countries, and record them with customs.
- Put the intra-group contract stack in place, with one arbitration seat and a named custodian for each evidence file.
A company that works through that list will usually find that the answer is not China or Vietnam but a defined split, with final assembly and test in Vietnam and the component, tooling and engineering base in China, held together by contracts that allocate title, risk and evidence. The firms that get this right treat the legal documents as part of the manufacturing design rather than as a closing formality.
Geason Law Offices advises foreign hardware companies on site selection, entity set-up and the contract stack for cross-border manufacturing. Our incorporation, commercial contracts and export controls teams work with the same file, so that the corporate structure, the supply contracts and the control analysis are consistent. For the operational detail behind this guide, see AI Hardware in China: Supply Chain Legal Basics, Vietnam’s AI Law and China Hardware Supply Chain.
Frequently Asked Questions↑ Back to article
Should we move AI hardware production from China to Vietnam?
Move only the steps Vietnam can perform competitively and document the transformation. Final assembly and test often move first, while precision components and tooling stay in China. The decision should follow tariff exposure, local content and your ability to prove origin.
What licences does a foreign company need to build a factory in Vietnam?
You need an Investment Registration Certificate and an Enterprise Registration Certificate, plus land or park lease documents, an environmental impact assessment or protection plan, a construction permit, fire approval and a discharge permit. Export Processing Enterprise status is separate and worth modelling.
Does assembling Chinese components in Vietnam avoid US tariffs?
Not automatically. US customs applies a substantial transformation test, and simple assembly of imported kits is often treated as transshipment, which carries a higher rate. Keep bills of material, cost records and production evidence, and obtain a customs ruling if the position is unclear.
Is Vietnamese labour law more flexible than Chinese labour law?
Vietnam has lower wage costs but a structured Labour Code with a 48-hour week, overtime caps, social insurance, union requirements and severance. China has higher employer social insurance costs and stricter termination rules. Neither is a low-compliance environment.
Do China's export controls apply to AI hardware made by our Vietnam factory?
They can, if the Vietnam entity receives controlled Chinese technology, equipment or components, or if Chinese staff transfer controlled know-how. China's Export Control Law and dual-use rules apply to the export from China, so map technology and component flows before restructuring.
How strong is IP protection in Vietnam compared with China?
China has stronger enforcement infrastructure, including specialised IP courts, criminal trade-secret cases and customs recordation. Vietnam offers registration and civil enforcement but slower procedures and lower damages. Register early and rely on contracts and operational controls in both.
What is the biggest legal risk in a China plus one structure?
Origin and technology transfer are the two pressure points. A structure that looks diversified but cannot prove substantial transformation risks retroactive duties, while undocumented licensing of Chinese technology can breach export control and transfer pricing rules.