A foreign AI hardware company signing its first Chinese manufacturing agreement usually does so in a hurry. The product is designed, the launch date is fixed, and the factory has survived three rounds of samples. The contract arrives as a short template from the factory, in Chinese with an English courtesy translation, and the price schedule is attached to the back. Legal review is squeezed into the week before the deposit falls due.

Six months later the same company learns that the mould for its enclosure was paid for inside the unit price and therefore belongs to nobody in particular; that the firmware it supplied sits on a server the factory administers; that the acceptance clause says “meets buyer’s specification”, which the factory reads as “meets the drawing” and the buyer reads as “works in the field”; and that leaving means abandoning tooling that cannot be moved without the factory’s cooperation.

None of those outcomes requires bad faith. Each is the default result of a contract that described the transaction and did not allocate the things that decide what happens when the transaction goes wrong. Chinese manufacturing is not uniquely hazardous. It is simply fast, and the parties who move fastest tend to be the ones who have not written down who owns what.

The single most useful test of a manufacturing contract is not whether it covers the price and the delivery date. It is whether the buyer could take the product to a second factory next month without the first factory’s consent. If the answer is no, the contract has left the buyer with a supplier it cannot replace, and every later negotiation will be conducted on that basis.

What Chinese law supplies, and what it leaves to the parties

Chinese law treats a manufacturing arrangement in one of two ways, and the difference is not academic. Where the factory supplies labour and process against materials and specifications that you provide, the arrangement is a contract for work: the factory owes a result, and the default rules on inspection, risk and payment follow the law of contract for work. Where the factory also supplies the materials and sells you a finished product, the arrangement is closer to a sale of goods, and the default rules on delivery, risk transfer and conformity apply instead.

Most OEM programmes are a mixture. You supply the design, the firmware and some critical components; the factory buys the rest, assembles, tests and packs. Which default regime applies to which part of the performance is rarely obvious, and the contract should say. A clause that states “this agreement is a contract for work” is worth more than it looks, because it settles which set of default rules fills the gaps that the parties left.

Three further features of the legal environment shape the drafting.

First, there is no discovery. A dispute in a Chinese court or tribunal will be decided on the documents each side produces, and neither side is obliged to hand over its internal records. If you want the factory’s process records, test data or sub-tier supplier list, the contract must create the right to them. Asking after the dispute has begun is too late.

Second, the written contract is enforced but not in the way a common-law lawyer expects. Concepts such as consideration, privity and the parol evidence rule do not operate as they do in England or the United States. What matters is whether the obligations are clear, whether performance can be evidenced, and whether the remedy is one a Chinese court or tribunal can order. A clause that says “the parties will cooperate in good faith on tooling transfer” is a statement of intent. A clause that says the factory must release the tooling to a named carrier within a stated period after a stated event, and that a failure to do so attracts a stated daily sum, is an order the tribunal can make.

Third, the contract does not need registration to be valid, but several adjacent items do. A licence of technology inside the arrangement may attract technology import and export formalities depending on how the technology is classified. A transfer of patent rights or a design patent must be recorded before it binds third parties. Software copyright is better registered than asserted. None of these is a condition of the manufacturing contract itself, and all of them affect whether the buyer’s rights survive the relationship.

The practical consequence is that the contract is not a formality to be signed after the commercial terms are agreed. It is the instrument that converts a commercial understanding into something a tribunal in Beijing, Hong Kong or Singapore can enforce, and it should be drafted with that tribunal in mind.

Clause architecture: how the parts depend on each other

Most manufacturing contracts fail not because a clause is missing but because the clauses do not fit together. The scope says one thing, the quality clause refers to a specification that has been superseded, the purchase order terms say something else again, and the order of precedence is silent. When a dispute arises, each side quotes the document that suits it.

A workable architecture has a deliberate order.

Definitions and interpretation come first, because half the later disputes are about what a word meant. “Product”, “Specification”, “Tooling”, “Defect”, “Delivery”, “Acceptance” and “Confidential Information” each need a definition that is narrow enough to be useful. A definition of “Defect” that covers any deviation from the specification will produce arguments about cosmetic variation; one that distinguishes critical, major and minor non-conformity will not.

A document hierarchy comes next. The agreement, the statement of work, the technical specification, the drawings, the approved vendor list, the packaging specification and the purchase order will conflict at some point. The contract should rank them and should state that a purchase order term has no effect unless the agreement expressly incorporates it. Factory purchase-order templates routinely contain arbitration clauses, liability caps and quality terms that the buyer never read.

Scope of work follows, and it should be a list of activities rather than a sentence. Design for manufacturability review, tooling design and fabrication, sub-tier sourcing, incoming inspection, surface mount and assembly, in-circuit and functional test, burn-in, conformal coating, packaging, labelling, certification support, and logistics to the named place. Each activity should say who performs it, who pays for it and what evidence of completion is produced.

Change control is the clause that earns its place later. An engineering change note, a product change notification, a component substitution and a process change at a sub-tier supplier are four different events, and each has a different risk. The contract should require written notice before the change, a sample and test report where the change affects form, fit or function, and the buyer’s written approval before implementation. It should also state what happens when the factory changes a component without approval: the buyer’s remedies, including the right to reject the affected lots and to recover the cost of containment.

Tooling and materials come before quality, because they define what the factory is holding. Quality and acceptance define when the buyer must pay. Price and payment define the commercial mechanics. Intellectual property, confidentiality and non-circumvention define what the factory may do with what it learns. Warranty and liability define the consequences. Term and termination define how the relationship ends, and transition defines what happens in the ninety days after it ends. Dispute resolution defines where all of this is decided.

There is a case for splitting the package into three instruments rather than one: a development and tooling agreement, a supply agreement, and an NNN agreement. The split has a practical advantage. The tooling agreement and the IP assignment survive the termination of the supply agreement by their own terms, so the buyer does not lose title to the mould at the moment it loses the supplier. A single document can achieve the same result with a survival clause, but the survival clause is easy to forget and the separate instrument is not.

Scope, specification and the documents that decide

The scope clause is where optimism enters the contract. “The Supplier shall manufacture the Product in accordance with the Buyer’s specifications” reads as though it settles everything. It does not. It leaves open what “manufacture” includes, which revision of the specification applies, what happens when the specification is incomplete, and who bears the cost of making the product manufacturable.

A specification that has been written for a design team is rarely a specification that a factory can build to. It will omit tolerances, surface finish, colour matching under defined lighting, test limits, firmware loading sequence, serialisation and labelling. The gap is normally closed during the DFM review, and the DFM review is normally conducted before anyone thinks to record its outcome. The contract should require the DFM output to be issued as a controlled document, signed by both parties, and incorporated into the specification hierarchy. Otherwise the buyer discovers two years later that the agreed relaxation of a tolerance exists only in an email.

Revision control deserves the same attention. The contract should identify the applicable revision by number and date, require the factory to build only to the released revision, and require the factory to quarantine or rework material built to a superseded revision at its own cost if the change was implemented without approval. A factory that continues to build to the previous revision after a change notice has, in commercial terms, built scrap.

Where the buyer supplies components, the contract must deal with free-issue material. Who bears the loss in transit, who insures it, what scrap allowance is permitted, how the factory reconciles consumption against the bill of materials, and what happens to surplus material at the end of the programme. Free-issue material is a classic source of quiet loss: the factory consumes more than the theoretical quantity, attributes the difference to yield, and the buyer has no records to challenge it. A reconciliation clause with a stated frequency and a stated evidence standard closes the gap.

Tooling: title, possession and the right to walk away

Tooling is the most underestimated clause in the contract and the most expensive to fix afterwards. Moulds, dies, jigs, fixtures, test rigs, burn-in racks, programming fixtures and automated optical inspection programmes are all assets that the buyer usually pays for and frequently does not own.

The first question is who pays. There are two common structures. Under direct payment, the buyer pays the tooling cost as a separate line, usually against a tooling purchase order, and owns the tooling from the moment of payment. Under amortisation, the tooling cost is spread across the unit price over an agreed volume, and ownership transfers to the buyer when the amortisation is complete. Both are workable. The failure mode is a contract that says neither, and a factory that treats the tooling as its own capital equipment because it paid the toolmaker.

The second question is what “ownership” means in practice. In Chinese law, title and possession are distinct, and a factory in possession of tooling can create practical problems even where the buyer holds title. A factory that has not been paid for processing work may assert a lien over property in its possession in connection with that work. The lien can arise under the general law whether or not the contract provides for it. The way to manage the risk is not a clause that denies the lien, which the law may not respect, but a structure that keeps the tooling payments current, separates the tooling obligation from the supply obligation, and gives the buyer a contractual right to remove the tooling on notice that is exercisable before a payment dispute escalates.

The third question is identification. A tooling clause without a tooling register is a clause about an object nobody can find. The register should list each item, its serial number, its location, its replacement value, the date of acquisition and the ownership status. Photographs of the tooling, with the serial number visible, are worth keeping. Where the tooling is large, a plate riveted to the mould stating the buyer’s name and the purchase order reference is a cheap form of evidence.

The fourth question is maintenance and life. Moulds wear. The contract should state the expected number of shots or cycles, who maintains the tooling, who pays for refurbishment, and what happens when the tooling reaches the end of its life earlier than expected because the factory ran it harder than the parties contemplated. A factory that runs a buyer’s mould on a second customer’s product at night will consume its life faster than the buyer’s forecast assumes, and the contract should prohibit that use expressly.

The fifth question is transfer. The right to remove tooling is only useful if it can be exercised. The clause should name the circumstances in which removal is permitted, the notice period, the party that bears the cost of crating and transport, and the cooperation the factory must provide: releasing the tooling, the tooling drawings, the process parameters, the test programmes and the fixture documentation. A mould without its process settings and its gauges is a lump of steel. Where the factory has developed the process, the buyer should negotiate a licence to the process know-how for use at the next factory, or a documented transfer obligation, at the outset.

Two further items belong in the tooling clause because they are lock-in in another form. The first is the test programme: the automated test sequence, the limits, the fixture and the software that runs it. The second is the sub-tier supply chain: the anodiser, the connector supplier, the display module house. A buyer who owns the mould but cannot obtain the qualified components has not solved the problem. The contract should require the factory to disclose the sub-tier list for critical components and to assist in qualifying a replacement factory, and the NNN agreement should prevent the factory from using that list for its own products.

Intellectual property in an AI hardware programme

The IP clause is where a hardware contract differs most from a general manufacturing contract, because the product contains three kinds of rights that behave differently: the industrial design of the enclosure, the embedded software and firmware, and the model or algorithm that gives the product its function.

Start with the conventional division, because it remains the correct frame. Background IP is what each party brings. Foreground IP is what is created during the relationship. Joint IP is what both contribute to. Improvements are a separate category, because a factory that finds a cheaper way to assemble your product has created something, and the contract must say who owns it.

For an OEM programme the allocation is straightforward in principle: the buyer owns the product design, the firmware and the model; the factory owns its general manufacturing processes and techniques, but not any process developed specifically for the buyer’s product. That last qualification matters. Factories routinely claim that a fixture, a test method or a process improvement developed for one customer is part of their general capability, and therefore available to other customers. If the improvement was funded by the buyer or is specific to the buyer’s product, the contract should say so.

For an ODM programme the negotiation is harder, because the factory brings a platform. The contract needs to distinguish the platform IP, which stays with the factory, from the customisation layer, which should belong to the buyer, and it needs to grant the buyer a licence to the platform that is broad enough to survive a change of factory. The licence should be irrevocable, transferable to a successor manufacturer, and long enough to cover the product’s service life. A licence that terminates with the supply agreement gives the factory a veto over the buyer’s exit.

For a JDM programme, where both sides contribute engineering, the allocation should follow contribution and should be recorded milestone by milestone rather than left to a general clause at the end. A schedule that identifies, for each deliverable, the contributing party, the resulting rights and the licence granted to the other side is more useful than a paragraph that says the parties will jointly own what they jointly create. Joint ownership is a poor default in any legal system, and it is a particularly poor default where one party controls the manufacturing data.

Embedded software needs separate treatment. The factory will need to load firmware, run test scripts and sometimes modify configuration. The contract should grant a narrow licence for those purposes, prohibit reverse engineering and extraction, require the factory to return or destroy the toolchain at the end, and address the situation in which the factory writes part of the firmware. Where the factory’s engineers contribute code, the assignment should be express, and the moral rights waiver should be included where the law permits it.

The model raises questions that a conventional manufacturing contract does not anticipate. If the product collects data in the field and the data is used to improve the model, the contract must say who owns the data, who may use it for training, and whether the factory has any access. If the model is loaded onto the device, the contract should treat the weights as confidential information and should restrict their extraction. If the model is served from a cloud endpoint, the manufacturing contract is the wrong instrument for the model licence, and a separate software licence and data processing agreement is needed.

Registered rights should be recorded rather than assumed. A patent assignment that is not recorded at the patent office may not bind third parties. A design patent covering the enclosure should be filed before the product is exhibited at a trade fair, because the novelty of the design is destroyed by prior disclosure in most systems. Software copyright registration is not compulsory, but it converts a claim into a certificate, and the certificate is what a customs officer or a court will ask for first. Customs recordal of the registered rights allows the border authority to detain infringing shipments, which is often the fastest remedy available.

Trade secrets carry the practical weight, because most of what matters in an AI hardware product is not patented. The NNN agreement is the right instrument: non-disclosure, non-use and non-circumvention. The non-use limb is what prevents the factory from building a competing product from your design. The non-circumvention limb is what prevents the factory from going to your customers. Neither is effective unless the confidential information is identified, marked and access-controlled in practice, because a trade secret claim requires the owner to show that it took reasonable steps to keep the information secret. A signed NNN agreement with no operational discipline behind it is evidence of an intention, not of secrecy.

Chinese law also regulates the reward and remuneration of employee inventors. Where an invention is made by the factory’s employees in the course of their work, the right to apply for the patent may belong to the factory as the employer, and the employee is entitled to a reward. That is a matter between the factory and its staff, but it becomes the buyer’s problem where the buyer has commissioned the development and expects to own the result. The contract should require the factory to procure the assignment of employee inventions and to confirm that the reward obligations have been met, and it should require the factory to indemnify the buyer against a claim by an employee inventor.

Quality, acceptance and the inspection regime

Quality disputes are the most common manufacturing conflict, and most of them are caused by a contract that adopted a standard without defining it. “Industry standard quality” means nothing. “GB standard” means little more, because there are many GB standards and the relevant one depends on the product.

A workable quality clause has five layers.

The first is the standard itself. Where a national or international standard applies, name it and its number: the GB standard for the product category, the ISO standard for the management system, IPC-A-610 for the acceptability of electronic assemblies. Where no standard fits, the buyer’s specification does the work, and it must contain pass and fail criteria rather than descriptive language.

The second is the defect classification. Critical defects are those that create a safety risk or make the product unusable. Major defects are those that impair function or would cause a customer to return the product. Minor defects are cosmetic or otherwise tolerable within a limit. Each class needs its own acceptance level, and the sampling scheme should be stated: the AQL tables in ISO 2859-1, and their Chinese equivalent GB/T 2828.1, are the conventional reference. A clause that says “AQL 1.0” without saying which class it applies to is a clause that will be argued about at the first inspection.

The third is the inspection regime. Incoming inspection of components, in-process inspection at defined stations, final inspection before packing, and pre-shipment inspection by the buyer or its agent are four different controls, and the contract should say which party performs each, to what sampling plan, and what record is produced. The records matter more than the inspection. A factory that inspects without recording cannot demonstrate conformity, and the buyer cannot audit what was not recorded.

The fourth is the reference sample. A golden sample, signed and dated by both parties and held in a sealed state, resolves most cosmetic arguments. A limit sample set, showing the boundary of acceptable appearance, resolves the rest. Neither is a substitute for a specification, but both are faster to apply on a production line than a document.

The fifth is acceptance and deemed acceptance. The contract should state when the buyer is treated as having accepted a lot, what the buyer must do to reject it, and what happens if the buyer does neither. A clause that deems acceptance after a period without inspection is a trap for a buyer with a slow quality team. A clause that allows the buyer to inspect on arrival and to reject within a stated period is more balanced, and it should be paired with a provision that acceptance does not waive latent defects.

Post-delivery quality needs its own mechanism. The warranty clause covers the period after sale; the field failure process covers what happens when a unit fails in a customer’s hands. The contract should require a failure analysis report within a stated period, a corrective action report in the eight-discipline format that electronics manufacturers use, and a root cause determination that distinguishes design, component and process causes. The distinction matters because it allocates cost: a design defect is the buyer’s, a process defect is the factory’s, and a component defect may be the sub-tier supplier’s.

The cost of poor quality should be allocated expressly. Rework labour, scrap, expedited freight, containment at the customer, and the cost of a recall are all real costs that a quality clause can leave unallocated. A chargeback schedule that states the recoverable items and the evidence required is more effective than a general right to damages, because it can be applied to the next invoice without litigating.

Price, payment and the exporter question

The commercial terms deserve more drafting attention than they usually receive, because the payment structure is the buyer’s main lever during production and the currency structure determines who bears the exchange risk.

Unit price should be stated with its basis. Does it include packaging, testing, certification support and delivery to the port? Is it fixed for the term, or adjustable by reference to component cost, exchange rate or volume? A price adjustment clause that names an index and a review interval is workable; one that says the parties will discuss adjustments in good faith is not.

The minimum order quantity and the capacity reservation are two different commitments and are often confused. The MOQ is the smallest order the factory will accept. The capacity reservation is the volume the factory undertakes to be able to produce. A buyer that has negotiated an MOQ but no capacity commitment may find that the factory’s other customers have taken the line.

Payment terms are the buyer’s security. The conventional structure is a deposit against the order and the balance against shipping documents, and the common split is thirty per cent in advance and seventy per cent before shipment. That structure protects the factory and leaves the buyer exposed until the goods arrive. A buyer with leverage should push for retention against acceptance, a warranty holdback, or payment against a pre-shipment inspection certificate. Where the amounts are large, a letter of credit converts the factory’s credit risk into a bank’s, at a cost that should be priced into the negotiation.

Set-off is worth a clause. If the buyer has a quality claim, the ability to set it off against the next invoice is faster than a claim. The clause should permit set-off for liquidated amounts that are not genuinely disputed, and it should survive termination.

The exporter question is the one that foreign buyers most often overlook. If the factory sells to you FOB, the factory is the exporter of record, and the export declaration, the classification of the goods and any export licence are its responsibility. If you buy domestically and export through a trading company, the structure changes: the trading company becomes the exporter, the value chain includes an extra margin, and the export tax treatment differs. The choice affects the price, the customs documentation, and who is liable if the goods are misclassified. It should be made deliberately, with the factory and the freight forwarder in the same conversation.

Currency is the last item. A contract priced in RMB shifts the exchange risk to the foreign buyer; one priced in US dollars shifts it to the factory, which will price the risk in. A currency clause that states the currency of the contract, the currency of payment, and the mechanism for adjusting the price if the rate moves beyond a band is more honest than either party absorbing a risk it did not price.

Liability: what the contract must say because the law will not

Chinese law supplies a framework for damages, and the framework is not the same as the one a common-law lawyer expects. Damages are compensatory. Liquidated damages are permitted, and a court or tribunal may adjust an agreed sum that is excessively high or excessively low relative to the actual loss. The possibility of adjustment cuts both ways: it protects a buyer against a penalty that a tribunal considers unenforceable, and it weakens a buyer’s reliance on a large liquidated damages figure as a deterrent.

The contract should therefore do four things.

It should set a liability cap and state what the cap does not cover. The cap is usually expressed as a multiple of the charges paid in a defined period, or as a fixed sum. The carve-outs matter more than the cap: IP infringement, breach of confidentiality, wilful misconduct, death or personal injury, and product recall costs are the items a buyer should insist sit outside the cap. A factory will resist; the negotiation is worth having, because a cap that covers an IP indemnity is a cap that limits the value of the IP clause.

It should provide a specific indemnity for third-party IP claims. The indemnity should cover defence costs as well as damages, should give the buyer control of the defence with the factory’s cooperation, and should address the remedies available if an injunction is granted: procurement of a licence, modification of the product, or replacement. A general indemnity for “all losses arising from breach” is less useful than a specific one, because the specific one is easier to apply.

It should allocate product liability and recall costs. Under Chinese product quality law the producer and the seller can both be liable to an injured consumer, and the internal allocation between them is a matter of contract. Where the defect is a manufacturing defect, the factory should bear the recall cost; where it is a design defect, the buyer should. A clause that states the test and the process for determining the cause is more valuable than a clause that states a conclusion.

It should state the delay regime. Liquidated damages for late delivery, a right to cancel after a stated period of delay, and a right to recover the cost of expediting or of buying substitute goods are the three levers. The rate should be high enough to concentrate attention and low enough to be enforceable.

Consequential loss is the item that foreign buyers assume is excluded and Chinese factories assume is not addressed. The contract should state the position expressly, and it should carve out the losses that are the foreseeable consequence of the factory’s failure: the cost of a line stop, the cost of a customer’s chargeback, the cost of a launch delay. A blanket exclusion of consequential loss, drafted by the factory’s lawyer, will exclude exactly the losses the buyer cares about.

Termination, transition and the last-time buy

The end of the relationship is when the contract is tested. A contract that describes the beginning in detail and the end in a sentence has left the most dangerous part to the default rules.

Termination for cause needs a definition of material breach, a cure period, and a mechanism for determining whether the cure has happened. A cure period the parties choose, stated in days, is better than “reasonable time”, because the tribunal will otherwise decide. Termination for convenience needs a notice period, a cost consequence, and a clear statement of what the buyer must pay for work in progress and for material ordered that cannot be cancelled.

The transition provisions are the operational core. On termination or expiry, the factory should be obliged to deliver the tooling, the fixtures, the test equipment, the process documentation, the approved vendor list, the remaining free-issue material, the finished goods, the work in progress and the firmware build environment. It should be obliged to co-operate with a replacement factory for a stated period, to answer technical questions, and to permit the buyer’s engineers to observe the process. It should be prohibited from using the buyer’s design, tooling or sub-tier relationships for anyone else.

A last-time buy provision is worth negotiating at the start, when the parties are friendly. It gives the buyer the right to place a final order for a stated quantity of finished goods and spare parts before the line is shut, at a stated price. Without it, the buyer discovers at termination that the components are obsolete and the factory has moved on.

Escrow is the answer to the software problem. If the product cannot be manufactured without the factory’s test software or the factory’s build scripts, the buyer should hold those items in escrow with a release trigger: the factory’s insolvency, its refusal to supply, or the termination of the agreement. An escrow agreement with a named agent and a defined verification process is more useful than a promise to deliver on request.

Disputes: arbitration, interim relief and the parallel IP route

The case for arbitration holds. Awards travel under the New York Convention, the tribunal can be chosen for technical competence, the proceedings are confidential, and the language can be English. CIETAC, HKIAC and SIAC are the conventional choices, and for a Shenzhen-centred supply chain the Shenzhen Court of International Arbitration is also worth considering.

Three drafting points are often missed.

The first is the scope of the clause. A clause that covers “disputes arising under this agreement” may not cover a claim in tort for misappropriation of trade secrets, or a claim against an affiliate that is not a party. The clause should be wide, should name the entities it covers, and should state that the tribunal has power to grant interim relief.

The second is interim relief. Arbitration takes time, and a buyer facing a factory that is about to ship infringing product needs a fast order. Chinese courts can grant property preservation and behaviour preservation in aid of arbitration, and the availability and speed depend on the court and the evidence. The contract should permit either party to apply to a competent court for interim measures without waiving the arbitration. The buyer should also know, before the dispute, which court would have jurisdiction over the factory’s assets.

The third is the parallel IP route. A trade secret misappropriation or a patent infringement can be pursued through the civil courts, through the administrative enforcement route, and, in serious cases, through the criminal route. Each has a different evidentiary standard and a different speed. The civil route can produce an injunction and damages; the administrative route can produce a rapid seizure and a penalty; the criminal route can produce the pressure that settles a case. A buyer facing a factory that has copied the product should take advice on all three before choosing, because the choice determines what evidence must be preserved and how quickly.

Documents, filings and timeline

The table below sets out the sequence for a typical OEM programme. Timings are indicative and depend on the product, the port and the district; each should be confirmed locally before it is written into a launch plan.

StepAuthority or partyTypical timingWhat foreign companies miss
Company and factory due diligenceSAMR public records, site visitTwo to four weeksChecking the contracting entity is the entity that holds the licence and the premises
NNN agreement signedPartiesBefore any design disclosureDisclosing drawings before the NNN is in place
Development and tooling agreementPartiesBefore tooling is orderedFailing to separate tooling title from the supply terms
Tooling fabrication and first article inspectionToolmaker, factorySix to twelve weeksNot witnessing the first article or recording the process settings
DFM review and specification freezePartiesBefore pilot runLeaving the DFM outcome in email rather than a controlled document
Pilot run and reliability testingFactory, third-party labFour to eight weeksSkipping the reliability test because the launch date is fixed
Product certificationCCC, radio type approval or sector regulator as applicableVaries by productDiscovering late that the product needs approval before sale
Software copyright and design patent filingCPCC, CNIPAFiling early; grant takes longerExhibiting the product before filing the design
Customs IP recordalGACCAfter registrationRelying on the registration alone at the border
Production and pre-shipment inspectionFactory, buyer or third-party inspectorPer orderInspecting without a written sampling plan
Export declarationFactory or trading company, customsPer shipmentNot agreeing who is the exporter of record

Where foreign companies get this wrong

Five failure modes recur, and each has a consequence that is visible in the first year.

Using the home-country template unchanged. A contract drafted for a US or EU supply relationship assumes doctrines that do not operate in China and omits mechanisms that Chinese practice requires. The clause on remedies may be unenforceable; the clause on inspection may not match the way Chinese factories record quality; the dispute clause may name a court that cannot give an effective remedy. The template is a starting point, not a draft.

Treating the IP clause as a sentence. “All intellectual property arising from this agreement belongs to the Buyer” is a conclusion without a mechanism. It does not identify the background IP, does not deal with improvements, does not assign the employee inventions, does not record the assignment, and does not survive termination in express terms. The factory’s lawyer will read it as a statement of aspiration.

Discovering tooling ownership at the point of exit. The buyer decides to move to a cheaper factory, asks for the mould, and is told that the mould belongs to the factory because the cost was amortised into the price and the amortisation is not complete. The negotiation that follows is conducted with the buyer’s launch date as leverage for the factory. The time to fix this is before the first purchase order.

Accepting a quality clause with no defect classification. The first shipment arrives with cosmetic variation. The buyer rejects; the factory says the variation is within the drawing; the buyer has no limit sample and no AQL by class. The dispute is resolved by commercial pressure rather than by the contract, and the buyer usually loses because it needs the goods.

Choosing a dispute forum that cannot enforce. A clause naming the courts of the buyer’s home country is close to useless where the factory, the assets and the evidence are in China. A foreign judgment is difficult to enforce in China, and the factory knows it. The clause should choose arbitration with a seat whose awards travel, or a Chinese forum where the remedy requires a Chinese court.

Cost and commercial impact

The cost of a properly drafted manufacturing package is small relative to the exposure it addresses, but it is not trivial, and it is worth being specific about what drives it.

The drafting cost depends on the number of instruments, the complexity of the IP structure, and whether the product requires certification or sector approval. A single supply agreement with an NNN annex is a different exercise from a three-instrument package with a tooling schedule, a development schedule and an escrow. The review cost depends on whether the factory’s template is in Chinese only, whether there is an English version with a Chinese governing-language clause, and whether the parties intend to sign both.

The enforcement cost depends on the forum. Arbitration is more expensive than litigation in fees, and cheaper in time and confidentiality. A dispute that runs to an award and enforcement in China will consume management attention for a year or more, and the recoverable costs will not cover the whole expense. That is the reason to spend on the contract: the cheapest dispute is the one the contract prevents.

The commercial impact of a weak contract is easiest to see in the price of change. A buyer that can move its tooling and its sub-tier supply chain negotiates a re-quote from a position of credibility. A buyer that cannot move anything negotiates from the factory’s schedule. The difference is usually larger than the legal budget for the whole programme.

Practical checklist

The commercial logic of a Chinese manufacturing relationship is that the factory will always know more about the process than the buyer, and the buyer will always know more about the product than the factory. The contract’s job is to make sure that asymmetry does not become leverage at the moment the buyer needs to move. A buyer that owns the design, owns the tooling, holds the process documentation and can qualify a second source has a supplier. A buyer that does not has a dependency.

Geason Law Offices advises foreign AI hardware companies on manufacturing agreements, tooling and IP ownership, quality and acceptance regimes, and supply chain disputes in China. For the surrounding structure, the guides on the China hardware and supply chain, IP strategy for hardware companies in China and export controls for AI hardware cover the adjacent questions, and the manufacturing and supply chain service page sets out how we work on a programme of this kind.

Frequently Asked Questions↑ Back to article

What is the most important clause in a Chinese manufacturing contract?

There is no single most important clause, but IP ownership, tooling ownership, quality standards, and termination rights are the ones that most often cause disputes. For ODM relationships in particular, IP ownership is the critical clause — if it is not clearly specified, you may end up in a situation where the factory claims partial ownership of your product design.

Should manufacturing contracts be under Chinese law?

Yes, generally. If you are manufacturing in China, the factory is in China, the goods are in China, and performance happens in China — Chinese law will apply regardless of what your contract says in many cases. You are better off drafting the contract under Chinese law with Chinese dispute resolution, rather than having a foreign law contract that may be difficult or impossible to enforce in China.

Who owns the tooling and moulds?

This depends on what your contract says. In China, if you pay for tooling directly, you should own it. If tooling cost is amortised into the unit price, ownership may transfer to you once you have paid a certain number of units or the tooling cost has been fully amortised. The biggest mistake is not addressing tooling ownership in the contract at all — then you are at the mercy of the factory when the relationship ends.

Can I use my home country NDA with Chinese factories?

You can try, but it is unlikely to be effective. NDAs need to be drafted under Chinese law and include Chinese-law specific provisions to be enforceable in China. Generic Western NDAs often have gaps that make them unenforceable or impractical to enforce in China. For example, the remedies, burden of proof, and enforcement mechanisms are different under Chinese law.

What is the difference between an NDA and an NNN agreement?

An NDA (Non-Disclosure Agreement) only protects confidential information. An NNN agreement (Non-Disclosure, Non-Use, Non-Circumvention) goes further: it also prevents the supplier from using your ideas/designs for their own products (non-use) and from going around you to contact your customers (non-circumvention). For manufacturing relationships in China, NNN agreements are strongly recommended over basic NDAs.

What dispute resolution mechanism works best for Chinese manufacturing contracts?

Arbitration is generally preferable to court litigation for cross-border manufacturing disputes because arbitration awards are more easily enforceable internationally under the New York Convention, you can choose arbitrators with industry expertise, proceedings are private, and the process is often faster than Chinese courts for complex commercial cases. Common choices include CIETAC in China, HKIAC in Hong Kong, and SIAC in Singapore. Arbitration can also be conducted in English, which is important for foreign parties.

What quality standards should be included in a manufacturing contract?

A good manufacturing contract should specify industry standards such as GB standards for China or ISO and IPC standards, your specific product specifications with detailed pass/fail criteria, acceptable quality level (AQL) for defects including critical, major, and minor classifications, inspection procedures and stages (IQC, IPQC, FQC, OQC), golden sample requirements, and corrective action procedures for quality issues. Quality disputes are the most common type of manufacturing conflict, so specificity here prevents costly disagreements later.

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