A few years ago, building consumer hardware in China, we found a cosmetic defect on a moulded part three weeks before a launch window. Not a functional failure, just a witness line in the wrong place on a surface the customer would touch. We drove forty minutes to the tool shop, sat with the toolmaker, agreed the change, and had corrected parts back on the line inside four days. Nobody wrote a change order. Nobody booked a flight. The cost of that fix was roughly the price of a decent dinner for eight people.

I think about that week every time an executive shows me a sourcing comparison where Vietnam beats Guangdong by 18% on unit cost. The spreadsheet is almost always correct and almost always irrelevant. It is measuring the wrong thing, because what you buy when you manufacture in Shenzhen or Dongguan is not a labor rate. It is a 50-kilometer radius.

The labor rate was the smallest line on the bill

Start with the arithmetic, because this is where the conversation usually ends before it has begun.

On most consumer electronics and small appliances I have worked with, direct labor sits somewhere between 5% and 12% of factory gate cost. Materials, components and tooling amortisation eat the rest. So when a sourcing team tells me a Vietnamese or Indian plant pays 60% of the Guangdong wage, they are describing a saving of maybe four points of cost of goods. Real money at volume, I will not pretend otherwise. But four points is a thin cushion, and here is what it has to absorb.

It has to absorb the revision cycles. In the radius, a tooling change is a car ride and a conversation. Outside it, the mould is either shipped back to China (two to four weeks each way, plus the risk of damage in transit) or modified locally by a shop that has never run your resin at your cycle time. A two-week iteration becomes a six-week iteration. If your product needs three revisions before it is stable, you have just moved your launch by a quarter.

It has to absorb quality escapes. A plating line that runs slightly out of spec does not announce itself. It shows up as field failures eight months later, and by then you have shipped 40,000 units. I have seen a single cosmetic finish problem generate rework costs that wiped out a full year of the labor saving that justified the move.

It has to absorb air freight. When the ramp slips, you fly. Air is roughly ten to fifteen times the cost of ocean per kilogram, and a few pallets of airfreighted product will erase the annual saving on those units in a single booking. Everybody knows this. Nobody puts it in the comparison.

And it has to absorb the boring physical realities. In the summer of 2023, industrial users in northern Vietnam were rationed on power. If your line stops for three days in peak build season, no wage differential saves you. In parts of India, you plan around the grid as a standing assumption and price in generation.

None of these appear in the sourcing comparison, because the sourcing comparison is built from quoted piece prices and quoted wages. The hidden costs of relocating production from China to Vietnam are not hidden at all. They are simply in a different column, owned by a different department, and they arrive later.

What you actually bought in Dongguan

Here is the thing that took me years to articulate properly. Your contract manufacturer is not your supply chain. Your contract manufacturer is the visible tip of thirty to eighty tier-2 and tier-3 suppliers that you have never visited and often cannot name: the connector house, the plating line, the anodiser, the pad printer, the foam die cutter, the flex circuit shop, the guy who makes the fixture that holds your part while the fixture that holds the other part is being repaired.

In the Pearl River Delta, most of those sit within a day’s drive. Many sit within an hour. That geography is not a convenience, it is the product. It means a problem discovered at 9am can have three quotes by lunch and a sample by Thursday. It means when your engineer describes a tolerance issue, the process engineer across the table has solved a version of it forty times, for forty other customers, and offers you the answer before you have finished the sentence.

That accumulated pattern recognition is the asset. It is also the part nobody invoices you for, which is exactly why it disappears from the analysis.

Relocation does not move a factory. It moves you outside the radius. The new plant may be excellent. The building may be newer, the floor cleaner, the management more attentive because you are a marquee customer rather than one account among two hundred. But the ecosystem around it is thinner, and every gap in that ecosystem converts into either a longer lead time, a lower first-pass yield, or a plane ticket.

The operator’s test

So I would retire the question “where is the piece part cheaper” and replace it with two questions.

First: how many of my tier-2 and tier-3 suppliers exist within one day’s drive of the proposed site? Not “can be sourced from”, which is a euphemism for “will be trucked or shipped in from China anyway”. Physically present, with a plant you could visit tomorrow morning. Make your CM produce the list. Count it. If the honest answer is that 70% of your sub-tier content still originates in Guangdong and gets kitted into Vietnam for final assembly, you have not diversified your supply chain. You have added a border crossing, an inventory buffer and a rules-of-origin argument with customs to a supply chain that remains exactly as concentrated as it was.

Second: who pays for the learning curve? Somebody always does. In a mature ecosystem, the supplier absorbs most of it, because they already climbed that curve on someone else’s product. In a greenfield, you pay it as tuition: your scrap, your yield ramp, your engineers’ flights, your launch slip. In my experience you should budget eighteen to twenty-four months and something like 1.5 to 2 times your planned transfer cost to reach parity on yield and cycle time. If the business case only works on day one pricing, it does not work.

Running two sites on purpose

None of this is an argument for staying put. Concentration risk is real, tariffs are real, and a board that asks for geographic optionality is asking a reasonable question. The answer is a deliberate two-site structure rather than a migration.

What should move first is labor-heavy, low-tolerance assembly with a stable bill of materials: cable and harness work, sewn and soft goods, mechanical subassembly with generous tolerances, final pack-out, kitting, anything where the process is mostly hands and the drawing has not changed in two years. Pick your most mature, highest-volume, most boring SKU. Not the new one. The temptation is always to launch the new product at the new site so the numbers look good from the start, and it is the single most reliable way to lose both.

What should not move is anything that requires fast iteration on tooling. Multi-cavity precision moulds, tight-tolerance parts with cosmetic surfaces, multi-step metal finishing, and above all new product introduction. Keep NPI inside the radius. Let the engineering-intensive work sit where the engineers are, then transfer processes outward only once they are frozen and documented to the point where a competent stranger can run them.

Practically: own your tooling and know where every mould sits. Pay for duplicate tools on the parts you cannot afford to have stranded, and accept that duplication costs real money and buys real optionality. Put a resident engineer of your own at the second site for the first year, not a visiting auditor. Run the second site at 20% to 30% of volume and hold it there deliberately, even when it is more expensive per unit, because that is what the insurance premium looks like. Measure the yield gap and the cash conversion cycle monthly, not the quoted unit price.

And be honest internally about what you are buying. You are buying resilience and tariff optionality, not savings. If you tell the board it is a cost programme, the first quarter of bad yield numbers will kill it.

Why nobody announces the return

A pattern I have watched with some interest: companies that made the move loudly, then moved volume quietly back.

The announcement gets its news cycle. The reversal gets nothing, because there is no upside in publishing it. So the second site stays open, runs a token line, keeps the geography on the slide, and the real volume drifts back to where the tool shop is forty minutes away and the plating line answers the phone. I do not read that as failure. I read it as a company discovering the actual shape of its cost structure and responding rationally, if not publicly.

The lesson worth taking is not that moving manufacturing out of China is a mistake. It is that a china plus one strategy is an ecosystem decision dressed up as a procurement decision, and the two have different owners, different time horizons and different arithmetic. Procurement optimises a quoted price. Operations pays for the radius.

Before you approve anything, go walk the new site, then get in a car and drive for an hour in each direction, and count what you find. That drive will tell you more than the spreadsheet will.

I write from twenty years of building businesses between Europe and Asia. If your company is facing this, start a conversation.