Minus 7 and minus 2. Those are the two numbers, and the gap between them is the entire decoupling story.
From 2017 to 2024, China's share of bilateral US imports fell by seven percentage points. Over the same stretch, China's share of the value-added embedded in US imports — its parts, components, and content, wherever they were finally shipped from — fell by just two.
One number counts where the box was stamped before it crossed the border. The other counts where the stuff inside actually came from. Only the second one measures dependence, and it barely moved.
That is the whole case against calling any of this a decoupling. Peterson Institute researchers, examining value-added flows through 2024, put it flatly: US tariffs changed the routes Chinese content takes to reach America far more than they reduced America's reliance on China.
The mechanism is not subtle once you see it. Tariffs discriminate by shipping origin, not by content. A duty applies to a good declared as coming from China. It does not follow a Chinese capacitor into a Vietnamese assembly plant and out the other side as a Vietnamese export.
So the content did what content does under that rule. It moved one step upstream. Chinese goods increasingly reach the US embedded in imports from third countries — the same inputs, one relabeling away from the tariff.
This is why the bilateral line falls fast while the value-added line sags gently. You are watching an accounting migration, not an industrial one. The supplier of record changes; the supplier does not.
You are watching an accounting migration, not an industrial one. The supplier of record changes; the supplier does not.
If you want proof that firms treat this as a paperwork problem rather than a sourcing problem, look at 2025. That was the year US tariffs hit Chinese production of consumer electronics for the first time — the category everyone expected to crater.
Smartphones, laptops, monitors, video game consoles. US imports of them did not decline sharply.
They didn't decline because the multinationals had already built the escape hatch. Alternative supply chains in India and Vietnam were standing and ready, and the firms simply switched their sourcing of record to them. The rerouting wasn't a scramble; it was a pre-positioned pivot executed the moment the tariff landed.
That speed is the tell. You do not stand up a new manufacturing base in a quarter. You do reroute an existing one in a quarter — if the base is already there, staffed, and quietly humming, waiting for exactly this policy.
Contrast the case where firms didn't bother. Clothing and footwear importers largely left their chains alone, in part because the tariff gap between China and third countries on those products was too small to justify the reshuffle. When the arbitrage was there, firms took it instantly; when it wasn't, they ignored the policy. Either way, the tariff isn't reshaping the world — it's reshaping the customs declaration.
Worth flagging one genuine confound, because not every "shift away from China" is a shift away from China. Much of the longer-term move of US sourcing toward Taiwan and Mexico reflects the imported inputs feeding the AI data-center buildout — an appetite for GPUs and networking gear that has nothing to do with decoupling and would be happening regardless. Even the movement that looks like reshoring is partly just the map bending around a new demand center.
Now the part that matters for anyone footing the bill. If the content didn't move, the tariff's burden didn't land where its defenders promised.
The promise runs under the label "optimal tariff" — the idea that a large buyer like the US can use import duties to force foreign suppliers to eat the cost, shifting the tax burden abroad and coming out ahead. Kimberly Clausing and Maurice Obstfeld's verdict on this is blunt: the term is misnamed, and in practice these tariffs harm US interests rather than producing any net gain in national welfare.
The logic collapses precisely because the content never budged. An optimal tariff only extracts rent from foreigners if foreign supply actually contracts and prices adjust in America's favor. When the Chinese input simply reroutes through Vietnam at an unchanged — or higher — landed cost, there is no foreign concession to capture.
Instead the cost lands at home. Clausing and Obstfeld find US firms are harmed through higher costs of intermediate inputs — the parts and components American companies buy to make things. The tariff becomes a tax on domestic producers using imported inputs, not a lever against Beijing.
And the political prize? The blue-collar employment picture, the whole justification for absorbing the disruption, has been "disappointing at best." The jobs did not come, because the production did not come. What came was a longer, costlier route to the same Chinese content.
So tally it honestly. Seven years of tariffs bought a two-point reduction in real dependence, a cosmetic rerouting through India and Vietnam, higher input costs for US firms, and a jobs record that its own architects can't defend. That is a lot of friction to move a number that mostly measures paperwork.
Which brings us to the forward problem, and it's the sharpest one. The second Trump wave is being priced and sold against the same flattering bilateral figures — the minus-7 kind of number, not the minus-2 kind.
That means it is aimed at a target it cannot see through. A tariff written against shipping origin will, again, move the origin and leave the content. The 2025 electronics episode already ran the experiment and returned the answer: the chains are pre-built, the switch is instant, the dependence persists.
Repeating the first error would be a mistake. Repeating it after the value-added data are in, with the mechanism documented and the failure measured, is expensive theater by design.
The live test is India. It's in its own tariff standoff with Washington right now, and the lesson from China's seven-versus-two is directly on point: bilateral tariffs reroute supply chains through third countries far more easily than they decouple actual demand.
The real question in that standoff was never the headline rate. It's whether anyone negotiating it is measuring value-added content or just watching the bilateral ledger — whether, this time, someone has learned to count the thing that matters.
On the evidence so far, no one has. The map keeps getting redrawn. The dependence keeps not moving. And the two numbers keep telling opposite stories to people who only want to read one of them.