Peak Globalisation — The Reshoring That Never Happened

The totals are where the deglobalisation argument fails first. World trade in goods and services grew about 4.7% in volume in 2025 while global GDP grew 2.9%, which is the arithmetic of deepening integration rather than retreat [3]. Merchandise trade volume alone rose 4.6% [1]. In the third quarter of 2025 the dollar value of world merchandise trade reached an all-time high, up 7.5% year on year [2]. Across the first nine months of that year volumes rose 4.5% and values 6.5% against the same period of 2024 [2]. Not one of those series turned down.

The physical measures agree, and they are harder to argue with. Global container volumes reached a record 192.9 million twenty-foot equivalent units in 2025, up 4.7% from 184.3 million a year earlier [25]. August 2025 set a single-month record of 16.61 million TEU [25]. Chinese ports moved 350 million TEU, up 6.8% [26]. Container throughput is the least manipulable indicator in trade statistics: a box either crosses a quay or it does not, and the long series shows only two down years since 1980, in 2009 and 2020 [44].

Composition is where the change shows. AI-related goods made up roughly 15% of world trade in 2025 but accounted for 42% of its growth [2]. The McKinsey Global Institute puts exports of AI-related goods up nearly 40% in 2025 and credits them with about one third of total trade growth [18]. Semiconductors, data-centre equipment and the components behind them are doing work in the headline number that the rest of manufacturing is not. Strip them out and the aggregate looks far less like a boom and far more like the flat decade that preceded it.

The forecast is where the turn appears, and it is a turn in growth rather than in level. The WTO expects merchandise trade growth to fall to 1.9% in 2026 from 4.6% in 2025, with services easing to 4.8% [1]. Goods and services together are projected to grow 2.7% against global GDP growth of 2.8% [1]. That convergence matters: it is the first year in which trade is not expected to outrun output. Director-General Ngozi Okonjo-Iweala framed the same numbers as resilience, buoyed by high-technology products and digital services [1].

Regional dispersion is wider than at any point since the pandemic. Over the first nine months of 2025 Asian export volumes rose 9.5%, African 6.1% and South and Central American 5.7%, while European exports fell 0.3% and CIS exports 1.7% [2]. The 2026 forecasts continue the split: Asian imports up 3.3%, African 3.2%, European 1.3%, North American 0.3% and CIS down 2.0% [1]. A single global trade cycle no longer describes what is happening to any particular region.

The long series puts all of this in proportion. The world trade-to-GDP ratio climbed from about 25% in 1970 to roughly 61% in 2008 and has not regained that level since [41]. That flattening is the factual core of the peak-globalisation claim. But Richard Baldwin has shown that the single peak is an artefact of adding four very different series together: China peaked in 2006, the United States in 2011, Japan in 2014, and the European Union shows no clear peak at all [20].

So the honest summary of the totals is narrow and specific. The trade-to-GDP ratio has been flat for roughly fifteen years, absolute volumes keep setting records, container throughput crossed its previous high again in 2025, and the policy environment is at its most restrictive since the 1940s [15][28]. Nothing in that combination describes a world trading less. It describes a world trading the same amount through different intermediaries, under legal arrangements that change several times a year.

The bilateral reallocation is the largest and fastest peacetime shift in a major trading relationship on record. Laura Alfaro and Davin Chor put China's share of US goods imports at about 21% in 2017, 13% by the end of 2024, and roughly 9% by August 2025, a level last seen around the year China joined the WTO [9]. Between March and August 2025 alone the share fell some four percentage points [9]. The speed is what distinguishes this episode; the direction had been set for seven years before it.

The customs data from the year after the April 2025 tariffs are starker still. Atradius records China's share of US imports falling from 13.4% in 2024 to 7.5% twelve months after the Liberation Day announcement, with Chinese machinery and electrical equipment shipments at less than half their 2024 level [27]. Vietnam gained 2.4 percentage points to 6.7% and Taiwan 5.2 points to 8.8% [27]. Mexican imports ran about 5% above 2024 overall and 32% above in machinery, making Mexico the largest single machinery supplier to the United States [27].

At corridor level the same picture holds. The McKinsey Global Institute finds the US-China trade corridor shrank by around 30%, redirecting more than 165 billion dollars of trade to other routes, while goods trade overall grew about 6.5% in 2025 [18]. Its conclusion is that trade is being reshaped along geopolitical lines rather than reduced, with countries trading more with geopolitically aligned partners and less with distant or politically sensitive ones [18]. That is a statement about direction, not about volume.

The concentration of the gains is the detail that gets lost. Only a handful of economies absorbed almost all of the displaced trade, and they were already large suppliers. Kearney records China falling below 10% of US manufacturing imports, from 20% four years earlier, a loss of 135 billion dollars, while the other thirteen Asian low-cost countries in its index gained 193 billion [7]. The arithmetic is unambiguous: the region kept the business, and one country inside it lost share to its neighbours.

Europe experienced a mirror image of the same process without choosing it. The EU trade deficit with China reached 359.9 billion euros in 2025, up 2.7% [30]. German goods exports to China fell 9.3% to 81.8 billion euros, the lowest in a decade and 23% below the 2022 peak, and in the first half of 2026 they fell a further 12% to just under 37 billion [32]. German car exports to China dropped 66% between 2022 and 2025, to their lowest level since 2009 [31].

The European case shows how much of the rewiring is demand-side rather than policy-driven. China overtook the United States as Germany's largest trading partner in 2025, but the composition of that relationship inverted: Germany now buys far more than it sells [32]. The Federal Reserve's analysis of Chinese industrial policy identifies exactly this pattern across advanced economies, with Germany suffering the most pronounced market-share losses because of its exposure to sectors where Chinese exports surged [11].

Put the pieces together and the shape of the change is clear. One bilateral corridor contracted violently, several others expanded to absorb it, one advanced economy lost export share in the products it had built its industrial model around, and the world total rose. None of that is deglobalisation. It is a redistribution of who stands between the factory and the customer.

Mary Lovely and Christine Wan traced Chinese value added through the composition of US imports using Asian Development Bank multiregional input-output tables covering 2007 to 2024 [6]. The method answers a different question from customs data. Customs records where a shipment was loaded; input-output tables record where the value inside it was created. Over 2017 to 2024 China's share of bilateral US imports fell seven percentage points while its share of the value added embedded in those imports fell about two [6].

The gainers in the customs data are exactly the economies most integrated with Chinese production. Taiwan added 4.1 percentage points of US goods imports between 2017 and 2025, Vietnam 3.7 and Mexico 2.3 [6]. Each of those three runs a manufacturing base that buys intermediate goods from China. Their gains in the bilateral statistics therefore overstate, by construction, the amount of Chinese content that left the American supply chain.

The Vietnamese case has now been measured at firm level, and the result is unusually clean. Economists at the Federal Reserve Board find that Chinese-owned firms raised their share of Vietnam's exports to the United States from 11% in 2018-19 to 25% in 2020-23, while Vietnamese domestically owned firms fell from 33% to 23% [12]. The number of Chinese-owned firms operating in Vietnam rose from 1,287 in 2018 to 2,864 in 2023 [12]. The production moved; the ownership largely did not.

The input side moved even less. Those same Chinese firms in Vietnam doubled their share of Vietnam's imports from China, from 11% to about 22% [12]. Newly established foreign firms remained heavily reliant on Chinese inputs after relocating [12]. The Federal Reserve authors draw the obvious conclusion: the rapid expansion of Vietnamese exports to the United States does not imply that the gains accrued primarily to Vietnamese-owned industry [12].

That last finding deserves restating because it inverts the policy intuition. To take export share from China in a strategic product, a country generally has to buy more from China, not less [10]. The supplier network is the asset being competed over, and access to it runs through Chinese component makers. A tariff aimed at the final assembly step leaves that structure intact and simply relocates the last stage to a jurisdiction with a lower rate.

Outright rerouting exists, and its scale is disputed. Academics at Harvard, Duke and Academia Sinica estimate more than 8 billion dollars of Chinese exports were rerouted through Vietnam to the United States in the first three quarters of 2025 [35]. A White House report puts the figure across all hubs, including Mexico and India, at roughly 67 billion for the year [35]. Against total US goods imports those are meaningful but not dominant numbers ⚖ Contested.

The disaggregated evidence cuts against the maximal version of the transshipment claim. CSIS finds four electronics product categories accounting for 87% of the growth in Vietnamese exports, while apparel, footwear and toys, the sectors most associated with simple relabelling of Chinese goods, barely moved [36]. Firms in Vietnam increased their imports of Chinese intermediate goods by 85% rather than replacing Chinese suppliers [35]. That is deep integration being reported honestly, not fraud being concealed.

The global aggregate for value chains confirms that nothing structural has yet broken. Global value chains carried 46.3% of world trade in 2024, against a peak of 48% in 2022 [4]. The share held by the ten most integrated economies fell from 76% in 2010 to 64% in 2024, which is participation widening rather than chains shortening [4]. Nearly two thirds of global trade still takes place inside value chains of some kind [39].

There is no contradiction between the two sets of numbers, and it is worth being precise about why. The bilateral collapse is real and large. The value-added shift is real and small. Both are correct measurements of different objects, and the entire reshoring claim depends on quietly substituting the first for the second.

The index is deliberately narrow, and that is its value. It tracks the manufacturing import ratio, defined as manufactured goods imported from fourteen Asian low-cost countries and regions expressed as a share of US domestic manufacturing gross output, and reports the year-on-year change in basis points [8]. A negative reading means imports grew faster than domestic production. The index has now been negative for consecutive years through the most aggressive tariff programme since the 1930s [7].

The 2026 reading improved without reversing. It moved from minus 115 to minus 91 while US manufacturing imports rose 4.6%, reaching a four-year high [7]. Certain product categories representing about 40% of Asian imports show small signs of reshoring, which is the strongest positive finding in the exercise [7]. Patrick Van den Bossche, the partner who leads the study, described most product categories as starting to rely less on imports but not yet enough to conclusively state that the corner has been turned [7].

The composition of the import growth explains most of the headline. Computer and electronics imports rose 29% in 2025 while domestic output in the same category grew 2.8% [7]. Two categories, computer and electronics plus apparel and accessories, account for 44% of all Asian imports into the United States [7]. The surge that keeps the index negative is concentrated in precisely the sector where AI hardware demand is running hardest [2].

Mexico offers the clearest test of nearshoring as an investment thesis, and the result is mixed at best. The country attracted a record 40.87 billion dollars of foreign direct investment in 2025, up 10.8%, but nearly all of it was reinvested earnings rather than new capital, and fresh investment fell 13% year on year in early 2026 [34]. Total domestic investment fell about 10% in 2025, and gross fixed investment fell 3.6% year on year in February 2026 [34].

The constraints are physical and institutional rather than commercial. The Dallas Fed identified electricity generation and water supply bottlenecks as limits on deeper integration into global value chains, alongside rule-of-law concerns following judicial reform [33]. Those are decade-scale problems. A manufacturer choosing a site in 2026 is pricing grid capacity in 2031, and Mexican industrial electricity is not currently a solved question [33].

Global investment data tell the same story at a higher level. UNCTAD reports global foreign direct investment up about 14% in 2025 to an estimated 1.6 trillion dollars, with the growth concentrated in developed economies and heavily shaped by flows through financial centres [13]. Underneath the headline, greenfield project announcements fell 16% and their announced value fell by almost one third [14].

Where greenfield capital did go is the decisive detail. Data centres accounted for more than one fifth of global greenfield project value in 2025, above 270 billion dollars, while greenfield investment in manufacturing outside strategic sectors declined [14]. The largest industrial investment wave in a decade is building compute, not import substitution. Those two things are routinely reported under the same heading and they are not the same activity.

None of this means the investment was wasted or that the capacity will never arrive. It means the timeline being claimed is not the timeline being built. Capacity growth of 1.5% over four years cannot displace an import base that grew 4.6% in a single year [7]. Until those two rates cross, the reshoring claim is a forecast rather than a finding.

Start with the concentration that the reallocation was supposed to address. China accounts for roughly 30% of global manufacturing value added and recorded a trade surplus of about 1.19 trillion dollars in 2025, the first time any country has passed one trillion [29][43]. The Federal Reserve puts that surplus above 6% of China's own GDP and above 1% of global GDP, roughly double the peak surpluses achieved by Japan or Germany at their most export-intensive [11].

European dependence is concentrated in exactly the places a tariff cannot reach. A European Commission study found that of 204 goods on which the bloc is dependent, one third of the sixty-four most critical came from China [30]. China supplies 98% of EU solar panels and 54.4% of its machinery and vehicles [30]. The Commission has declared the trade relationship unsustainable and is assembling a de-risking toolkit of subsidy probes, trade defences, export controls and outbound investment review [30].

The breadth of the Chinese position is what makes it durable. Federal Reserve researchers find China gaining global market share across nearly all manufacturing sectors over the past decade, from apparel and textiles to vehicles and electronics, while advanced economies lost share broadly [11]. Imports lagged exports badly: commodities now make up 44% of Chinese imports, and manufactured goods from advanced economies grew much more slowly [11].

The link to policy is measurable rather than inferred. Sectors with higher policy intensity showed faster export growth between 2017 and 2024 and larger increases in trade balance, and the fifteen most policy-supported sectors accounted for 76% of the aggregate surplus increase [11]. Computing machinery, pharmaceuticals, motor vehicles and electronic components received the most interventions [11]. Whether industrial policy works in general remains contested; in this case the correlation is unusually tight ⚖ Contested.

The enforcement problem is the most immediately practical of the five. Determining whether a good shipped from a connector economy is substantially transformed there or merely relabelled requires firm-level, product-level origin auditing across dozens of jurisdictions. The estimates currently in circulation, from 8 billion dollars through one country to 67 billion across all hubs, differ by nearly an order of magnitude precisely because nobody has that data [35].

The capacity gap compounds the enforcement gap. If the domestic alternative does not exist, tighter origin rules raise costs without changing suppliers, because the importer has nowhere else to go. That is the mechanism by which a policy designed to relocate production instead relocates paperwork. Kearney's finding that imports rose 4.6% in the year the index improved is what this looks like in aggregate [7].

Legal instability is a cost in itself, independent of the rate. Between February and May 2026 the United States lost its primary tariff authority at the Supreme Court, replaced it within four days under a different statute, lost that too at the Court of International Trade, and had the judgment stayed on appeal [16][17]. A firm choosing a supplier for a five-year contract is pricing a legal process, not a tariff schedule.

The macro cost estimates remain the weakest link in the chain of evidence, and they should be presented that way. IMF staff put long-run losses from fragmentation between 0.2% and almost 7% of global output, a range wide enough to accommodate almost any policy conclusion [19]. The upper bound is roughly the combined annual output of Germany and Japan, and the losses fall hardest on smaller and middle-income economies with the least capacity to absorb them [42].

The policy series has moved faster than any trade series it was meant to affect. Global Trade Alert and the WTO trade monitoring database both record new restrictions reaching record highs across 2023 to 2025, reversing decades of gradual liberalisation [28]. More than 2,500 restrictions were imposed in the first ten months of 2025 alone, almost five times the count for the same period a decade earlier [28].

The joint WTO and IMF trade policy activity index puts the acceleration in historical context. Averaged over January to May 2026 it ran nearly twice its 2024 level and about a quarter above its 2025 average, a new high for a series whose previous peaks were the 2018-19 tariff escalation, the 2020 pandemic response and the 2022 energy shock [5]. The index describes the most pronounced rise in policy activity since the aftermath of the global financial crisis [5].

Tariff levels followed. The Yale Budget Lab measured the pre-substitution average effective US rate at 16% before the IEEPA tariffs were struck down in February 2026, the highest since 1936 [15]. By 8 April 2026 it stood at 11.8%, the highest since the early 1940s [15]. The projected end-2026 rate is 9.7% if the Section 122 surcharge expires and 12.2% if it is made permanent [15].

The legal ground shifted underneath all of it. On 20 February 2026 the Supreme Court held 6-3, in an opinion by Chief Justice Roberts in Learning Resources v. Trump, that the International Emergency Economic Powers Act does not authorise the president to impose tariffs [16]. The ruling invalidated tariffs collected over the preceding year and opened refund claims for importers across the entire affected period [16].

The replacement arrived in four days. On 24 February 2026 a 10% global surcharge took effect under Section 122 of the Trade Act of 1974, an authority limited to 150 days and conditioned on the existence of fundamental international payments problems [17]. On 7 May the Court of International Trade held that the statutory condition was not satisfied; on 12 May the Federal Circuit stayed that judgment, allowing collection to continue pending appeal [17].

Underneath the tariff headlines a quieter reordering is under way through bilateral instruments. The WTO counts more than 185 targeted trade deals signed by the end of 2024, of which only 55 official texts are publicly available [4]. Around 80% of mineral-related deals were concluded after 2022, and critical-mineral agreements raised trade values between their signatories by roughly 12% [4].

The European toolkit is being assembled on the same schedule. Brussels has moved from declaring the relationship with China unsustainable to a set of concrete instruments: foreign-subsidy investigations, strengthened trade defences, the anti-coercion instrument, tighter export controls and a push to review outbound investment in semiconductors, artificial intelligence and quantum computing [30]. Measures forcing supply-chain diversification have been floated for chemicals, metals and clean energy [30].

UNCTAD's read of the same period is that slower growth, rising protectionism and structural shifts in value chains are jointly redefining trade flows, with diversification strengthening resilience while reducing efficiency [39]. That trade-off is real and it is not free. Diversification bought with duplicated tooling and shorter production runs shows up later as higher unit costs, and those costs land on importing economies.

The pattern across the whole period is a policy apparatus moving faster than the productive base it is trying to redirect. Restrictions can be legislated in weeks, struck down in months and replaced in days [16][17]. Factories take five years. The mismatch between those two clock speeds is the central operational fact of the current trade regime.

The peak case is straightforward and rests on a real series. The world trade-to-GDP ratio reached roughly 61% in 2008 and has not recovered it in eighteen years [41]. Global value chain participation has slipped from 48% of world trade in 2022 to 46.3% in 2024 [4]. Trade restrictions are at record levels and the average effective US tariff is at its highest since the early 1940s [28][15]. Ton-miles are projected to grow only 0.3% a year to 2030 [24].

The rebuttal attacks the aggregation rather than the data. Baldwin shows that the four largest traders peaked at different moments for different reasons: China in 2006, the United States in 2011, Japan in 2014, and the European Union not clearly at all [20]. Forcing dissimilar series into a single peak, he argues, conveys a false impression of a common turning point that never existed [20].

The services margin is the substantive part of the counter-argument. Digitally delivered services reached about 56% of global services exports in 2025, worth 5.3 trillion dollars within total services exports of 9.6 trillion, having grown at an average 7.1% a year over the past decade [23]. Baldwin, Freeman and Theodorakopoulos argue that the future of trade lies specifically in intermediate services, the margin that goods-based measures do not capture at all [21].

South-South trade is the second piece of evidence against a global retreat. South-South merchandise trade reached 6.2 trillion dollars in 2024, up 7%, and its share of world merchandise trade rose from 11% in 2000 to 26% in 2024 [22]. 57% of developing-country exports now go to other developing economies, against 38% in 1995 [22]. That is integration deepening in the part of the world economy that is growing fastest.

The regionalisation thesis holds up least well of the three claims. Intra-ASEAN trade has stagnated slightly above 20% of the bloc's total, against about 43% for ASEAN+3 and roughly 61% for the European Union [37]. Intra-ASEAN goods trade fell from 25.1% in 2010 to 22.5% in 2024, and services from 19.8% to 13.8% [38]. In the fastest-growing trading region on earth, regional integration has gone backwards while the regionalisation narrative was being written.

What would settle the argument is a measurement that neither side currently produces at annual frequency: value added by origin, for the whole world, with a lag of months rather than years. The PIIE exercise demonstrates that the method works and that it changes the answer, but it depends on input-output tables that arrive long after the customs data they correct [6]. Policy is therefore being made against the faster and less accurate series.

Until that gap closes, the responsible position is the narrow one. Goods globalisation has plateaued as a share of output; services globalisation has not [20][21]. Bilateral concentration has fallen sharply; upstream concentration has barely moved [6]. Anyone who reports only one half of those pairs is describing a world that does not exist.

The first finding is that the volume story is settled and the volume story is boring. Trade grew faster than output in 2025, container throughput set records, and value chains still carried 46.3% of world trade [3][25][4]. The trade-to-GDP ratio has been flat since 2008, which is a plateau rather than a decline [41]. Anyone forecasting a collapse in cross-border exchange has now been wrong for six consecutive years.

The second finding is that the reallocation is real, large and almost entirely bilateral. China went from about 21% of US goods imports in 2017 to roughly 9% in August 2025, and the corridor between the two economies shrank around 30% [9][18]. Taiwan, Vietnam and Mexico absorbed most of the displaced volume [6]. That is a genuine structural change in who ships to the United States.

The third finding undoes most of the strategic value of the second. The Chinese value added inside US imports fell about two percentage points while the bilateral share fell seven [6]. Chinese-owned firms supplied a quarter of Vietnam's exports to the United States by 2023, up from 11%, and doubled their share of Vietnam's imports from China [12]. The assembly moved; the supply base largely stayed.

The fourth finding is that reshoring, in the specific sense of domestic output displacing imports, has not happened. The Kearney index remained negative in 2026, manufacturing imports rose 4.6% to a four-year high, and capacity grew 1.5% across four years of tripled investment [7]. Greenfield capital went to data centres rather than to import substitution [14].

The fifth finding concerns cost and who carries it. IMF staff put long-run fragmentation losses between 0.2% and almost 7% of global output, with the burden falling hardest on smaller and middle-income economies dependent on open markets [19][42]. Diversification that duplicates tooling and shortens production runs raises unit costs, and those costs land downstream [39]. The resilience being purchased is real; it is not free, and it is not yet being measured against its price.

The sixth finding is institutional rather than economic. Trade policy authority in the largest importing economy changed statutory basis twice and survived two adverse court rulings within four months [16][17]. Policy activity is at a series high and rising [5]. A firm making a ten-year sourcing commitment is pricing legal volatility alongside freight and labour, and legal volatility has no forward market.

The conclusion the data supports is narrower than either side of the public argument. Globalisation did not peak in the sense of retreating; goods trade plateaued as a share of output while services and South-South trade kept expanding [20][23][22]. Supply chains did not come home; they acquired an extra jurisdiction between the input and the customer. The dependence that the policy set out to reduce is still there, one step further upstream, and considerably harder to see.