Is the World Economy Coming Apart, or Being Rewired?
From the The Next Global Economy: AI, Fragmentation, and Development to 2045 collection
In 2025 the average effective tariff rate facing American consumers rose to around 18 percent, by the Yale Budget Lab's estimates, the highest level since the 1930s. It was easy to read the number as an obituary for globalization. Export controls on advanced chips, sanctions on a major energy producer, and large new factory subsidies seemed to point the same way: the integrated world economy was coming apart.
The trade data tell a less dramatic and more interesting story. For about two decades before 2008, goods trade grew much faster than world output. Since then it has held roughly steady as a share of global GDP. It stopped outgrowing the economy; it did not collapse.
The difference matters because the two readings imply different worlds. A world that is deglobalizing is one in which countries retreat and flows shrink. A world that is being rewired is one in which the flows persist but their routes, their rules, and their chokepoints change. The second description fits the evidence better, and it carries a further complication: a general-purpose technology, artificial intelligence, is arriving in the middle of the rewiring.
Reading the next two decades well means holding several forces at once rather than betting on any single one. Four ideas carry most of the weight.
Flat is not falling
Economists have a name for the post-2008 pattern: slowbalization. Volumes hold up while the geography underneath shifts. The clearest example is what followed the US tariffs of 2018 and 2019. China's share of American imports fell, and the shares of Vietnam and Mexico rose.
Research at the World Bank by Caroline Freund and co-authors found something more subtle. The countries that gained the most American market share tended to be those already most integrated with Chinese suppliers. Much of the diverted trade was assembled from Chinese inputs and shipped from a new address. These connector economies absorbed the rerouting without dissolving the dependence beneath it.
The lesson generalizes. Tariffs and sanctions change the last leg of a supply chain far more easily than its deeper structure. The costs, meanwhile, landed largely at home: a 2019 study by Mary Amiti, Stephen Redding, and David Weinstein found complete pass-through of the 2018 tariffs into the domestic prices of the imported goods. Whether the larger 2025 round achieves its stated goals of reshoring, leverage, and revenue remains a live debate, and each goal deserves a separate verdict.
The state as a permanent participant
Rerouting is not a market accident. It follows from a broad return of the state as an economic actor. The US CHIPS and Science Act of 2022 set aside about $39 billion in manufacturing subsidies for semiconductors, the Inflation Reduction Act added clean-energy incentives the same year, and Europe, Japan, and Korea responded with programs of their own. Export controls, sanctions, and investment screening, once exceptional, became routine instruments of economic statecraft.
The useful question about any of these tools is not whether the state should act, but what its action builds. Subsidies that develop suppliers, skills, and learning can leave lasting capability. Subsidies that compete for the same factory mostly raise the price each government pays for it. Sanctions and controls raise costs for their targets, who adapt by substituting, stockpiling, and trading through third countries.
China is the largest case of state-directed adjustment. The property correction that began in 2021 ended a growth engine built on debt-financed construction, and Beijing's answer, framed as new productive forces, pushed scale in electric vehicles, batteries, and solar. Whether the resulting export surge reflects overcapacity or competitiveness is contested, and the honest answer is likely some of each. Either way, China's domestic choices now set world prices in the technologies of the energy transition.
AI lands on physical ground
Artificial intelligence is often discussed as software, weightless and instantly global. Its economics say otherwise. Large models depend on advanced chips, data centers, and electricity, which makes compute a factor of production in the way steel and oil once were. The International Energy Agency estimated data-center electricity use at around 415 terawatt-hours in 2024 and, in its base-case projection, sees that figure roughly doubling by 2030.
Those inputs sit on the same fractured map as everything else. Advanced chips are subject to export controls, and grid capacity depends on local permits and national energy policy. The minerals behind power systems and batteries pass through a processing chokepoint: in the IEA's 2025 assessment, China is the leading refiner for 19 of 20 strategic minerals, with an average share near 70 percent. The fragmentation story and the AI story turn out to be one story.
The productivity payoff may also arrive later than enthusiasts expect. Electricity and computing both showed a lag between adoption and measured gains, because firms first had to reorganize around them. Economists call this the productivity J-curve. Early firm-level studies are promising, such as a 14 percent average productivity gain among customer-support agents given a generative AI tool, reported in the Quarterly Journal of Economics in 2025, but national statistics may take years to show comparable effects.
Slow variables, hard limits
Beneath trade and technology run variables that move slowly and bind tightly. The first is money. The inflation surge of 2021–23 and the rapid rate increases that followed ended a decade of near-zero real interest rates. Positive real rates reprice public debt, corporate investment, and asset values at the same time, and they arrive while public debt in many advanced economies sits near postwar highs.
The second is people. China's population began falling in 2022, and South Korea's fertility rate reached 0.72 children per woman in 2023. United Nations projections place most growth in the working-age population to 2050 in sub-Saharan Africa and South Asia. Whether that surge becomes a dividend depends on jobs, capital, and cities, and on a development ladder that may be harder to climb now that factories need fewer workers.
These constraints interact with AI in both directions. Faster productivity growth could ease the burden of aging workforces and heavy debt; slower growth would leave both problems larger. The AI productivity question is therefore a fiscal and demographic question as much as a technological one.
Adaptability as the durable advantage
Put the forces together and a single forecast looks fragile. Two uncertainties dominate: how fast AI raises productivity, and how deep geopolitical fragmentation runs. Crossing them yields four plausible worlds, each favoring different kinds of economies. Technology leaders gain most from fast AI, connector economies gain from fragmentation that is deep but managed, and commodity exporters and young low-income economies face very different prospects depending on which world arrives.
The practical response is a short list of signals watched over time: productivity data, trade diversion, chip controls, fertility, debt costs, and the dollar's share of reserves. Each headline then becomes evidence about which world is unfolding, rather than a verdict on globalization itself.
The world economy is not being dismantled. It is being rewired under several pressures at once, and no single one will decide the outcome. The advantage over the next two decades belongs to the economies and the people able to adjust to several of those pressures together, not to those that bet everything on one.