The AI Boom Isn't Shrinking the World. It's Speeding It Up.
Undersea cables, chip supply chains, power grids, and mines are outpacing the trade rules meant to govern them, and the gap decides which protections count as friction.
Undersea cables, chip supply chains, power grids, and mines are outpacing the trade rules meant to govern them, and the gap decides which protections count as friction.
Despite two years of headlines about reshoring, friend-shoring, and strategic autonomy, the physical infrastructure of global exchange is accelerating. Undersea cable buildout, semiconductor supply chains, power-grid expansions, and mineral-extraction projects are moving at a pace that formal trade metrics, and political rhetoric, have not caught up to.
International internet bandwidth now measures 1,835 terabits per second, according to Telegeography. UNCTAD's policy accelerator on cross-border data flows argues that those flows have outstripped the growth of goods, services, and people flows, a reversal that contradicts the political language of deglobalisation. The form of integration has changed. The direction of integration has not.
Tech, logistics, and finance rules are being pushed to move faster; environmental, labour, and Indigenous protections are being treated as friction to strip away. A Social Europe essay by Alan Shipman labels the moment "hyperglobalisation 2.0", borrowing the German sociologist Hartmut Rosa's concept of "social acceleration." Stripped of the academic vocabulary, the move is simple. The defining feature of current globalisation is the speed of exchange, not the size of the flows. Cables, chips, and grids are not metaphors. They are the actual substrate.
The IMF's 2026 scenario exercise on the global economic and financial implications of AI treats the same infrastructure as a macro variable, modelling how compute, energy, and capital deployment shift under different AI uptake paths. That puts the question on the macro desk, not just the technology desk, which is where the trade-policy and standards debates will be fought.
The critical-mineral layer is where the asymmetry becomes physical. Data-center buildout is feeding demand for cobalt, nickel, copper, and rare earths at a scale that mining and permitting systems were not built to absorb. Permits that take a decade to clear become political targets, not because the project is unimportant but because they slow the velocity that the rest of the AI stack depends on. A peer-reviewed paper in AI & Society by Cox and colleagues puts it bluntly: "energy creates intelligence". AI is an infrastructural geography problem, tied to grids, siting, and mineral extraction. When the public conversation treats those physical dependencies as abstract "AI," it loses the ability to ask who absorbs the costs.
The WTO's 2025 World Trade Report and its AI-TPOI trade-policy observatory offer a useful counterpoint. The index attempts to measure AI's effect on trade openness and fragmentation, and the early evidence is messier than either the boom narrative or the doom narrative suggests. Formal trade policy is moving slowly because trade agreements are slow, while the underlying data and capital flows are moving quickly. The two are not in sync. The WTO is the institution most visibly behind the curve.
That gap, between physical velocity and political perception, is where the next decade of trade politics will be decided. The AI race is not a single race. It is a stack of races, and each layer, from data and compute to energy and minerals, has its own permit office, its own standards body, and its own constituency that can either be accelerated or treated as friction. The next WTO ministerial, the next round of critical-mineral agreements, and the next set of cable-landing permits will tell us which way the gap closes. The political debate has not yet caught up to the infrastructure that is already being built.