Taiwan holds the leading edge logic, South Korea owns the high bandwidth memory, and Japan and Vietnam stake AI built into robotics and industrial systems and assembly: four bets, not two.
The contest shaping the next decade of artificial intelligence is not the two-player race it is usually framed as. Look at the supply chain, and a different geography appears: Taiwan, South Korea, Japan and Vietnam are each staking a distinct node of the AI value chain, and refusing to be rule-takers in someone else's game.
Taiwan sits at the centre of advanced AI chip production through TSMC, the only firm that can fabricate the most advanced processors at scale. Nvidia CEO Jensen Huang, who was born in Taiwan, called the island the "epicentre" of the AI revolution during a recent visit, per The Conversation. The Taiwanese government has made that posture official. In late 2025, foreign minister Francois Chih-chung Wu said the most advanced chipmaking capabilities must remain on the island, articulating what The Conversation frames as the island's de facto doctrine. The Taiwan AI Basic Act, passed in 2025, codifies that intent, and the Ministry of Digital Affairs' industry initiatives have moved it into procurement and infrastructure. Even as TSMC builds fabs in Germany, Japan and the United States, its leading-edge nodes stay reserved for Taiwan.
South Korea is concentrating capital in the layer just below: memory. Samsung and SK Hynix together hold roughly 80% of the global market for high-bandwidth memory, the specialised chips that feed AI processors data fast enough to train and run large models. The two firms have committed nearly US$600 billion (£446 billion) to two new fabs in the country's south-west, a capital bet that re-anchors the memory layer inside Korean borders for the next decade. The logic and memory halves are not duplicates. Nvidia-designed processors fabricated by TSMC, stacked with Samsung or SK Hynix HBM, form the coupled stack that most frontier AI accelerators are built around today. South Korea is staking the half of that coupling that can be priced, contracted and exported independently of the other.
Japan and Vietnam are staking the next layer down, and the layer above, in ways that are less mature but no less deliberate, per The Conversation's four-economy analysis. Japan is using its base in precision manufacturing and robotics to position for the "physical AI" buildout, the deployment of AI inside factories, vehicles and industrial systems where its incumbents already hold global share. Vietnam, with lower-cost assembly capacity and a growing electronics supply chain, is positioning to capture packaging and integration work that the leading-edge logic and memory layers rely on. The Conversation frames all four bets as a value-chain sovereignty strategy, in which no single country tries to replicate the whole stack but each one refuses to be cut out of it.
The four bets are mutually reinforcing rather than duplicative. A TSMC processor without Samsung or SK Hynix HBM is a stranded chip; a Korean memory fab without Taiwanese logic is missing its customer. That coupling is also the soft underbelly of the four-country map. Foundry allocation for advanced packaging, including CoWoS, was fully booked through Q1 2026, and TSMC's AP7 advanced-packaging line and Arizona P6 build-out are running behind demand, which means the four-country stack is currently capacity-constrained, not capacity-redundant.
The clearest pressure on the map is internal. "Sovereignty" is partly aspirational: the four bets still depend on US design intellectual property, since Nvidia holds the architecture that TSMC fabricates and Samsung packages, on US capital markets for the long-dated fab financing, and on US export controls for the boundary conditions that make east-Asian leadership in the most advanced nodes possible at all. A re-verticalization of the stack, whether by a single firm reabsorbing memory and logic, or by a US-China settlement that re-opens the export-control regime, would compress the four-country map back toward two. Until then, the working assumption for industrial policy on either side of the Pacific is that AI supply runs through at least four Asian capitals, not two.