The Compute Corridor: Why Every American Military Move Points to AGI
Every time the United States moves a carrier group, threatens a tariff, or floats an idea as outlandish as buying Greenland, pundits reach for the nearest frame: domestic politics, electoral signaling, alliance reassurance, fossil fuel nostalgia, great power competition in the abstract.
But step back.
Look at the map.
Every recent American kinetic action -- and every conspicuous non-kinetic "what if" -- clusters around global trade chokepoints and energy corridors: the Strait of Hormuz, the Bab el‑Mandeb, the South China Sea, the Red Sea, the Panama Canal, and the Arctic routes opening through Greenland's shadow.
On the surface, it looks like old‑school geopolitics. The 20th‑century script, dusted off.
But that reading is incomplete.
Because in 2026, oil is not just about oil. Trade routes are not just about trade. And Taiwan is not just about Taiwan.
It is about controlling AGI.
The Concentration Problem
Consider the hard numbers behind that claim. Taiwan Semiconductor Manufacturing Company's dominance is not rhetorical: TSMC accounts for roughly 55-60% of global foundry revenue and more than 90% of leading-edge nodes below 7nm, according to Counterpoint Research and TrendForce. Nvidia, in turn, commands an estimated 80-95% share of the data-center GPU market used for AI training, with virtually all of its advanced chips fabricated by TSMC. This concentration is historically anomalous, closer to a single chokepoint than a competitive market.
Taiwan sits at the center of the modern compute stack. TSMC controls roughly 72% of the global foundry market and close to 90% of advanced‑node manufacturing. Nvidia -- which supplies the GPUs that train nearly every frontier AI model -- consumes the overwhelming majority of those wafers. This is not a diversified supply chain; it is a single point of failure.
Energy as Compute Bottleneck
Training frontier AI models is an energy event.
The International Energy Agency estimates global data center electricity consumption at roughly 460 TWh in 2022, rising sharply with the post‑2023 AI training boom; multiple projections now place data center demand between 800 and 1,050 TWh by 2030 under current growth trajectories. Goldman Sachs estimates incremental AI-related power demand alone could exceed 160 GW globally by the end of the decade, roughly equivalent to adding another Japan to the grid. These are not marginal loads; they are system-shaping ones. Global data centers consumed more than 400 TWh of electricity in 2024 and are projected to approach 1,000 TWh by 2030, driven primarily by AI workloads. Individual training runs are on track to draw gigawatts of power. Energy stability translates directly into compute stability.
This is why chokepoints matter.
The Strait of Hormuz carries about 20-21 million barrels of oil per day, roughly one-fifth of global petroleum liquids consumption (U.S. EIA). The Bab el‑Mandeb funnels energy and container traffic between Asia and Europe; after Houthi attacks in late 2023 and 2024, major shippers rerouted around the Cape of Good Hope, adding weeks of transit time and materially raising freight and energy costs. The Strait of Malacca handles roughly 23-25% of global seaborne trade by volume. These figures are not symbolic; they directly translate into fuel prices, grid stability, and the cost of running hyperscale compute.
The causal chain is straightforward: AGI depends on frontier compute; compute depends on advanced semiconductors; advanced semiconductors depend on Taiwan; Taiwan's security depends on US naval dominance; naval dominance depends on control of energy flows and maritime chokepoints.
Every American move that hardens these corridors buys time.
The CHIPS and Science Act explicitly frames this time horizon. The U.S. has committed over $50 billion in incentives to reshore advanced semiconductor manufacturing, with TSMC's Arizona fabs (targeting 4nm and eventually 3nm nodes), Intel's Ohio and Arizona expansions, and Samsung's Texas investments all designed to narrow -- not eliminate -- the Taiwan gap by the late 2020s. Simultaneously, the Department of Energy has fast-tracked permitting and financing for large-scale generation and transmission projects tied to data center clusters, implicitly acknowledging that compute is now a strategic load, not a commercial afterthought. Time to reshore fabs. Time to scale domestic energy. Time to prevent a sudden compression of the AI race via a Taiwan shock.
This is not to say AGI is the only driver of US strategy. Sea lanes mattered before AI. Oil mattered before GPUs. But AGI is increasingly the meta‑layer that reframes all of it. What once powered the industrial economy now powers the intelligence economy.
From this perspective, Greenland is not a meme, the Red Sea is not a sideshow, and Taiwan is not just another flashpoint. They are structural nodes in the race to scale intelligence.
Oil is a proxy. Sea lanes are a proxy. Alliances are a proxy.
Seen this way, U.S. naval posture, export controls on advanced chips, restrictions on EUV lithography, and even industrial policy all collapse into a single strategic question: who controls the throughput of intelligence. AGI is not constrained by ideas -- those diffuse quickly -- but by atoms, electrons, and logistics. The map still matters, but now it maps compute.
The real prize is whoever gets to define -- and deploy -- AGI at scale.
Counterarguments and Stress Tests
A skeptic might object on several fronts.
First, the diversification argument: TSMC is building fabs in Arizona and Japan; Intel and Samsung are scaling advanced nodes; the chokepoint is narrowing, not widening. This is true directionally but misleading on timelines. Arizona's first 4nm fab won't reach volume production until 2025-2026, and even optimistic projections put US-based advanced capacity at 10-15% of global supply by 2030. The gap is narrowing, but slowly.
Second, the substitution argument: if Taiwan were disrupted, couldn't hyperscalers shift to older nodes or alternative architectures? In principle, yes. In practice, frontier AI training is hardware-bound at the margin. A 20% efficiency loss in chip performance translates to months of delay in model capability. The race is measured in weeks, not years.
Third, the multipolarity argument: China is building its own semiconductor stack (SMIC, Huawei), and export controls may accelerate rather than prevent decoupling. This is the strongest objection. But even aggressive Chinese timelines place domestic 5nm-equivalent capacity at scale no earlier than 2027-2028, and the EUV lithography bottleneck (ASML, Netherlands) remains unresolved. The US strategy is not to prevent Chinese progress indefinitely -- it is to maintain a lead long enough to reach AGI first.
None of these objections invalidate the core thesis. They complicate it. The race is not guaranteed; it is contested. But the map of contestation -- Taiwan, Hormuz, Malacca, Greenland -- remains the same.
Data: Chokepoints and Compute
Sources
- IEA Energy and AI Report: https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai
- Goldman Sachs AI Power Demand: https://www.goldmansachs.com/insights/articles/AI-poised-to-drive-160-increase-in-power-demand
- U.S. EIA Strait of Hormuz: https://www.eia.gov/todayinenergy/detail.php?id=65504
- EIA World Oil Transit Chokepoints: https://www.eia.gov/international/analysis/special-topics/WorldOilTransit_Chokepoints
- TrendForce Foundry Market Share: https://www.trendforce.com/presscenter/news/19700101-12510.html
- Nvidia GPU Market Share: https://carboncredits.com/nvidia-controls-92-of-the-gpu-market-in-2025-and-reveals-next-gen-ai-supercomputer/
- CHIPS and Science Act (NIST): https://www.nist.gov/chips
- White House CHIPS Fact Sheet: https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2024/08/09/fact-sheet-two-years-after-the-chips-and-science-act-biden-%E2%81%A0harris-administration-celebrates-historic-achievements-in-bringing-semiconductor-supply-chains-home-creating-jobs-supporting-inn/
- UNCTAD Review of Maritime Transport 2024: https://unctad.org/system/files/official-document/rmt2024ch1_en.pdf
