The Asymmetry

On paper, it is not even close.

The United States commands seventy-nine percent of global AI investment — $159 billion in 2025 alone. Its companies design ninety percent of the chips that make artificial intelligence possible. Its labs produce the frontier models. Its hyperscalers — Meta, Amazon, Alphabet, Microsoft — have committed $320 billion to AI infrastructure in a single year, with projections that push toward half a trillion annually by the early 2030s. Its grid is adding 86 gigawatts of new power capacity in 2026. Its nuclear renaissance is underway — Three Mile Island may reopen by 2028 to feed Microsoft's data centers, while Meta has announced projects unlocking 6.6 gigawatts of nuclear power for AI. Its Arctic strategy is materializing: a billion-dollar data center in Narvik, $15 billion committed to Nordic infrastructure, cables being laid beneath the Arctic Ocean to bypass the chokepoints.

The United States is energy-independent. It hosts fifty-one percent of the world's data centers. It attracts — for now — the world's best AI talent. Immigrants contribute to thirty-six percent of American innovation.

If this were a chess match, the evaluation bar would be tilted so far to one side it would look broken.

And yet.

The paradox of dominance in an interconnected system is that the more powerful you become, the more you depend on things you cannot control.

Consider the supply chain that makes American AI supremacy possible. NVIDIA designs the chips in Santa Clara, California. But those chips are fabricated in Hsinchu, Taiwan, by TSMC — which controls over ninety percent of advanced AI chip production at the cutting-edge nodes. The lithography machines that TSMC uses come from a single company, ASML, in Veldhoven, the Netherlands — the sole manufacturer of the extreme ultraviolet equipment required for sub-7nm fabrication. The rare earth magnets inside every data center server, every cooling system, every power transformer are processed through Chinese facilities that handle ninety percent of global supply and ninety-four percent of permanent magnet production. The data flowing between those servers — between continents, between training clusters, between the models and the users — travels through undersea cables, fifteen of which pass through a twenty-six-kilometer strait in the Red Sea controlled by Houthi rebels backed by Iran.

TSMC itself relies on China for roughly thirty percent of the consumables used in its most advanced chip production. The company building America's technological future cannot operate without materials from America's principal strategic competitor.

This is not an oversight. It is the architecture.

The global AI supply chain was not designed by a strategist. It was assembled by markets — over decades, across borders, optimized for cost and efficiency, with no regard for geopolitical resilience. Every node exists where it does because that was the cheapest place to put it, or the only place with the technical capability, or the location where some geological accident deposited the right minerals in the right concentrations. The result is a system of extraordinary productive power and extraordinary fragility — a machine that works beautifully until any single component fails.

What does the asymmetry actually look like? Not on a balance sheet — in the world?

Start with Taiwan. The CHIPS Act has allocated $30.9 billion across forty projects, and the first TSMC fab in Arizona has been certified. Commerce Department projections suggest the United States will manufacture twenty percent of the world's leading-edge logic chips by 2030, up from zero today. But that timeline stretches five to ten years into the future, and the fab in Arizona is producing chips on nodes that TSMC in Taiwan has already moved past. The ecosystem density of Hsinchu — the network of suppliers, specialists, and institutional knowledge that has accumulated over forty years — cannot be replicated by writing checks. As Carnegie's analysis concluded: reshoring has limits.

Meanwhile, that island sits 110 miles from a nation that considers it sovereign territory and has not renounced the use of force to reclaim it. The silicon shield — the theory that Taiwan's chip monopoly protects it from invasion because the world economy cannot afford the disruption — cuts two ways. It makes Taiwan worth protecting. It also makes Taiwan worth seizing.

Move to the Strait of Hormuz. The United States is energy-independent in aggregate — its natural gas reserves and production capacity can power its own AI infrastructure without importing a single barrel of oil. But energy independence is not energy isolation. When oil prices spike because Hormuz is threatened, the price of everything the United States imports rises too — the rare earths, the processed materials, the components manufactured in countries that are not energy-independent. America's allies in Europe, Japan, South Korea, and Taiwan are deeply dependent on Gulf energy. An energy crisis that cripples allies cripples the alliance.

Move to the Red Sea. Forty-four cable damage events across thirty-two locations in eighteen months. The median restoration time is forty days; in conflict zones, longer. The cables carry seventeen percent of global internet traffic, and while the internet has proven more resilient than catastrophists predicted — traffic reroutes, latency increases but service continues — the margin between degraded performance and genuine disruption is thinner than most people realize. Arctic cables offer a bypass, but they require specialized vessels for deployment and maintenance, and the buildout timeline stretches to 2030 and beyond.

Move to China's rare earth leverage. In October 2025, Beijing expanded export controls to require licenses for any product containing more than 0.1 percent Chinese-origin rare earth content when connected to semiconductors, AI, or defense applications. The foreign-direct-product rule reaches any product anywhere in the world. Building alternative processing capacity takes ten to fifteen years and billions in capital investment. Western projects — MP Materials, Lynas, Greenland's nascent mines — exist on paper and in pilot stages, but the gap between mining raw ore and producing refined, semiconductor-grade material is vast and expensive to close.

These are not separate vulnerabilities. They are the same vulnerability expressed through different materials. The system is one system.

China sees the asymmetry differently — and not without reason.

From Beijing, the picture looks like a nation under siege. American chip export controls are a "burn" — slow, cumulative, designed to degrade Chinese capability over time. The entity list has swollen past 140 companies, with over fifty added in a single action in March 2025. The message is clear: America intends to maintain its lead not just by running faster but by hobbling the competition.

China's response has been more resourceful than Washington anticipated. DeepSeek's January 2025 breakthrough — training a frontier-competitive model for $5.6 million on hardware the United States had deliberately downgraded — sent $589 billion in NVIDIA market capitalization evaporating in a single day. The collapse of the "compute-only" thesis proved that algorithmic innovation can compensate for hardware constraints. Google DeepMind's CEO, Demis Hassabis, acknowledged that China's leading models are closing the gap rapidly.

China is building its parallel stack. Huawei's Ascend 910C chips, fabricated by SMIC using deep ultraviolet lithography — less advanced than ASML's EUV machines, but functional. DeepSeek releasing models with first-day optimization for Huawei's CANN software stack, deliberately building a non-NVIDIA, non-CUDA ecosystem. Chinese AI firm Zhipu training major models entirely on domestic hardware. China targeting triple its AI chip output in 2026.

The performance gap is real — Huawei's best reportedly runs at sixty to seventy percent of NVIDIA's latest — but the gap is narrowing, and efficiency gains are compressing the timeline. China's rare earth controls are the mirror image of America's chip controls: a "choke" to match the "burn." Analysts at War on the Rocks argue the semiconductor controls will outlast the rare earth weapon, because chips represent a durable capability advantage while raw materials are a one-time leverage that depletes through use. Perhaps. But in the short term, both sides have their hand on the other's throat.

And then there are the nations caught in between — the ones whose territory, resources, or geography make them nodes in a system they did not design.

India hosts the AI Impact Summit and joins both the Quad and BRICS, hedging with a sophistication that has become its defining diplomatic posture. Indonesia buys fighter jets from China while maintaining close US military ties. Saudi Arabia signs a $600 billion AI deal with Washington while expanding trade through BRICS. Brazil, Turkey, South Africa, the UAE — all practicing what analysts call "multi-alignment," using their resources, their geography, and their strategic ambiguity as leverage.

Their choices will determine whether the bifurcation hardens into permanent separation or remains porous enough for the global system to function. India, with 1.4 billion people and the world's fastest-growing major economy, is building its own AI stack — training models on domestic data, developing chips through joint ventures, positioning itself as the democratic alternative to Chinese AI for the developing world. Indonesia controls critical nickel reserves essential for data center battery systems and is leveraging that position to demand technology transfer. Brazil sits atop rare earth deposits that could rival China's and is fielding offers from both Washington and Beijing. These are not passive bystanders. They are the swing voters in a geopolitical contest that will determine how the AI infrastructure of the next half-century is built, governed, and distributed.

These countries are not choosing sides. They are choosing not to choose, and in doing so, they are revealing something important about the system itself: it cannot actually bifurcate cleanly. The supply chain is too entangled. China provides twenty-four to thirty-two percent of imports for the major hedging nations. American technology runs on Taiwanese chips that depend on Chinese consumables processed with Dutch machines using gases that once came from Ukrainian steel mills. Pull one thread and the entire fabric shifts.

The asymmetry, then, is not simply that the United States is stronger. It is that strength in this system is paradoxical. The more dominant you become at any single node, the more dependent you become on the nodes you don't control. America can design the best chips in the world, but it cannot fabricate them at scale on its own soil — not yet. China can process ninety percent of the world's rare earths, but it cannot design the frontier chips those earths go into. Neither can secure the cables, the energy, the minerals, and the manufacturing simultaneously.

The game-theory term for this is "bifurcation without separation" — two systems pulling apart but remaining entangled through supply chains neither can sever without severing themselves. Some researchers have proposed a framework called Mutual Assured AI Malfunction — MAIM — drawing an analogy to nuclear deterrence. The idea is that each side's ability to disrupt the other's AI infrastructure creates a stable equilibrium.

But MAD worked — to the extent it did — because nuclear weapons sit in silos. The AI supply chain does not sit anywhere. It moves. It flows through straits and under oceans and across borders. You cannot deter a cable cut the way you deter a missile launch. You cannot threaten retaliation against an anchor dragged by a fishing vessel whose flag state denies involvement.

The asymmetry is real. America holds the stronger hand. But in a system this interconnected, holding the stronger hand does not mean you control the game. It means you have more to lose when the table flips.

What happens next depends on which of three futures is unfolding — and whether anyone has the vision to choose a different one.