Inside the US-China Chip War: Semiconductors, Rare Earths, and the New Cold War

Semiconductors are, on their face, an unlikely centerpiece of great-power rivalry — small pieces of etched silicon, invisible inside the devices that run everyday life. But chips sit at the exact intersection of military capability, AI development, and economic competitiveness, which is why the fight over who can build, buy, and sell them has become one of the defining fronts in the US-China relationship. What started in 2022 as targeted export restrictions has, by 2026, hardened into something closer to a bifurcated global technology ecosystem — two increasingly separate supply chains, two sets of chip architectures, and two AI infrastructures that are growing less compatible by the year.

From targeted controls to a full framework

The most recent major escalation came via a Bureau of Industry and Security rule that tightened export controls on advanced semiconductors bound for China and Macau, introducing total processing power thresholds and moving licensing for high-end AI chips — including Nvidia's H200 and AMD's MI325X — from a general presumption of denial to case-by-case review. That sounds like a technical adjustment. In practice, it determines whether China's AI labs and cloud providers can access the chips capable of training frontier AI models, which is precisely the point.

The policy path getting there hasn't been linear. Within the span of about twelve months, the H200 was banned, then unbanned, then hit with a 25% tariff, then wrapped in a licensing framework that industry analysts immediately flagged as internally contradictory — banning a chip outright, then taxing its sale, then requiring case-by-case approval, all within the same policy cycle. That volatility is itself a data point: export control policy toward China has become a live political football, shifting with administrations and negotiating leverage rather than settling into a predictable long-term regime.

China's countermove: rare earths

Beijing's retaliation hasn't focused on chips it can't yet build domestically at the leading edge — it has focused on what it controls almost completely: rare earth elements and critical minerals essential to chip manufacturing, defense systems, and electric vehicles. China's near-monopoly on rare earth processing (as opposed to mining, which is more geographically distributed) gives it a chokepoint the US doesn't have an easy substitute for, and it has used export controls on these materials as direct pressure in response to chip restrictions. In 2026, China further expanded its own export control list to include US firms, including a rare-earth miner, signaling this is now a genuinely two-sided fight rather than a unilateral US policy.

Congress wants to go further

Legislative pressure in Washington has pushed toward tightening the net further still. Lawmakers have advanced roughly twenty new export control measures, including the Match Act, which would require allied countries — chiefly Japan and the Netherlands, home to critical semiconductor equipment makers like ASML — to align more closely with US restrictions on selling advanced manufacturing equipment to China. This matters because unilateral US controls are far less effective if allies keep selling China the equipment it needs through a side door; the Match Act is an attempt to close that door across the entire allied semiconductor supply chain, not just the American portion of it.

The case for skepticism about controls

Not every expert believes this strategy is working as intended. Analysis from groups like CSIS has pointed to the limits of export controls as a tool for meeting the broader China challenge: restricting access to leading-edge chips has accelerated China's investment in domestic semiconductor manufacturing and alternative chip architectures, potentially building exactly the self-sufficient rival capability the controls were meant to prevent, just on a longer timeline. There's also evidence that enforcement of the current rules has "cooled down" in practice even as the rhetoric around them hasn't, suggesting a gap between policy on paper and policy as actually implemented.

What a bifurcated tech world looks like

The long-term trajectory here isn't a negotiated settlement — it's technological sovereignty on both sides, meaning two increasingly separate chip ecosystems, incompatible AI infrastructure stacks, and supply chains that no longer assume the free flow of components across the US-China relationship the way they did for the previous three decades of globalization. For businesses operating in either market, and for any economy trying to plan around global supply chains, the operating assumption for the rest of this decade has to be that semiconductors are no longer just a commercial product — they're a strategic asset being actively fought over, with the terms of that fight still being written in real time.