The Chip War's Hidden Lever: Why Rare Earths, Not Semiconductors, Define the US-China Tech Contest
The most telling detail from the Trump-Xi summit did not come from the podium. While AI safety cooperation made the headline and chip export restrictions occupied the photogenic portions of the joint programme, it was the quiet agreement on rare earth exports that moved the needle on the ground. According to reporting from The Epoch Times, rare earth export terms improved while semiconductor negotiations stalled. That inversion tells a more accurate story about leverage in the US-China tech relationship than the dominant Western framing admits.
For years, Washington has presented the semiconductor restriction regime as its primary instrument of competitive pressure against Chinese technology firms. The export controls, the foreign direct product rule, the allied restrictions on chip-making equipment — these have been portrayed as the cutting edge of a containment strategy. Meanwhile, the underlying dependency on Chinese processing of critical minerals has been treated as a political inconvenience rather than a strategic vulnerability. The summit's outcome suggests Beijing understood this asymmetry better than Washington wanted to acknowledge.
The Leverage That Sits Closer to the Ground
Semiconductors are manufactured. Rare earths are extracted and refined. The distinction matters because manufacturing capacity can be built — it takes time, capital, and know-how, but it can be built. Refining capacity sits in geography that does not move. China processes roughly 85 percent of the world's rare earths, according to figures cited by multiple mining industry trackers. That is not a temporary characteristic of the market; it reflects decades of investment, favourable geology, and industrial integration that took hold when rare earths were a niche commodity. By the time Washington recognised the strategic weight of those supply chains, the infrastructure was already there.
The rare earth category includes seventeen elements with names like lanthanum, neodymium, and dysprosium that most readers will not have encountered in daily life. But their applications are everywhere: electric vehicle motors, wind turbine magnets, fighter jet guidance systems, smartphone speakers. Restricting chip exports to Huawei was a policy choice. Becoming a low-dependency actor in the rare earth refining space is a civilisational infrastructure project. The gap between those two challenges is the gap between the public narrative and the structural reality.
The Epoch Times reporting notes that Trump and Xi also discussed artificial intelligence cooperation frameworks at the Geneva-adjacent meeting. That framing — AI safety as a domain of potential bilateral cooperation — sits awkwardly beside the simultaneous expansion of US chip restrictions. It suggests both governments are running parallel tracks: competitive pressure and managed cooperation, depending on which dimension of the relationship is being discussed in which room.
What Decoupling Actually Means in Practice
The political rhetoric of decoupling has obscured a more complicated operational reality. The United States is not, and cannot be, fully decoupled from Chinese processing infrastructure in the near term. American EV manufacturers source permanent magnets from Chinese-controlled supply chains. American defence contractors source specialty minerals through processing networks that pass through Chinese facilities. The rare earth import figures from the Commerce Department show a steady upward trajectory in tonnage over the past four years despite the political temperature around China. The numbers and the rhetoric run in opposite directions.
China, for its part, has every structural incentive to keep those flows intact. The rare earth refining industry employs hundreds of thousands of workers and sits at the base of an industrial pyramid that includes the electric vehicle sector Beijing is pouring capital into. A complete embargo would be self-harm. But controlled leverage — the ability to signal disruption without executing it — is itself a tool. The improvement in export terms at the summit reads, from that angle, less like a concession and more like a demonstration of the alternative that remains available if the relationship deteriorates further.
The chip restriction regime, meanwhile, operates with a different logic. It is designed to slow Chinese advancement in advanced logic semiconductors, the chips that power AI training clusters. The evidence that it is working in that purpose is mixed. Chinese investment in domestic fabrication capacity — SMIC and its peer institutions — has accelerated. Advanced node development is proceeding more slowly than Beijing's public ambitions suggest, but it is proceeding. The restrictions create friction, not foreclosure.
The Structural Asymmetry Both Sides Understand
Washington's most hawkish China hands will read the summit outcome as a failure of pressure. The semiconductor restrictions did not produce a negotiating breakthrough; the rare earth terms improved anyway. Beijing extracted something on supply chain stability without conceding ground on technology competition. That reading has merit.
But the symmetric critique is equally valid: the rare earth improvement is real, but it is bounded. It does not represent a reversal of the extraction-heavy model that has made China dependent on processing infrastructure it built when nobody was watching. It represents a pause, a managed continuation of an arrangement that neither side wants to upset in the near term but both are working to render unnecessary over the long term. Both governments are spending heavily on domestic alternatives. American rare earth mining projects — Mountain Pass in California, the rare earth initiatives in Montana and Wyoming — are receiving federal capital. China's domestic semiconductor programme is absorbing state investment at a scale that would have seemed implausible a decade ago.
The summit, in this reading, is an interlude. A moment when both sides judged that the costs of managed friction were lower than the costs of managed rupture. The rare earth agreement is the price Beijing charged for sitting in the room. The chip stall is the space Washington is protecting from that same room. Neither side is bluffing. Both are building for a future in which those dependencies no longer exist.
What the Summit Signals
The immediate signal is continuity: both sides prefer functional disagreement to catastrophic rupture. The AI safety language is real — both governments understand that unconstrained AI development carries risks they have an interest in sharing, even if they disagree fundamentally about what safe AI means. The rare earth improvement is real — both governments understand that supply chain disruption in either direction carries costs neither can currently absorb.
The medium-term signal is acceleration. Every year of parallel investment in domestic chip manufacturing and domestic mineral processing narrows the space for this kind of managed arrangement. The rare earth leverage that China demonstrated at this summit will diminish as alternative refining capacity comes online — American, Australian, Canadian, even ultimately Chinese-owned but geographically distributed. The semiconductor leverage that Washington deployed will diminish as Chinese domestic production climbs the node curve, however slowly.
The stake is not which country wins the tech competition. It is whether the tech competition produces a stable transition or a disorderly rupture. The summit suggests both governments, at least for now, prefer the former. Whether that preference survives the next election cycle, the next geopolitical shock, or the next acceleration in AI capability is a question neither the podium nor the joint statement answered.
This publication's approach: the dominant Western wire framing positioned the summit as a chip-competition story. The structural imbalance in leverage — Washington's manufactured semiconductor restrictions versus Beijing's geology-derived rare earth processing dominance — warranted foregrounding in the analysis, as did the awkward fit between AI safety cooperation language and simultaneous technology restriction expansion.