The Chokepoint
Beneath Everything

What Rare Earths Actually Are

The term "rare earths" is a misnomer on both counts. The seventeen elements classified as rare earth elements are not particularly rare in the Earth's crust; cerium is more abundant than copper. And they are not earths in any geological sense that the term implies. What makes them rare, in the economically meaningful sense, is not their abundance in the ground but the difficulty of separating them from the rock in which they occur, and from each other once extracted. They are geochemically gregarious; they appear together, in low concentrations, mixed with other elements including radioactive thorium and uranium, and they require sophisticated chemical processing to isolate into the high-purity forms that industrial applications require. The processing involves solvent extraction, ion exchange, and a series of chemical steps that generate significant radioactive waste. The infrastructure required to do this at scale is expensive, technically demanding, and politically difficult to site.

China built this infrastructure over forty years, beginning in the 1980s under Deng Xiaoping's explicit industrial policy directive. Deng reportedly said that the Middle East has oil; China has rare earths. The policy was patient and deliberate: accept low prices, undercut foreign competitors, allow mines in Australia, South Africa, the United States, and Brazil to become economically unviable, and accumulate the refining capacity that the rest of the world had neither the patience nor the tolerance for waste streams to build. By the mid-2010s, China controlled roughly 85 percent of global rare earth production and over 90 percent of refining. The 2010 episode in which China briefly restricted rare earth exports to Japan during a territorial dispute over the Senkaku Islands demonstrated what that leverage looked like when deployed. The West took note, resolved to diversify, and largely failed to do so before the AI era made the stakes dramatically higher.

The specific rare earths that matter for AI infrastructure are the magnet rare earths: neodymium, praseodymium, dysprosium, and terbium. These four elements are the inputs for neodymium-iron-boron permanent magnets, the most powerful commercially available magnets, which appear in the motors that spin the cooling fans and storage drives of every server rack in every data center on Earth. A large AI data center contains tens of thousands of servers, each with multiple cooling fans and drives, each with permanent magnets. The magnets cannot yet be replaced with any commercially viable alternative. The data centers cannot run without them. The magnets cannot be made without the rare earths. The rare earths cannot be refined at commercial scale anywhere outside China except in small quantities in Malaysia, Estonia, and a few other locations that together amount to less than 10 percent of global capacity.

The October Escalation and
the Busan Suspension

The sequence of events that produced the current one-year clock began in April 2025, when China's Ministry of Commerce imposed export controls on seven heavy rare earth elements, requiring export licenses for shipments of the materials and related compounds, metals, and magnets. The immediate effect was visible within weeks: export volumes fell sharply, European rare earth prices rose to six times their Chinese equivalent, and carmakers in the United States, Europe, and Japan struggled to obtain permanent magnets, with some forced to cut production rates or temporarily shut factories.

In October 2025, China escalated. The new controls expanded the list of controlled elements to twelve, added restrictions on the export of rare earth production equipment and technologies, and introduced the critical new category: case-by-case review for exports of rare earth materials destined for the manufacture of advanced semiconductors. The specifications in the Chinese control regime were almost identical to those in the American export control regime targeting China: logic chips with process nodes of 14 nanometers or below; memory chips with 256 layers or more; AI processors with specific interconnect and compute thresholds. Nvidia's H100 fell within the scope. Apple's A18 Pro fell within the scope. Samsung's ninth-generation V-NAND chips fell within the scope. The South China Morning Post quoted a Chinese analyst calling it "the first time for rare earth export control policies to mention semiconductors" and describing the measures as "a new level of regulatory intensity."

China had also deployed, for the first time, an extraterritorial mechanism modeled on the American foreign direct product rule: foreign-produced items containing at least 0.1 percent Chinese-origin rare earth elements, or produced using Chinese rare earth processing technologies, would require Chinese government approval for export. The mirror held up to Washington was precise and deliberate. You extended your semiconductor controls extraterritorially to constrain our chip supply. We are extending our rare earth controls extraterritorially to constrain yours.

The markets reacted with sufficient alarm to force emergency consultations. The Busan summit in October 2025 produced the one-year suspension: China agreed to delay implementation of the new controls until November 10, 2026, in exchange for a US agreement to suspend planned additional tariffs and extend the broader trade truce. Both sides understood they were buying time rather than resolving the underlying competition. The suspension altered none of the fundamentals, as a January 2026 analysis in War on the Rocks noted explicitly: "The pause quieted financial markets but altered none of the fundamentals. Both sides understood they would revisit the one-year freeze in 2026."

The Summit That Bought More Time

The Beijing summit of May 14 and 15 was, in its strategic logic, a second attempt to buy time before the November deadline. Trump arrived with Elon Musk and Jensen Huang in his delegation, a signal that could not have been more explicit about the commercial stakes the administration was managing. Musk's Tesla depends on rare earth permanent magnets for its drive motors. Huang's Nvidia depends on rare earth inputs for the chips that are the company's entire business. Both men's presence in Beijing communicated, more clearly than any diplomatic communiqué, that the US side understood the chokepoint and was there to negotiate around it.

The summit's public outcomes were warm and vague. Xi declared it "historic" and a "landmark." Both sides agreed to develop a "constructive China-U.S. relationship of strategic stability." Trump said Xi had told him China would not supply military equipment to Iran, which Trump called "a big statement." The trade envoys reported "overall balanced and positive outcomes." Xi warned of potential "clashes and even conflicts" if the Taiwan issue was not handled properly. No specific agreement on rare earth export controls appeared in any public readout.

This absence is the most important fact about the summit. The November deadline is the most consequential near-term structural risk in the global AI supply chain. If the controls are reimposed in November, the case-by-case review process for advanced semiconductor inputs becomes active; Nvidia's supply chain faces mandatory disclosure of end users and applications to Chinese authorities; chipmakers with any defense-adjacent business face heightened scrutiny and potential denial; and the AI buildout that has been priced into global equity markets at roughly five times the magnitude of the dot-com bubble encounters the physical constraint it has assumed away.

The summit produced atmospherics consistent with a second suspension. It did not produce a second suspension. What it produced was the same strategic logic that produced the first: both sides have sufficient incentive to maintain the pause, the pause is structurally temporary, and neither side has resolved the underlying competition that makes the chokepoint real. Trump leaves Beijing having demonstrated that engagement is possible. Xi leaves having demonstrated that engagement is conditional. The November clock continues to run.

The Brazil Mirage

The Trump-Lula meeting on May 7 was, in its strategic logic, the administration's attempt to demonstrate that an alternative to Chinese rare earth supply exists and is being secured. Brazil holds the world's second-largest rare earth reserves, approximately 21 million tons against China's 44 million tons, with only 25 percent of Brazilian territory fully mapped. In late April, USA Rare Earth completed a $2.8 billion acquisition of Serra Verde, Brazil's only commercial-scale rare earth operation, with significant US government equity participation exceeding 15 percent from the Department of Commerce and $565 million in financing from the US Development Finance Corporation. A 15-year offtake agreement with US government agencies and private buyers was established. Washington was not merely courting an ally; it was directly investing in the supply chain it needed.

The Bloomberg headline from May 12 captured the situation precisely: "Don't Buy the Hype on a US-Brazil Rare Earths Deal." The piece noted that Brazil's long history with rare earths "suggests the geopolitical enthusiasm may far exceed the industry's near-term reality." Despite holding the world's second-largest reserves, Brazil has spent decades failing to translate geological potential into sustained production. Environmental licensing delays run between five and ten years. The tailings dam disasters at Mariana in 2015 and Brumadinho in 2019, which killed hundreds of people, have hardened public and regulatory scrutiny of new mining developments. Brazil has the geology. It does not have the processing infrastructure. Building rare earth separation and refining at commercial scale requires the same kind of patient, decades-long industrial investment that China made in the 1980s and 1990s, and that the United States declined to make at the same time.

Lula's response to Trump made the political constraint explicit. Brazil's rare earth reserves are open to investment from any country willing to process minerals domestically. China was not excluded. Trump's desire for preferential US access was declined, politely but clearly. "We have no preference," Lula said. "What we want is to share with whoever wants to invest in Brazil." The Peterson Institute for International Economics described the meeting's outcome as "meh": three hours, lunch, a working group, and no announcements on critical minerals. Serra Verde's $2.8 billion acquisition gives Washington a route into Brazilian production, but it is a route that runs through five-to-ten year permitting timelines, the absence of domestic refining capacity, and a host government that has explicitly declined to offer Washington preferential terms over Beijing.

The Brazil alternative is real in geological terms. It is a mirage in supply-chain terms, on the timescale that the November 2026 clock requires. The gap between the second-largest reserves in the world and a functioning alternative supply chain is measured in years, in capital, in refining infrastructure, and in political will that Lula has clearly indicated he is not prepared to deploy exclusively on Washington's behalf.

The Self-Strengthening Trap, Revisited

The previous dispatch in this series, "The Self-Strengthening Trap," identified a pattern in how complex systems fail when they make public commitments to rise that outrun their capacity to deliver on them. The Late Qing's Self-Strengthening Movement built modern naval vessels and arsenals while leaving the administrative, fiscal, and supply-chain foundations of the empire unreformed. The fleet was modern. The system that supported it was not. When the test came at Weihaiwei in 1895, the gap between the two produced collapse rather than transformation.

The AI buildout has constructed the same pattern at a different scale and a different speed. The public commitment to rise is explicit and enormous: hundreds of billions of dollars in data center construction, model development, and infrastructure deployment; the explicit promises of Nvidia, Microsoft, Google, Amazon, OpenAI, and Meta that AI will transform every domain of human activity within a decade. The concentration index that the previous dispatch identified (Oracle at 2.79, Google at 2.82, comparable to the figures for Western Rome and Late Tang in the Collapse Signature research) measures the degree to which the system's throughput depends on a narrow set of critical inputs whose supply is not controlled by the system making the promise.

Rare earths are the physical expression of that concentration. The AI buildout's capacity (the data centers, the chips, the models) is real and growing. Its throughput (the actual movement of materials from mine to refinery to magnet to server) depends at the critical refining stage on a single country's export licensing decisions. Its resilience (the ability of the system to sustain throughput when the critical input is constrained) is low, because the alternative supply chains exist only as geological potential and diplomatic aspiration, not as operational infrastructure. The Amplification Channels dispatch identified five channels through which financial system stress propagates; one of them was supply-chain disruption in AI infrastructure. The rare earth chokepoint is the supply chain disruption that the IMF was modeling.

The temporal-mismatch pattern here is specific. The AI buildout is proceeding at platform speed: model releases every few months, data center construction measured in quarters, equity market pricing of AI revenues measured in weeks. The diplomatic resolution of the rare earth dependency is proceeding at summit speed: two meetings in seven months, a one-year suspension, a second summit that produced no extension, a November deadline that now belongs entirely to the next round of negotiations. The alternative supply chain development is proceeding at geological and industrial speed: decades for permitting, years for construction, indefinite timelines for refining infrastructure that does not yet exist outside China at commercial scale. None of these three speeds is commensurable with the others. The buildout is already here. The chokepoint is already real. The alternatives are years away, at best.

What to Watch

The series has reached seventeen dispatches by making falsifiable observations rather than sweeping predictions. The rare earth chokepoint is the next structural condition the series will track, and three specific variables will determine whether the dependency resolves, deepens, or produces the kind of threshold event that earlier dispatches have documented in adjacent domains.

The November 2026 deadline. The most important near-term event in the global AI supply chain is not a model release or an earnings report. It is whether China's rare earth export controls are re-suspended, reimposed, or renegotiated by November 10, 2026. A second suspension would buy another year and allow the diplomatic and industrial processes to continue at their current pace. A reimposition would activate the case-by-case review requirement for semiconductor inputs, introducing a new layer of Chinese regulatory discretion into the supply chain of every advanced chip manufacturer in the world. A renegotiation that actually resolves the underlying framework is the least likely outcome but the only one that removes the structural uncertainty. Watch what Beijing demands and what Washington is prepared to concede in the months between the Beijing summit and the November deadline.

The Serra Verde timeline. USA Rare Earth's acquisition of Serra Verde is the most concrete step the United States has taken toward alternative supply. The 15-year offtake agreement is real. The US government equity stake is real. What is not yet real is the refining and separation infrastructure required to turn Brazilian ionic clay deposits into the high-purity rare earth oxides that chip manufacturers need. Serra Verde currently produces mixed rare earth carbonate, a partially processed intermediate product. Converting that into separated oxides requires downstream processing that Brazil does not currently have at commercial scale. Watch whether the US government investment package includes funding for downstream processing infrastructure, and whether Brazilian regulatory approvals for that infrastructure proceed at the speed the November clock requires.

The pricing signal. Since bottoming in late 2024, rare earth concentrate prices have risen more than 50 percent. Baogang Group and Northern Rare Earth, China's two largest producers, raised concentrate prices 37 percent in a coordinated move last year. European rare earth prices reached six times the Chinese domestic price during the April 2025 export control episode. These price signals are the market's current best estimate of the chokepoint's severity. If prices continue to rise through the summer and fall of 2026 as the November deadline approaches, the market is pricing increasing probability that the controls will be reimposed. If prices stabilize or fall, the market is pricing confidence in a second suspension. The price of neodymium oxide and dysprosium oxide in European and American spot markets is the most honest real-time indicator of whether the diplomatic atmospherics from Beijing are translating into actual supply security.

This dispatch connects three threads that the series has been developing independently.

The Amplification Channels dispatch identified supply-chain disruption in AI infrastructure as one of the IMF's five propagation channels for systemic financial risk. At the time of writing, that channel was theoretical. It is now a clock. The one-year suspension expires in November. The market exposure to a reimposition of Chinese rare earth controls on semiconductor inputs is not a scenario in the risk models of most institutional investors; it is an event in the calendar.

The Self-Strengthening Trap dispatch identified the pattern by which systems that make public commitments to rise while leaving structural dependencies unresolved tend to fail at the moment when the dependency is tested rather than transformed. The AI buildout's rare earth dependency is precisely that structural gap: the commitment to rise is public and enormous; the physical foundation of that commitment runs through a chokepoint that neither summit this month resolved.

And the Frozen Conflict dispatches have been documenting how the Iran war accelerated Trump's diplomatic pivot toward Beijing; the war froze the Middle East, freed bandwidth for the China relationship, and made the Beijing summit possible on a timeline that would otherwise have taken longer to construct. The geopolitical and the technological are not separate domains. The Iran war and the rare earth chokepoint are connected through the same executive's calendar, the same diplomatic bandwidth, and the same structural dependency on decisions made in Beijing.

The series will continue to track the November deadline. The chokepoint beneath everything is not beneath the news. It is the news, six months early.