Latest / The AI Supercycle / The Critical Materials Constraint
Transcript
- [0:10]Hello and welcome to the AI Supercycle Podcast, where we follow the capital flows and physical constraints shaping the largest industrial build-out of our generation. From semiconductors, AI factories and hyperscale data centers, to energy markets and power grids, critical materials, orbital compute, and embodied intelligence. We track where the capital is being deployed, where the binding constraints are emerging, and what it means for traders, investors, and the broader economy. Today is Sunday the 28th of June 2026. I'm Tim Hardwick, founder of Quantum Field's Market Intelligence. Over the past two episodes, we've worked through the first two binding constraints in the AI supercycle, high bandwidth memory in episode 2 and delivered power in episode 3. Memory is now a consensus trade. Global chip stocks surged on Thursday after Micron's blockbuster results and investors grew more confident about persistent demand and tightening supply. Power is increasingly on the front page as the debate continues on how to generate and deliver enough power to run the AI data center build out. None of this is new information to anyone listening to this podcast. So this week we arrive at the binding constraint that is not yet consensus. Critical materials, the physical stuff that comes out of the ground. [1:29]And stepping back for a moment, I want to note something about the shape of what we are building here. Over the past three episodes, we have been moving through what is really a physical continuum. Episode 2 took us into the clean rooms where memory is fabricated. Episode 3 took us to the substations and the grid. This week, we go further down, underground, to the mines that supply the raw inputs for everything above, as well as the processing plants, above ground. And next week, we go in the other direction entirely, upward, into orbit, from mines to rocket launches and satellite payloads that are increasingly part of what I call the AI continuum. Every layer in this continuum depends on the ones below it, and can constrain the ones above it. [2:15]A copper shortage constrains transformer production. Transformer shortages constrain grid expansion. Grid constraints limit data center connections. And if the terrestrial layers cannot scale fast enough, the question becomes whether you can bypass them altogether. [2:30] Underground Constraint Begins [2:31]That is the conversation for episode 5. Today we start at the bottom. Because here is the thing about memory and power. You can build a chip fab, you can build a power station, but you cannot build a copper mine where there is no copper. You cannot legislate a lithium deposit into existence. You cannot accelerate geological time. Every chip, every transformer, every cable and every cooling system, Every data centre and every piece of grid infrastructure that we discussed last week requires physical materials that were formed over millions of years and extracted from a finite number of places on Earth, and increasingly processed through a very small number of countries. This is the constraint where the two-clock problem does not just become acute, it becomes geological. [3:18]But before we go underground, let us start with what landed on the desk in the last few days, because the market just gave us a live demonstration of what it looks like when a binding constraint reprices the entire value chain. If you want to understand what is coming for critical materials, start with what just happened to memory. On Wednesday evening, Micron delivered fiscal Q3 earnings that beat expectations on every metric, with gross margins at levels the company has never previously reached. But the structural disclosures carry more weight than the numbers. Micron has signed 16 strategic customer agreements with take-or-pay commitments and substantial customer cash deposits. [4:01]Management stated that the company can currently fulfil only half to two-thirds of customer demand and they have no visibility of when supply catches up with demand. The reasons it's cited are not semiconductor problems. They are industrial infrastructure problems, greenfield fab construction, permitting, labour shortages, energy infrastructure. The same categories of constraint we discussed in episode 3 on power and the same categories we are about to discuss on materials. Then came Thursday afternoon. Apple raised prices across its MacBook and iPad lines, telling customers it could no longer absorb soaring memory and storage chip costs. Price increases ranged from the mid-teens to 25%. Micron rallying and Apple selling off on the same trading day are not separate stories. They are the same binding constraint, repricing simultaneously at both ends of the value chain. However, Apple did bounce back somewhat on Friday, but the signal from Thursday still remains. I bring this up at the start of the episode on critical materials because it's a preview. What just happened to memory is what we can expect to happen progressively across copper, rare earths and the wider critical materials complex. [5:18]The only difference is the timescale. Memory operates on a supply clock measured in quarters. Materials operate on a supply clock measured in decades. When the repricing arrives, it will be slower, but it will also be harder to reverse. [5:32] Iran Shock and Market Moves [5:32]But before we get into critical materials, let's take a quick look at the US and Iran situation and its impact on energy prices. The situation has deteriorated significantly since we last talked about this. The Memorandum of Understanding signed on 17 June did not survive its second week, On Thursday, Iran struck a container ship with a drone as it exited the strait, The US responded with strikes on Iranian missile storage and radar sites on Friday, On Saturday, a tanker carrying Qatari oil was hit in the strait and Iran launched drone attacks on Bahrain, striking a residential building while targeting US military facilities in Kuwait, The escalation has spread geographically, from the Strait to Bahrain and Kuwait, and vertically, from commercial shipping to military infrastructure to civilian targets. Brent Futures reopen tonight and we'll see what the market makes of this. But here is the remarkable thing. Even as this escalation unfolded, oil barely flinched. Brent has fallen from above $119 at the peak of the war to roughly $72 on Friday. The market has round-tripped the entire Iran war premium. [6:47]Also this week, we saw the Kospi had triggered the circuit breakers again on Friday, their fifth time this year. The Apple sell-off transmitted throughout Asia, and Samsung and SK Hynix dragged the index sharply lower. [7:02]We've got two more developments worth flagging. The first is OpenAI. They announced on Friday that it is deferring the full public rollout of GPT 5.6 at the request of the US government, and this limits initial access to a vetted group of partners whose details have been shared with federal authorities. Now, this is the first time a frontier model launch has been explicitly shaped by national security considerations at the government's request. Whatever you think about the policy merits, the signal is unmistakable. [7:35]Frontier AI is being treated as strategic infrastructure, not consumer software. That framing has implications for everything we discuss on this podcast. Because strategic infrastructure gets built with strategic materials, and strategic materials attract strategic policy. And Samsung Group is reportedly preparing to announce a 900 trillion won investment in South Korea over the next decade That's roughly 583 billion dollars Covering chips, AI data centers, batteries and displays, That is an astonishing capital commitment from a single corporate group And every dollar of it will require copper, rare earths, high purity materials and energy infrastructure, The capital formation side of the AI supercycle continues to accelerate. The question this episode asks is whether the physical supply chains can keep pace. We are building the cathedral before we have secured the quarry, and Samsung just announced plans for an even bigger cathedral. [8:40] G7 Mineral Alliance [8:40]Okay, so let's talk about the G7 Critical Minerals Alliance. The most consequential development for our topic today came out of the G7 summit in Evian. Leaders formally launched a Critical Minerals Resilience and Production Alliance, The most structured attempt yet by the Western economies To reduce dependence on China for the materials that underpin modern technology. [9:05]The headline target, reduce reliance on any single non-G7 supplier for rare earths and permanent magnets to below 60% by 2030, with an ambition to reach 50% as soon as possible. There are 195 projects already in motion, attracting investments of roughly 64 billion euros across the critical minerals value chain. [9:29]Australia has formally joined the alliance despite not being a G7 member, which tells you something about how seriously mineral-producing nations are taking this. Now I want to be a bit more measured, a bit more specific about what this actually means. This is a policy framework, not a mine. Policy frameworks do not produce copper. They do not refine rare earths. They do not shorten the 10-20 year timeline between discovering a critical material and producing refined metal. [9:59]What the G7 alliance does is signal political will, and political will eventually becomes capital allocation. But the word eventually is doing a lot of heavy lifting in that sentence. We are going to come back to that timeline problem because it is the central analytical question in everything we discuss today. [10:18] Copper Tariff Watch [10:18]Meanwhile, a discussion with more immediate market impact is arriving in the next few days. The US Commerce Secretary has until June 30th to deliver an updated recommendation to President Trump on whether to impose tariffs on refined copper imports. The United States currently imports roughly 57% of its refined copper consumption. The original Commerce Department proposal called for a phased universal tariff, 15% on refined copper from January 2027, increasing to 30% in 2028. The market is already positioning. The COMEX LME spread has widened. Metal has been flowing into American warehouses, and Trump signed a proclamation on the 1st of June adjusting the broader Section 232 metals framework in ways that look like preparation. Here is why that matters for the AI supercycle specifically. If the United States imposes tariffs on refined copper, the cost of building AI infrastructure in America goes up. [11:18]Every data center requires enormous quantities of copper and so does every piece of grid infrastructure connecting to the electricity network. The hyperscalers can absorb it. Smaller operations and grid builders may find their economics changing materially. And it creates a fascinating tension. The US government is simultaneously trying to accelerate AI infrastructure build-out and making one of its essential inputs more expensive. Industrial policy pulling in two directions at once, which, if you have followed the trade policy for any length of time, should not come as a great surprise. [11:52]And then there is China. The suspension of Beijing's expanded rare earth and critical mineral export controls, agreed following the Trump-Z talks last October, expires in November. Those controls are suspended, not withdrawn. The underlying architecture is intact, and critically, the April 2025 controls on seven heavy rare-earth elements, the ones most essential for high-performance permanent magnets, were never suspended at all. CSIS research showed that U.S. yttrium imports from China collapsed from over 333 metric tons in the eight months before April restrictions to just 17 metric tons after. The IEA estimates that a full re-implementation of China's export control framework could put $6.5 trillion of annual economic activity outside China at risk. When you consider that China accounts for roughly 70% of global rare earth mining, 90% of processing, and 94% of permanent magnet manufacturing. We're going to spend some more time on this, because the geopolitical dimension of the materials constraint is inseparable from the physical one. [13:04] Mining Runs on Geological Time [13:05]Okay, so let's move on. In the last two episodes, I described the two clock problem as a mismatch between the financial clock, which reprices in minutes, days and weeks, and the physical constraint clock, which reprices over years. Memory operates on a supply clock measured in quarters. Power operates on a supply clock measured in years to decades. Materials operates on a supply clock measured in decades to geological time. So let me put some specific numbers on this. From the point at which a geologist identifies a promising copper deposit to the point at which refined copper is being produced, the typical timeline is between 12 and 25 years. That includes exploration, resource definition, feasibility studies, environmental assessments, permitting, community consultation, financing, construction and commissioning. [14:00]The Cayoveco copper mine in Peru is one of the most recent large-scale greenfield projects to reach production. It was discovered in the 1960s and it began commercial production in 2022. 60 years. Even well-funded projects with political support routinely take 12 to 15 years. There are no shortcuts. You cannot compress the geological assessment. You cannot skip the environmental permitting. And unlike a chip fab, where you are building on a flat site with established infrastructure, mines are located wherever the geology put the ore, which is frequently in remote locations with limited infrastructure, complex water requirements, and communities whose consent is not optional. The AI industry operates on a capital deployment clock measured in months. The hyperscalers can go from board approval to operational data center in 18 to 36 months. The mining industry operates on a capital deployment clock measured in decades. This is the most extreme version of the two-clock problem in the entire AI supercycle, and it is the one that the market is least prepared for. [15:08] Copper Under Strain [15:09]Okay, so let's talk a little bit about copper. And if you want to understand the material's constraint in the AI supercycle, start with copper. Copper is to electrical infrastructure, what oxygen is, to breathing. Technically, you could discuss alternatives. But practically, everything stops without it. Copper closed Friday at just over $6.14 per pound on COMEX. That is down roughly 9% from the record high of $6.71 on the 13th of May. The pullback has been driven by a stronger dollar and hawkish-fed expectations, as well as other factors in there too. But zoom out and the picture is very different. Copper is up sharply year on year. And the structural story is only getting tighter. [15:53]Goldman Sachs recently raised its year-end 2026 copper forecast and significantly increased its estimate for the global copper deficit. The IEA's analysis shows that copper concentrate treatment charges, the fees smelters charge to process ore, have fallen to zero in the 2026 annual benchmark. Chinese smelters, which account for half of global refined copper output, have agreed to cut production by over 10% because of the economics of processing have become unworkable. When smelter fees go to zero, it tells you that the competition for raw material has become so intense that processors are willing to work for nothing just to keep their plants running. That is not a sign of a balanced market. Now, why does this matter to AI? A single hyperscale data center campus uses thousands of tons of copper in cabling, busbars, power distribution and cooling systems. But the data center itself is only part of the story. [16:55]Remember what we discussed last week. Every data center needs grid connection. The transformers, the transmission lines, the substations, the power generation facilities. All copper intensive at every stage. Grid and power infrastructure are projected to drive more than 60% of copper demand growth through 2030. S&P Global projects that AI data centres alone could account for 2.5 million metric tonnes of annual copper demand by 2040. And this is layered on top of the electrification of transport, the build-out of renewable energy, defence spending and the reshoring of manufacturing, all competing for the same metal at the same time. There is a structural point here that I think the market is underestimating. The analysts covering AI infrastructure and the analysts covering mining are, for the most part, different people sitting at different desks, reading different research, and attending different conferences. The AI analysts model demand growth, the mining analysts model supply response. Very few are modeling the collision between the two on the same page. The analytical gap is where the mispricing lives. [18:07] Uranium Supply Tightens [18:08]Okay, so let's move on and talk about uranium, the supply-side story. In episode 3, we discussed uranium in the context of nuclear power generation, a strategic energy source for AI data centers. Today, we return to uranium from the other side, the mining and supply chain perspective, because the materials constraint here is a distinct analytical question from the power constraint. Uranium is trading at roughly $85.50 per pound, maintaining a narrow range since early April after erasing a surge earlier this year. The spot price hit $101 in late January before selling off on broader risk events. It has now settled into this range and the spot market looks quiet. But do not let the spot market fool you. The real story is the term market. Long-term contract prices have climbed to $94 per pound, the highest since 2008. Cameco's president noted at PDAC earlier this year that the volume of uncovered utility requirements, future uranium demand not yet secured by contracts, has reached record levels. The forward demand that has yet to come to market has, in his words, never been bigger. [19:22]Now layer on the AI demand. As we covered last week, Hyperscale is signing power purchase agreements directly with nuclear operators. Microsoft has the Three Mile Island restart. Amazon has deals with Talon Energy at Susquehanna. Google is contracting with Kairos Power for small modular reactor capacity. [19:42]Meta and Microsoft have both signed agreements for fresh nuclear capacity as well. Every one of these agreements increases the demand for uranium fuel over multi-decade operating lifetimes. The supply side is struggling to respond. Kazakhstan remains the dominant producer, accounting for roughly 40% of global output, with all of the geopolitical concentration risk that implies. Canadian production from the Athabasca Basin delivers some of the highest-grade ore in the world, but new projects face lengthy permitting timelines. The Sprott Physical Uranium Trust has been buying aggressively, purchasing over £5 million year-to-date, which tightens the spot market further. Here is the structural point. Nuclear power stations have operating lifetimes of 40 to 60 years. When a hyperscaler signs a power purchase agreement with a nuclear operator, they are creating uranium demand that extends decades into the future. The AI supercycle is not just creating a one-off spike in uranium demand, it is adding a permanent structural layer to the demand base. And unlike oil or gas, where a price spike can unlock shale production within quarters, uranium supply expansion operates on a mining clock that is measured in years at the absolute minimum. The two-clock problem again, with geology setting the pace. [21:02] Critical Materials Wideview [21:03]So now let's turn our attention to the wider critical materials complex. Copper and uranium are the largest and most liquid commodity expressions of the materials constraint, but the critical materials story extends well beyond these. Rare earth elements are perhaps the most strategically sensitive. Global rare earth oxide demand is estimated at roughly 170 to 180,000 metric tonnes, with projections suggesting growth to over 200,000 metric tonnes by 2030. The elements themselves are not actually rare in the geological sense. The name is a historical accident. What is rare is the capacity to mine, separate and refine them at the industrial scale outside of China. This is the distinction that matters, and it is the distinction that makes the G7 Alliance announcement very significant if the political commitment translates into actual processing capacity, which it has not done yet. Gallium and germanium are essential for semiconductors and optical transceivers, the components that enable high-speed data transfer within and between data centers. [22:10]China produces approximately 98% of the world's refined gallium supply. Following the initial export controls in 2023, prices spiked 25-30%. The current suspension of enhanced controls runs until November 2026. After that, the architecture is in place to re-impose them at any time. And then there is tin, which has emerged as a somewhat unexpected critical material in the AI build. TIN's primary application is solder, the metallic glue that connects components on every printed circuit board in every server, GPU and networking switch. It accounts for over 50% of global TIN consumption. With 190 GW of new hyperscale data center capacity announced as of early 2026, the demand for solder, and therefore TIN, has surged. The supply chain is concentrated in Asia, with China accounting for roughly half of the global refined supply and seven of the ten largest smelters located in the region. It is not the kind of commodity that makes headlines, but it is the kind of commodity that can delay a server rack. The common thread across all of these materials is concentration. [23:23]Concentration of geological deposits. Concentration of processing capacity. And concentration of geopolitical exposure. The AI supercycle is the most materials-intensive technology build-out in history, and it is arriving into supply chains that were already strained by the energy transition, defence spending and the reshoring of manufacturing. There are only so many atoms to go around and everyone wants them at the same time. [23:52] China’s Control Leverage [23:52]So now let's take a look at the geopolitical dimension. We covered Beijing's export control architecture and the market share in the breaking developments, so I'll not repeat those numbers. But what I do want to draw out is the strategic logic because it transforms a supply chain problem into something qualitatively different. This is not a blunt instrument. It is a precision tool. The controls do not ban exports outright. They create a licensing framework that allows Beijing to map technology flows, monitor end-users, and selectively restrict supply at moments of geopolitical leverage. The State Council's new provisions on the security of industrial and supply chains, announced publicly in March 2026, integrate export controls, countermeasures, data security obligations, and investment screening under a unified national security mandate. The architecture is designed for adaptive, multi-domain escalation, the ability to tighten or loosen controls across multiple materials simultaneously in response to diplomatic conditions. For the AI supercycle, this creates a risk category that does not exist in the memory or power constraints. [25:04]HBM supply is concentrated in South Korea, a US ally. Power generation is largely a domestic infrastructure question, but critical materials pass through processing choke points controlled by a strategic competitor. The physical constraint and the geopolitical constraint are the same constraint, viewed from different angles. The G7 alliance is the beginning of a response, but reducing China's share from 90% to below 60% requires building processing capacity that does not currently exist, in jurisdictions where permitting, environmental standards and capital costs are significantly higher. This is a decade-long project at minimum. In the interim, the supply chain remains exposed. [25:47] Base Case Outlook [25:48]Okay, so let's move on to the base case on critical materials, the part of the episode where the framework commits to a reading. My base case on the critical materials constraint over the next three to five years is that it tightens progressively and that the market has barely begun to price the structural nature of the supply-demand mismatch. Unlike memory, where supply additions are measured in quarters, and power, where supply additions are measured in years, The materials constraint operates on a geological and geopolitical clock that offers no quick fixes. On copper, the structural deficit is widening. New mine supply cannot respond to the simultaneous demands of AI infrastructure, grid expansion, electrification and manufacturing reshoring on any timeline shorter than a decade. [26:34]The US tariff decision on refined copper due in the next week will determine the near-term price dynamics but does not alter the underlying supply-demand equation. Copper prices have meaningful further upside over a multi-year horizon, with the risk skewed to the upside by supply disruptions and policy-driven demand acceleration. On uranium, the demand base has been permanently expanded by hyperscalar nuclear commitments, and the supply side is constrained by the same geological timelines that govern all mining. The long-term contract price at $90 tells you more about the real state of the market than the spot price at $85. On rare earths and the wider critical materials complex, the concentration risk is severe and the timeline for diversification is long. The G7 alliance is a necessary step but not a sufficient one. [27:26]The November 2026 expiry of China's export control suspension is a hard catalyst. Markets will begin positioning well in advance. Any deterioration in US-China relations before that date could accelerate the timeline considerably. The framework assigns a probability of approximately 65% that the critical materials constraint becomes a significant market factor within the next 18 months. Defined as a supply disruption or geopolitical escalation that reprices materials and exposed equities by more than 15%. [28:01]The principal risk to this view would be a sustained improvement in US-China relations that leads to a permanent withdrawal rather than mere suspension of export controls, combined with a meaningful slowdown in hyperscaler capital expenditure. Both appear unlikely at present, but the range of outcomes on the geopolitical dimension is unusually wide. The indicators I am watching most closely are the June 30 copper tariff recommendation, the COMEX LME copper spread, uranium term contract pricing, rare-earth pricing differentials between Chinese and non-Chinese sources, and any changes to MOFCOM's export licensing framework ahead of the November expiry. So this is the base case as it reads today. As new evidence arrives, we'll revise this. [28:48] The Three Constraints [28:48]So that is the materials constraint, the third binding constraint on the AI supercycle, and in many ways the hardest to solve. You can build a fab, you can build a power station, but you can't build a copper deposit. The three binding constraints we have now covered, memory, power and materials, form an interconnected system. Memory requires power. Power requires materials. Materials require capital, permitting and geopolitical stability. A bottleneck in any one of them constrains the others. And as Micron and Apple demonstrated this week, when these constraints bind, they reprice the entire value chain. [29:30]The AI super cycle is tightening all three at the same time. You can find all of our work at qfmi.substack.com The Market Pulse and the Spotlight articles are free and always will be. The Weekly Outlook, the Weekend Debrief and the Monthly Strategic Research Report sit behind a paid subscription. Subscribe and you get the full picture. Next week, in episode 5, we will cover the latest market developments and begin exploring an emerging thesis in the AI supercycle, orbital data centers. What happens when the constraints we have spent three episodes discussing, power, materials, grid infrastructure, which are all terrestrial problems, and someone offers a pathway that bypasses all of them? That episode's going to be a good one. Thank you for listening to the AI Supercycle podcast. See you next week.