AI Runs on Rare Earth Metals. China Has Them.

I use AI tools every single day building products, but I never once think about obscure elements like neodymium, lanthanum, or cerium.

But I really should think about them more often. AI fundamentally runs on hardware that requires rare earth metals for manufacturing, and almost all of these critical materials come from China. That’s a genuinely big problem.

Table of contents

What Are Rare Earth Metals?

Rare earth metals aren’t actually rare in the sense of being uncommon in Earth’s crust, they exist in small quantities almost everywhere. What’s genuinely rare is finding economically viable concentrations that can be profitably mined and processed.

There are 17 different rare earth elements with genuinely weird names like dysprosium, terbium, and praseodymium. These elements get used in essentially all modern electronics including display screens, hard drives, batteries, computer chips, and server infrastructure.

You literally cannot build modern AI hardware without reliable access to these materials.

Why AI Needs Rare Earths

Training large AI models requires absolutely massive compute infrastructure running 24/7. Modern GPUs and TPUs rely heavily on rare earth elements throughout their manufacturing process, data centers need them for efficient cooling systems, power distribution, and networking equipment, massive storage systems depend on them for magnetic storage, and overall power efficiency depends critically on rare earth permanent magnets.

The dependency chain is brutally simple: no metals means no chips, and no chips means no AI development at scale.

China Controls the Supply Chain

China currently controls approximately 70% of global rare earth production and an astonishing 90% of all rare earth processing and refining capacity.

This dominance isn’t because China happens to have all the rare earth deposits in the world, since the US, Australia, and other countries have substantial proven reserves too. China strategically invested in extraction and processing infrastructure for decades while other countries chose not to, willingly accepted significant environmental costs that other nations weren’t willing to bear, systematically built unmatched expertise and technical knowledge, and scaled production to levels nobody else could match.

Now they completely control global supply chains.

What That Control Means

Back in 2010, China deliberately restricted rare earth exports to Japan during a diplomatic dispute, and the results were immediate where global prices spiked dramatically and electronics manufacturers worldwide panicked about supply security.

China has exactly the same leverage over AI hardware manufacturing right now. If geopolitical tensions escalate significantly, what realistically stops China from implementing restrictions again? They could impose higher export prices, prioritize domestic Chinese companies over foreign buyers, or restrict exports entirely. Absolutely nothing prevents this scenario.

American AI companies desperately need hardware to train models and serve customers, that hardware fundamentally needs rare earth materials for manufacturing, and China overwhelmingly controls those rare earth materials. This creates a massive strategic vulnerability that almost nobody in the AI industry actively thinks about.

The Chip Chokepoint

The US government has aggressively restricted advanced chip exports to China, so NVIDIA can’t sell their latest GPUs there and TSMC can’t manufacture cutting-edge chips for Chinese companies.

But these export restrictions only work effectively if the US and allies can manufacture chips independently without Chinese cooperation. If China retaliates by restricting rare earth exports, then nobody can manufacture at scale including not just the US but also Taiwan and South Korea.

This creates mutual dependency where China desperately needs advanced chips for AI development while the US desperately needs rare earth materials for chip manufacturing. It’s a tense standoff currently, but if real conflict erupts both sides lose catastrophically and global AI development grinds to a halt.

What the US Is Trying to Do

MP Materials operates the only active US rare earth mine: Located at Mountain Pass, California, it successfully mines raw ore but then ships everything to China for processing because there’s no domestic processing capacity at scale.

Investment in domestic processing is finally happening: The US is beginning to build domestic processing facilities, but this takes many years or potentially a full decade to match China’s current capacity and expertise.

Recycling could recover substantial materials: Old discarded electronics contain significant rare earth content that we currently just throw away wastefully, but serious recycling efforts are finally starting to change this.

Allied cooperation helps diversify but remains incomplete: Australia and Canada are working to diversify the global supply chain which is definitely better than full China dependence, but the alternative supply chains remain fragile and incomplete.

All these initiatives are still many years away from achieving genuine independence from Chinese supply chains.

Why This Matters for AI Development

If you’re building AI tools right now, you’re probably thinking intensely about model selection, API costs, and prompt engineering strategies, but you’re almost certainly not thinking at all about global supply chains and rare earth metal availability.

But geopolitics fundamentally shapes what’s actually possible in AI development, and the connections are direct and immediate. If rare earth prices spike dramatically due to supply restrictions, GPU manufacturing costs spike correspondingly, which means compute costs increase substantially across the board, which ultimately prices smaller companies and independent developers completely out of the market. If China restricts rare earth supply entirely in response to escalating tensions, nobody can manufacture GPUs at scale regardless of how much money they have available, which means you literally can’t get the hardware needed to scale your AI applications. If the US and China fully decouple their technology ecosystems, we end up with two completely separate AI development worlds with incompatible technical standards and fundamentally different values baked into the systems.

Your supply chain vulnerability isn’t just a boring logistics problem to delegate to operations teams, it’s a strategic constraint that directly limits what’s possible in AI progress and development.

The Environmental Reality

Rare earth mining is genuinely dirty work that nobody wants in their backyard. The extraction process produces radioactive waste materials that require careful disposal, creates toxic chemical runoff that can contaminate groundwater for decades, and causes widespread environmental damage to surrounding ecosystems that takes generations to recover from.

China strategically accepted these substantial environmental costs in order to dominate global rare earth production. The United States deliberately chose not to develop domestic mining capacity primarily due to legitimate environmental concerns and strong public opposition to having mines in their communities. That decision is completely understandable from an environmental protection perspective.

But the current situation creates obvious hypocrisy where we depend entirely on China doing the environmentally dirty work of mining and processing rare earths while we build supposedly “clean” AI applications using their materials and pretending our hands are clean.

If we genuinely want strategic independence from Chinese supply chains, we need to honestly pick one of three difficult options: accept the environmental costs of domestic mining with strong modern regulations to minimize damage, continue our current dependence on China indefinitely, or pay dramatically higher hardware costs for limited supplies from expensive alternative sources. We absolutely can’t avoid all three uncomfortable choices simultaneously.

What I Think Needs to Happen

We desperately need to build domestic rare earth processing capacity, not just mining operations, because processing is the actual bottleneck that China controls. We need to implement electronics recycling programs at massive scale to recover valuable rare earth materials from millions of tons of discarded devices currently sitting in landfills. We need sustained research investment into alternative chip designs and manufacturing processes that require substantially fewer rare earth materials. We need to honestly accept the environmental trade-offs of domestic production while implementing strong modern regulations to minimize damage. Most urgently, we need to dramatically speed up our timeline because waiting a full decade for independence is far too slow given current geopolitical tensions.

The Builder Perspective

As someone building AI products professionally, I want cheap and fast compute resources, reliable supply chains that never interrupt development, and US strategic independence from authoritarian control. I want AI development that isn’t held hostage by geopolitical tensions or trade wars.

But those goals fundamentally conflict with each other in ways I can’t ignore. Cheap reliable compute means continuing to rely heavily on China’s rare earth supply chains. Achieving genuine strategic independence means accepting substantially higher hardware costs and slower development timelines during the transition period.

There’s no easy answer that satisfies all competing priorities simultaneously. Ignoring the problem definitely doesn’t help anything. Every time I spin up a new model for training or inference, I’m using hardware manufactured from rare earth materials extracted and processed by a country that the United States is increasingly in serious geopolitical conflict with. That dependency isn’t remotely sustainable long-term.

What You Should Know

If you’re building AI products, you need to actively watch supply chain developments because they’ll directly affect your compute costs and technical capabilities going forward. If you’re investing in AI companies, you should ask hard questions about their compute dependencies because without reliable GPU access they literally can’t scale regardless of how good their technology is. If you’re making technology policy, you need to push for dramatically faster domestic rare earth production timelines. If you’re just using AI tools as an end user, you should understand that the screen magic you take for granted depends entirely on obscure metals mined in countries you’ve probably never visited, processed through supply chains shaped by geopolitical forces you can’t control.

The Bottom Line

AI is ultimately software running on physical hardware. That hardware fundamentally requires rare earth metals for manufacturing at scale. China overwhelmingly controls global rare earth extraction, processing, and supply chains.

This creates a massive strategic problem for US AI development, for American independence from authoritarian influence, and for anyone who wants to build AI applications without constant geopolitical constraints hanging over their work. Solutions definitely exist including domestic processing capacity, recycling programs, and alternative materials research, but all of these solutions take many years to implement at scale. That’s time we might genuinely not have given rapidly escalating tensions.

Yeah, I think about obscure elements like neodymium now when I’m building with AI. You probably should too.