China AI Supply Chain Chokehold: Gallium, Rare Earths, and the Materials War | The ByteDive

Your phone screen, your car’s radar, even the cooling system inside NVIDIA’s latest GPU — they all depend on a metal you’ve probably never heard of. China controls 98% of it. And as of March 2026, its price has risen tenfold.

This isn’t about chips anymore. The semiconductor story has moved upstream — past the fabs, past the lithography machines, all the way to the raw materials that make everything possible.

In Part 1 of this series, we dissected how HBM memory became the bottleneck of the AI revolution. In Part 2, we followed ASML’s EUV monopoly — the single machine that decides who can build cutting-edge chips.

Now we go deeper. Below the fabs, below the equipment, there’s a layer most people never think about: the raw materials. Gallium. Germanium. Rare earths. Synthetic diamonds. Four categories of materials, one country controlling almost all of them.

Think of it like a building. HBM is the penthouse suite. EUV lithography is the elevator. But the materials we’re about to discuss? They’re the foundation. Crack the foundation, and every floor above it collapses.

This is the story of how China turned the periodic table into a geopolitical weapon — and what the rest of the world is scrambling to do about it.

TL;DR — China controls the raw materials that make AI chips possible.

• Gallium retail prices surged to $2,101/kg (investment-grade; industrial benchmark ~$245/kg) since 2023 as China weaponized export controls on four critical materials.
• Synthetic diamonds are emerging as the breakthrough thermal solution for AI chips, with a $600M US plant under construction.
• The US has virtually zero strategic stockpile of heavy rare earths while Japan holds up to 180 days — a gap that could decide who wins the AI race.

The Invisible China AI Supply Chain Chokehold: Four Materials China Controls

Let’s start with a number that should unsettle anyone in the semiconductor industry: 98%. That’s China’s share of global gallium production (CSIS). Not 50%. Not 70%. Ninety-eight percent.

Gallium is what makes your 5G phone antenna work. It’s the backbone of GaN (gallium nitride) power chips that manage energy in data centers. The US Department of Defense says 11,000 military components require gallium compounds, and 85% of those supply chains include at least one Chinese supplier (CSIS).

But here’s the thing — gallium isn’t even the scariest one. It’s what you might call the “weak card” in China’s hand. The real weapon sits further down the periodic table.

The China AI Supply Chain Chokehold by the Numbers

Before we go deeper, let’s see the full picture:

(Sources: USGS MCS 2025, IEA, CSIS)

Notice the pattern. Mining share is high. But refining share is even higher. China doesn’t just dig these materials out of the ground — it processes them into usable forms that nobody else can replicate at scale.

This is the critical insight: mining is solvable. You can find gallium deposits in Australia, bauxite residues in Germany, ores in Greece. But refining? That’s where China’s real moat lies. It takes years of infrastructure investment, specialized chemistry knowledge, and massive scale to build refining capacity.

How China Weaponized the Periodic Table

China’s approach has been methodical. Not a sudden shutdown, but a slow squeeze — like gradually turning a valve:

• August 2023: Export licensing requirements for gallium and germanium (Fastmarkets)
• December 2024: Full ban on gallium/germanium exports to the US
• April 2025: Export controls on 7 heavy rare earths, including dysprosium and terbium (CSIS)
• November 2025: Temporary suspension of US-bound rare earth exports
• 2026.11.27: Current suspension expiry date — the clock is ticking

The dysprosium play deserves special attention. NdFeB permanent magnets — the kind that power EV motors, wind turbines, and precision-guided munitions — require dysprosium for heat resistance. China’s new rule: if a magnet contains even 0.1% Chinese-origin content, it triggers a licensing requirement (CSIS). That’s not a trade barrier. That’s a tripwire.

The Price of Dependence: China AI Supply Chain Chokehold Sends Gallium 10x Higher

Numbers don’t lie. Here’s what happened to gallium prices as China tightened the screws:

(Sources: TradingEconomics, Fastmarkets, Skillings, Strategic Metals Invest)

The retail/investment-grade gallium price — what small-lot buyers pay in European warehouses — has surged tenfold. The industrial benchmark (SMM) sits lower at roughly $245/kg, but even that reflects a doubling from pre-control levels. Germanium has been even more volatile — reaching $8,597/kg retail, more than doubling in 2025 alone (TradingEconomics).

Here’s the math that keeps defense planners awake at night: global gallium demand runs about 700 tons per year. Non-China production capacity? Roughly 170 tons — just 24% of demand. That’s not a supply chain risk. That’s structural dependence.

The Diamond Gambit: When Thermal Management Becomes Geopolitics

Now here’s where the story takes an unexpected turn. One of the most promising solutions to the AI chip thermal crisis comes from a material associated with jewelry: diamonds.

Not the kind you put on a ring. Synthetic diamonds — lab-grown crystals with a thermal conductivity of 2,200-2,400 W/m-K. That’s six times better than copper (IEEE Spectrum). When you’re running a GPU cluster at full capacity for AI training, heat is the enemy. Diamond heat spreaders can reduce AI chip peak temperatures by up to 30%.

Enter Element Six, a subsidiary of De Beers. In February 2026, they broke ground on a $600 million synthetic diamond grit manufacturing plant in Georgia — the first major project under a US-Japan $550 billion infrastructure investment deal (CBS Atlanta, Japan Times).

Meanwhile, an Intel/IBM/Mitsubishi consortium is developing open-standard diamond packaging specifications. Akash Systems already deploys “Diamond Cooled” servers that reduce GPU temperatures by 10-20 degrees Celsius (IEEE Spectrum). The technology works. The question is whether production can scale fast enough.

The TSMC Retreat and Samsung’s China AI Supply Chain Chokehold Gamble

The real-world impact of China’s materials pressure is already reshaping the semiconductor industry. Case in point: TSMC’s GaN exit.

In early 2026, reports confirmed that TSMC plans to exit the GaN (gallium nitride) foundry business by mid-2027 (Digitimes). The reasons: Chinese competitors offering GaN chips at dramatically lower prices, combined with gallium supply instability making long-term cost planning impossible.

GaN is not a niche market. Power semiconductors built on GaN are critical for EV chargers, 5G base stations, and data center power management. The market is projected to grow from $355 million in 2024 to $3 billion by 2030 — a 42% CAGR (Yole Group).

Samsung, interestingly, is moving in the opposite direction. The Korean giant began offering 8-inch GaN foundry services in 2025, betting that its closer proximity to alternative gallium supply chains gives it an edge that TSMC lacks.

The Stockpile Race: China AI Supply Chain Chokehold Drives Strategic Reserves

If China’s materials dominance is the disease, stockpiling is the emergency treatment. And the diagnosis varies wildly by country.

(Sources: USGS, Fastmarkets, JOGMEC, Al Jazeera)

Read that first row again. The United States — the country spending the most on AI infrastructure, pouring hundreds of billions into data centers and chips — has “virtually zero” heavy rare earth stockpile (USGS 2026).

Project Vault, announced in early 2026, aims to fix this with $12 billion in funding. The target: build a 60-day strategic reserve. But 60 days is a band-aid when your adversary controls 91% of refining capacity.

The FORGE Alliance and Korea’s Strategic Position

In February 2026, the US launched the FORGE Initiative — a 55-country alliance (plus the EU) designed to secure critical mineral supply chains (Fastmarkets). South Korea was named inaugural chair.

Korea’s strategy extends beyond stockpiling. The country is investing in urban mining, deep-sea mineral exploration, and bilateral deals with Australia and Canada. The Saemangeum rare earth processing facility, budgeted at KRW 240 billion, aims to create domestic refining capacity — directly attacking China’s most defended advantage.

The Alternative Pipeline: Can Anyone Break China’s AI Supply Chain Chokehold?

The honest answer: not quickly. But the pieces are moving.

MP Materials, the only active rare earth mine in the US (Mountain Pass, California), hit production records in 2025 — 2,599 tons of NdPr, doubling year-over-year (Globe and Mail).

Australia’s Lynas Rare Earths produced 16,100 tons of rare earth oxide in 2026 fiscal year, up 53% year-over-year, beginning separation of heavy rare earths like dysprosium, terbium, and samarium.

The EU’s Critical Raw Materials Act sets an ambitious target: reduce gallium dependence on China from 71% to 17% by 2030 (EU Commission). That’s a 54-percentage-point swing in four years.

But here’s the uncomfortable reality: these projects measure progress in years. China’s export controls can change overnight.

What the China AI Supply Chain Chokehold Means for Your Career

If you work anywhere in tech, this story matters more than the latest LLM benchmark. AI models need chips. Chips need materials. And the materials are controlled by a single country that has shown it’s willing to weaponize them.

For the semiconductor industry: watch Samsung’s GaN bet closely. If Samsung succeeds where TSMC retreated, it validates the thesis that materials supply chain proximity matters as much as fabrication technology.

For the AI infrastructure buildout: the $300 billion+ being poured into data centers assumes materials will be available. Diamond thermal management, GaN power delivery, rare earth magnets for cooling fans — every layer depends on materials China dominates.

For investors: the NVIDIA investment thesis and the entire AI security landscape both carry materials risk that’s rarely priced in.

For career strategy: materials science, supply chain management, and geopolitical risk analysis are becoming core competencies in tech — not just in mining companies.

Bottom Line. The AI revolution runs on materials most people have never heard of. When a single country controls 98% of one, 91% of another, and 85% of a third, the supply chain isn’t fragile — it’s a weapon waiting to be used. The race to build alternatives has begun, but the timeline mismatch between China’s ability to restrict exports (overnight) and the West’s ability to build refining capacity (years) is the most dangerous gap in the entire AI value chain.

Career Takeaway. The semiconductor industry is no longer just about design and fabrication. Materials sourcing, stockpile strategy, and geopolitical risk are becoming first-order variables. The professionals who see the full stack, from raw materials to AI models, will navigate the next decade better than those who only watch the software layer.

References

1. CSIS — “Beyond Rare Earths: China’s Gallium Supply Chains and the AI Race”
2. Fastmarkets — “China gallium germanium export ban suspension timeline 2025-2026”
3. Tom’s Hardware — “Chip material prices double as China tightens controls”
4. TradingEconomics — “Gallium and Germanium commodity prices March 2026”
5. CBS Atlanta — “Element Six breaks ground on $600M synthetic diamond plant”
6. Japan Times — “US-Japan $550B investment deal”
7. IEEE Spectrum — “Diamond thermal management for next-generation AI chips”
8. TechCrunch — “Element Six and Orbray achieve diamond substrate milestone”
9. CSIS — “China’s rare earth magnet export restrictions”
10. Digitimes — “TSMC to exit GaN foundry business by mid-2027”
11. Fastmarkets — “Project Vault: US $12B strategic mineral stockpile initiative”
12. USGS — “Mineral Commodity Summaries 2025”
13. JOGMEC — “Japan national stockpiling system”
14. EU Commission — “Critical Raw Materials Act 2026”
15. Globe and Mail — “MP Materials hits rare earth production records”
16. Lynas Rare Earths — “FY2026 production results”
17. Yole Group — “GaN power semiconductor market forecast”

This article is part of the Semiconductor Anatomy Series (S1) — a deep dive into the layers that make the AI revolution physically possible.

Disclaimer: This article contains references to investment themes and company valuations. It is intended for informational and educational purposes only and does not constitute investment advice.