Tearing apart the metal holding every AI chip together

Tin.

The financial press loves to call it a "niche metal." Bloomberg runs it in a sidebar. CNBC doesn't run it at all. When tin hit record highs in January, CRU's headline was that prices looked "disconnected from fundamentals" — that Chinese speculative flows were driving it, not real supply-demand mechanics. Classic. The small market gets the small coverage and the small coverage produces the wrong conclusion.

Here's what they missed while they were looking at the trading screen: tin is the nervous system of every piece of computing hardware on the planet. Over 50% of global tin consumption goes into solder — the material that physically connects semiconductor chips to circuit boards. Every GPU Nvidia ships. Every server rack Microsoft bolts into a data center. Every AI chip that TSMC packages using its latest stacking technology. None of it works without tin solder joints holding the architecture together.

The old model said tin tracks consumer electronics cycles. Smartphone shipments up, tin up. That was mostly true in 2018. I leaned on that framework myself for a while. But the demand base has shifted underneath everyone. As chip stacking density rises with each generation of AI hardware, the amount of tin each unit requires goes up, not sideways. CITIC Securities estimates AI data servers alone will generate roughly 2,500 tonnes of new tin consumption this year. That's incremental demand nobody's models were built to capture.

The sell-side is still pricing tin as a consumer electronics derivative. The machine has quietly become an AI infrastructure input. And the supply side — the supply side is where it gets ugly.

Story off. Numbers on.

Six thousand tonnes. That's what's sitting in LME-approved warehouses as of July 31. For context, the world consumes roughly 390,000 tonnes of refined tin per year. LME deliverable stock now represents about five days of global demand. Five days. And it's been draining for 22 straight sessions with no sign of reversal.

The supply architecture is broken in three places simultaneously. Start with Myanmar. The Man Maw mine — the country's largest tin-producing asset and historically the primary feeder of concentrate to Yunnan smelters — has been suspended since August 2023. Concentrate exports from Myanmar to China collapsed 77% year-on-year through July 2025. Chinese smelters are running below 70% capacity because they literally cannot get enough ore to process.

The Wa authorities formalized a dewatering cost-sharing mechanism in February, splitting pumping costs across 11 mine portals. That sounds like progress until you realize they're still dewatering — not mining. The ITA's optimistic case has Myanmar producing maybe 20,000 tonnes of contained tin this year. The old run rate was multiples of that. And "optimistic" is doing a lot of heavy lifting in that sentence.

Second fracture: the DRC. The Bisie mine in North Kivu accounts for nearly 6% of global tin output. M23 rebel forces already forced a five-week shutdown of operations in early 2025. The mine has since resumed at full capacity, but Bisie sits in contested territory where any renewed advance could halt production again. Together, Myanmar and the DRC represent 20% of global production and 60% of Chinese tin ore imports. Myanmar is offline; the DRC is one offensive away from the same.

Third fracture: Indonesia. The world's largest tin exporter. In February, the energy minister announced the country is studying a complete ban on tin exports, following the nickel playbook. No timeline set. But PT Timah — which controls 81% of the country's tin mining land — sits under state holding company MIND ID, which has already been renamed and reorganized under the new downstream-processing mandate. When Jakarta says "we're studying it," the permit architecture is usually already being rewritten behind the scenes.

Coface projects tin supply will grow 3% this year while demand rises 3.5%. That gap may look narrow on a spreadsheet. It does not feel narrow when you're the buyer trying to secure physical delivery and LME has five days of stock left.

The Indonesia angle is the one that keeps me up. Myanmar is priced in. The DRC is priced in — badly, but in. Indonesia is not priced in, because the market is treating the export ban study as political noise. They did the same thing when Jakarta floated the nickel ban. That wasn't noise either.

Indonesia is the world's largest refined tin exporter. Bangka Belitung province alone accounts for over 80% of the country's tin mining land. When the energy minister stood up in February and said "we will no longer allow exports of raw materials," he wasn't improvising. He pointed directly at the nickel ban as the template — the same policy that took processed nickel exports from $3.3 billion to $33.9 billion in seven years. That's the number Jakarta sees when it looks at tin. Value capture downstream.

Meanwhile, the demand signal that nobody in the mainstream press is connecting to tin continues to intensify. There are 190 gigawatts of hyperscale data center capacity announced as of early 2026. Every rack, every board, every chip-to-substrate connection in those facilities requires tin solder. As advanced packaging moves to finer pitch and taller chip stacks, the tin content per unit goes up. The AI buildout is eating tin from the bottom of the supply chain, and the analysts covering AI don't cover tin, and the analysts covering tin don't cover AI. The gap between those two desks is the whole trade.

Here's what makes tin particularly nasty in a squeeze: it's a tiny market. Global refined production is roughly 350,000–400,000 tonnes a year. Copper does that in a few days. A relatively small amount of speculative money can move tin prices violently — and when the physical tightness underneath is real, the speculative flows and the fundamental flows start feeding off each other. January was a preview. Prices ripped 25% in a month. CRU called it "disconnected from fundamentals." I'd argue the fundamentals just hadn't been properly diagnosed yet.

I've watched small-market metals do this before. The pattern is always the same: quiet accumulation of physical tightness, dismissive coverage, a sudden repricing event that leaves the sell-side scrambling to rewrite their models in a weekend. Tin is the smallest LME base metal by market size. It doesn't need a lot to move. It needs one more supply valve to close.

The trigger options are clearly visible. Jakarta sets a timeline for the export ban. M23 takes another run at Bisie. Man Maw's dewatering stalls. Or — most prosaic and most likely — a single large physical delivery contract against LME fails because there are 6,000 tonnes on warrant and someone just asked for 3,000 of them.

When a delivery contract fails on a market this thin, the sequence is mechanical. Backwardation explodes. Nearby contracts reprice. Physical premiums — already elevated — go vertical. Solder manufacturers, who buy tin on rolling contracts, eat the spike immediately. Electronics contract manufacturers pass it downstream. PCBA costs rise. It doesn't show up as a "tin crisis" on CNBC. It shows up as margin pressure in the next quarter's earnings calls from companies that solder boards — which is every hardware company on the planet.

The January spike already demonstrated this. Tin surged above $50,000 a tonne, SHFE hit 443,380 yuan. Solder cost analysts started publishing emergency client notes. Then it corrected, and everyone exhaled and forgot about it. The physical picture underneath didn't improve during the correction. It got worse. LME stocks have fallen another 33% since.

Where does capital not go? Diversified miners where tin is a rounding error. The broad base metals ETFs that dilute tin exposure into copper and aluminum noise. And anything with Indonesian supply chain dependency, because a ban — even a staged one — reprices the entire forward book overnight.

The ITA projects a 13,000-tonne annual supply deficit by 2030. That's the base case, without an Indonesian export ban. With it, the deficit doubles. The project pipeline to replace lost supply is nearly empty — Russia's Seligdar isn't bringing its Amur plant online until late 2028 at the earliest. There is no quick fix here either.

The financial press says niche. The physical market says five days of deliverable stock. I know which one I'd bet on. The machine doesn't care whether you're paying attention to it or not. It just keeps draining.