Tearing apart the tin solder supply chain

Story off. Numbers on.

LME tin stocks are sitting below 4,500 tonnes. Global refined tin consumption runs at roughly 380,000 to 400,000 tonnes per year — about 1,050 tonnes per trading day. Less than six days of cover is sitting in registered LME warehouses. The long-run average is three times that. When stocks drop below 5,000 tonnes, the cash-to-three-month spread historically goes into backwardation — meaning the market starts paying a premium to get metal now rather than later. We're already there.

The Indonesia plug is real but limited. PT Timah, the state miner, has increased output, but Indonesian tin grade has been declining for a decade — ore bodies are getting shallower and lower quality, and the smelters need to process more rock to get the same metal. Total Indonesian refined output peaked around 2013 and has been structurally lower since. The country cannot simply turn a dial and replace Myanmar's concentrate. The geology doesn't cooperate.

Now add the demand shift. Standard through-hole and surface mount soldering — the bread and butter of consumer electronics — uses tin-lead or SAF305 alloys at relatively thin deposits. Advanced chip packaging is different. Flip-chip BGA, 3D IC stacking, chiplet interconnects — these architectures use more solder per unit and require higher-purity alloys. TSMC's CoWoS packaging for Nvidia's H100 and B200 chips uses significantly more solder than a standard consumer GPU. Every AI server rack is, from a tin perspective, a completely different animal than the devices it replaces.

The two curves — supply declining on one side, demand composition shifting on the other — are moving toward each other. The LME price at around $31,000 per tonne today reflects the first curve partially. It does not yet reflect the second one at all.

Nobody in the semiconductor supply chain analysis is talking about the solder recycling rate.

Tin is recyclable. In legacy electronics — old motherboards, consumer devices, industrial kit — a meaningful percentage of solder tin gets recovered and recirculated. The recycling industry has historically provided a 30 to 35% supply buffer that smooths out primary mine shortfalls. Here's the problem: the AI hardware cycle is too new. The H100s and B200s going into data centers right now won't reach end-of-life recycling streams for five to seven years. The scrap that would normally offset a supply gap doesn't exist yet. It's sitting in racks in Virginia and Singapore running inference jobs. It will not be available as recycled tin until the 2030s.

There's a second weak link, and it's quieter. The Wa State Army in Myanmar is not a government. It is an armed ethnic organization that controls territory and runs commercial operations independently of Naypyidaw. The "environmental review" it announced in August 2023 was, by most accounts, a renegotiation with Chinese smelters over processing fees and revenue splits. I traded base metals through the 2015 Wa State export disruptions and the pattern was identical then: the mines stop, the language is environmental, the negotiation is financial. It resolved in 2015. It resolved again in 2016. But this time the duration has stretched past anything the previous suspensions hit — and the junta's hold on the country continues to deteriorate in ways that make a simple commercial renegotiation harder to close. The political risk that reopens these mines is substantially higher than it was eighteen months ago.

The CTA positioning data doesn't show any particular speculative interest in tin. It's not a crowded trade. It's not a fashionable trade. It's a small market — total LME tin futures open interest is a rounding error compared to copper or gold. Which is exactly why the physical squeeze, when it bites, tends to bite hard and fast. Small markets don't have the liquidity to absorb a covering event gracefully.

Tin squeezes don't telegraph themselves. That's the feature, not a bug. The market is small enough that by the time the headline writers notice the price, the first wave has already moved.

The trigger sequence doesn't require Myanmar to reopen or collapse. It just requires one large electronics contract manufacturer — a Foxconn, a Jabil, a Flex — to place an unexpectedly large spot solder order when their forward contract book runs low. That hits a market with less than five thousand tonnes of LME inventory. The spread dislocates. Others notice the dislocation and accelerate their own purchasing to avoid being caught short. That's not a supply crisis — that's a normal procurement panic in an illiquid market. It resolves in weeks. But during those weeks, the price does something ugly.

The longer-duration story is different. If Myanmar stays closed through 2027 — which is entirely plausible given the political trajectory of the junta conflict — the deficit compounds. Indonesia cannot fill the gap. Australian and Bolivian production is marginal. There are no greenfield tin projects in the permitting pipeline at scale. The supply side of this market has essentially no new capacity coming for three to five years. Meanwhile, every AI server farm that goes online adds incrementally to a demand base that the 2020-era supply models were not built to accommodate.

The semiconductor earnings calls this week talked about packaging innovation, CoWoS capacity expansion, and advanced node yield improvements. Not one mentioned what the packaging is actually made of. It's a small metal with an unglamorous name and a supply chain running through a conflict zone that nobody's editors want to explain. That combination — invisible to the narrative, visible in the inventory data — is usually where the interesting trades live.