Tearing apart the tin solder squeeze

Tin.

The headline this week is relief. Myanmar's Man Maw mine — the mega-deposit in autonomous Wa State that shut in August 2023 and took roughly 10% of global tin concentrate with it — is restarting. Reuters ran it. The wires repeated it. The consensus read is straightforward: supply is coming back, the squeeze is loosening, prices should cool from the $50,000 range where they've been grinding all year.

That's the dashboard. The engine is telling a different story.

The International Crisis Group — the people actually on the ground — called the Man Maw restart a "slow restart" and said output remains "well below pre-suspension levels." China's H1 2026 imports of Myanmar tin concentrate have surpassed all of 2025, but that's not the victory it sounds like. All of 2025 totaled just under 40,000 tonnes — sharply depressed from pre-shutdown norms. The bar they cleared was low.

Meanwhile, the demand side has quietly undergone a structural shift that most of the sell-side hasn't updated their models for. Tin is not a battery metal or a wire. It is solder. Every chip that ships — every AI accelerator, every HBM stack, every data center server board — is held together by tin solder joints. And the AI buildout has added an entirely new demand layer on top of a market that was already tight. CITIC Securities estimates AI data servers alone will generate 2,500 tonnes of new tin consumption in 2026. Fastmarkets models gross AI-chain tin procurement exposure at 11,340 tonnes this year, rising to 26,510 by 2030. That's not a rounding error. That's a second Indonesia.

I watched traders buy the "restart" headline and sell tin three times since Man Maw went dark. They were right twice. The third time, the stocks didn't refill.

Story off. Numbers on.

LME three-month tin settled at $54,633 on September 3. It touched $56,800 in January and spiked to an intraday all-time high near $59,000 on June 2. The metal hasn't revisited sub-$40,000 since late 2025. The market structure has shifted: this isn't a spike. It's a new floor.

Refined supply is growing roughly 3% in 2026. Demand is growing 3.5%. That half-point gap doesn't sound like much until you remember this is a 390,000-tonne market. Coface calls it the first supply deficit since 2021. The ITA puts it more bluntly: the pipeline of new tin projects is too thin and investment has been insufficient for years. They forecast a 13,000-tonne deficit by 2030 without new mine investment.

On the supply side, the geometry is ugly. PT Timah — Indonesia's state-owned producer and historically the world's swing supplier — is targeting 30,000 tonnes of refined output in 2026. They produced 5,630 tonnes in Q1, up 82% year-over-year. Sounds good until you learn that President Prabowo ordered 1,000 illegal tin mines shut down in Bangka Belitung. The Indonesian Navy seized 514 tonnes of illegally mined minerals in 2026 alone. The informal sector that used to pad Indonesia's numbers is being dismantled.

And SHFE? Trading volumes in January exceeded one million tonnes in a single session — more than twice global annual physical usage — before Shanghai authorities intervened. That's not price discovery. That's a casino bolted onto a commodity that moves in thousands of tonnes, not millions.

The restart headline gave the paper market a reason to sell. The warehouse data gives the physical market a reason to panic. Those two stories can't coexist for long.

Here's what nobody is modeling: the demand multiplier hiding inside every new chip generation.

Advanced semiconductor packaging — CoWoS, HBM stacking, 3D chiplets — doesn't just use tin solder. It uses exponentially more of it per unit of compute. An HBM4 stack bonds up to 16 DRAM dies with through-silicon vias. Each interposer carrying those stacks requires thousands of micro-bump solder joints. TSMC's CoWoS capacity is scaling toward roughly 130,000 wafers per month by year-end, feeding into 190 GW of announced hyperscale AI data center buildout — and every wafer routes through tin.

The analysts pricing tin still treat it as a legacy electronics commodity — tin tracks PC shipments and smartphone cycles. That was true in 2018. It is not true now. Nvidia, AMD, and Broadcom are shipping AI accelerators at volumes that didn't exist three years ago, and every one of those chips consumes more solder per die than the generation it replaced. The demand curve changed slope and the models didn't notice.

And then there's the supply-side weak link nobody mentions in polite company. The ITA projects tin mine production at roughly 360,000 tonnes for 2026. China — still the world's largest producer — has seen output edge higher, but new project development lags demand growth. Peru and Bolivia have reserves but not capital. The DRC contributes meaningful concentrate but from artisanal operations that don't scale on command. The entire global supply response to a 50% price increase over two years has been… 3% more refined output. That's not a supply response. That's a shrug.

I've seen this exact setup before — a restart headline giving the paper market permission to sell while the physical inventory silently drains. I was on the wrong side of a nickel trade with the same structure in early 2022. The headline said "supply returning." The warehouse said otherwise. The warehouse was right. The warehouse is always right eventually.

Man Maw's "slow restart" might add 15,000–20,000 tonnes of contained tin this year. The AI demand layer alone eats most of that before it reaches the exchange. The machines are reading a headline. The loading docks are counting pallets.

The sequence is mechanical. If you've watched commodity squeezes, you already know the choreography. The only variables are trigger and timing.

First wave: the Man Maw restart numbers come in soft for Q3 and Q4. The ICG already flagged this — output "well below pre-suspension levels." The Wa State Army isn't running a commercial mining operation. It's running a revenue extraction operation inside a civil war. Dewatering deep shafts, rebuilding concentrator capacity, and reestablishing logistics through contested territory takes years, not quarters.

Second wave: Indonesia's formal sector can't absorb the loss of the informal tonnage fast enough. PT Timah is targeting 30,000 tonnes but managed only 5,630 in Q1. The Navy is actively seizing illegal shipments. A thousand mines are shuttered. The crackdown is real, and the gap between formal output targets and the informal supply they're destroying is measured in tens of thousands of tonnes per year.

Third wave: physical buyers — solder paste manufacturers, PCB assemblers, the entire electronics assembly chain — realize the LME can't deliver. Premiums detach from the exchange price. This is where the cost transmits into the real economy. Every server board. Every EV inverter. Every 5G base station. Tin solder is in all of them, and there is no substitute at scale.

Where capital doesn't go: broad commodity ETFs where tin is a single-digit allocation buried under copper and aluminum. The leverage is in concentration. Companies whose revenue is directly tied to tin spot, with unhedged physical inventory, in jurisdictions where the mines are actually running.

The risk? Man Maw ramps faster than ICG expects. Indonesia loosens the crackdown under industry pressure. AI capex decelerates in H2. Any of those would compress the deficit and give the paper-market bears a reason to reload. I don't discount it. I've been burned by supply narratives that resolved faster than the geological timeline suggested. But the LME warehouse doesn't lie, and right now it's saying the same thing the loading docks said in zinc, in nickel, and in cobalt before the deficit became consensus.

Physical layer says tight. Paper layer says relief. In my experience, when those two disagree, bet on the metal you can hold in your hand. It just takes longer than your P&L wants it to.