Invest1 distinct publisher3 min readUpdated
China's export licence list, not fab capacity or megawatts, is what set the 2026 tungsten price. The arithmetic behind the gas that deposits the metal is smaller than most capex plans assume.
The Investor · Invest desk

Compiled by The InvestorSomething wrong?How this is made
Start with the capacity arithmetic the report leaves implicit. If Kanto Denka and Central Glass together made roughly 2,200 tons of tungsten hexafluoride a year, and that was about a quarter of world capacity [5], then the entire global business for the gas that deposits tungsten inside a chip runs on the order of 8,800 tons a year [1]. A quarter of that went offline on a single date [4]. There is no version of a data centre buildout where the input at the narrowest point is measured in low thousands of tons and nobody notices.
The demand curve does not explain the price. Global tungsten demand is projected to go from about 143,000 tons in 2025 to 210,000 tons in 2035 [7], which is roughly 47% over ten years, or about 3.9% a year compounded [2]. A 622% move [1] means prices multiplied by something like 7.2 [3]. Single-digit annual tonnage growth does not do that. What moved was access: China holds about 80% of supply [2], began restricting which firms could ship abroad in 2025 [14], and by 2026-2027 the authorised exporter list stood at 15 companies [3]. Defence demand tied to the Ukraine and Iran conflicts is pulling from the same pool [15].
The instructive detail is where the failure actually landed. Not a Chinese mine and not a fab, but two converters in Japan, which stopped because they had run out of metal to process [4]. That is the step buyers tend to model as a service rather than a supply chain. Samsung and SK Hynix have tungsten embedded in their process flows with no quick substitute, and are reported to be under pressure to diversify into alternatives that are limited and slow [6].
Which is where the equity story deserves a cold reading. Almonty Industries brought its Sangdong mine in South Korea to full production in July 2026 [9], and the same report notes that one mine does not close a global deficit, with Sangdong's output small against what the restrictions removed and non-Chinese project lead times measured in years [11]. On 17 August 2026 the company announced a share buyback, citing its own view that the stock was undervalued given the supply environment [10]. A producer buying its shares during a declared structural shortage is making a statement about its share price, not about tonnage.
Provenance matters here too. This is one account, carried by Crypto Briefing via Mining News North [12], and its framing that the binding constraint is neither chips nor electricity [13] is the argument of the mining trade press. The part that stands on its own is narrower and harder: deficits are expected through at least 2028 [8], and the practical question for anyone signing wafer contracts is whether their gas supplier can get metal at all.
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
Tungsten prices surged as much as 622% in 2026, driven by Chinese export restrictions, rising military demand and AI chip fabrication demand.
By 2026-2027, only 15 companies were authorised by China to export tungsten.
Japan's Kanto Denka and Central Glass, two of the world's leading producers of tungsten hexafluoride (WF6), ceased operations entirely on 1 July 2026, stating they had run out of tungsten to process.
Almonty Industries brought its Sangdong mine in South Korea to full production in July 2026; it is one of the few non-Chinese tungsten mines of meaningful scale.
The report frames tungsten, rather than chips, capital or electricity, as the bottleneck on global AI ambitions.
Starting in 2025, China imposed export restrictions limiting the number of firms permitted to ship tungsten abroad.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One syndicated trade retelling, no primary documents
Every figure traces to a single item on cryptobriefing.com credited to miningnewsnorth.com. There is no price benchmark, regulatory notice, company filing or named forecaster behind the headline numbers, and the pivotal chipmaker exposure claim is hedged as 'reportedly'. The specific dated events are internally consistent and the 8,800-ton implication is sound arithmetic on the source's own figures, which keeps this above the floor, but nothing here is independently corroborated.
Named, dated physical events; magnitude unconfirmed
This is not a technology-uptake story but a supply-chain one, and the source does supply discrete dated events rather than sentiment: two producers halting on 1 July 2026, a licence list narrowed to 15 exporters, a mine reaching full production in July 2026 and a buyback on 17 August 2026. What is absent is any measured downstream consequence - no fab output loss, no inventory drawdown, no procurement disclosure from Samsung, SK Hynix or any fab operator - so real-world bite on AI capacity remains unquantified.
Bottleneck framing outruns the demonstrated impact
The lede declares tungsten the binding constraint on world AI ambitions ahead of chips, capital and electricity, and leads with a 622% peak price - yet the demand path it cites is only about 3.9% a year, no fab has been shown to slow, and the substitution, recycling and secondary-supply channels that normally absorb such shocks are never addressed. The underlying supply facts are real and the small-market arithmetic is genuinely underappreciated, which is why this is a substantial overstatement rather than an extreme one.
Scarcity story flatters a named issuer; aggregating publisher
The only company positioned as a solution is a listed tungsten pure-play that, on the source's own account, was buying back its shares three weeks before publication while arguing its stock was undervalued given the supply environment - a party with direct interest in a widely believed scarcity narrative. The chain is also commercially motivated at the publishing end: a crypto/finance outlet reprinting a mining trade item, with no disclosure of any position in the securities discussed. Nothing here shows the facts were shaped by those incentives, only that the alignment is visible on the page.
Low - single unverified publisher
Confidence is capped by having one publisher, one provenance chain and no primary documentation. The dated, specific, internally consistent events (producer shutdowns, mine ramp, buyback) and the checkable capacity arithmetic support a modest floor; the unattributed forecasts, unbenchmarked price move and visible issuer incentive prevent anything higher.
invest
A $27bn memory ETF with a quarter in Micron is not diversified exposure1 distinct publisher
build
China's accelerator swap makes Cambricon supply, not export policy, your ship-date risk1 distinct publisher
product
The cheapest part in the car is now the one that stops the line1 distinct publisher
build
Neither can build it alone: Korean memory and Taiwanese packaging now set the accelerator clock1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
cryptobriefing.com
1 article · August 23, 2026