Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
48 Days. That's How Long Before the Helium Runs Out for AI Chips.
TL;DR
A helium plant in Qatar has become a real choke point for AI — Nate argues that Ross Leffan, which historically accounted for roughly 33% of global helium supply, being offline and partially damaged puts chipmaking in Taiwan and South Korea directly at risk.
There is no practical helium substitute for advanced chip fabrication — helium is required for EUV lithography support, plasma etching temperature control, and vacuum leak detection, and advanced fabs can burn through 5,000 to 20,000 cubic meters per month at 99.9999% purity.
The 35-to-48-day clock matters because helium literally evaporates in transit — stranded ISO containers headed to Taiwan and South Korea can lose their liquid helium cargo if delays persist, turning a shipping disruption into actual lost supply.
This is not just a chip-gas story; it is also an LNG and electricity-cost story — Qatar’s helium is tied to LNG production, so shutdowns also raise East Asian fab energy costs, which Nate says will flow through into higher chip prices and more expensive AI inference in 2027–2028.
China may come out structurally stronger if the disruption drags on — Nate points to China’s push on domestic helium, including a Guangdong plant accepted by ASML, and the potential Power of Siberia 2 pipeline as giving China a path to cheaper energy and a more independent chip stack.
His practical advice is simple: buy compute sooner, not later — with HBM already sold out, DRAM already up 70%, helium spot prices doubled, and surcharge contracts up 30%, he warns that laptops, phones, servers, and AI infrastructure are all likely to get pricier.
The Breakdown
A trillion-dollar AI build meets a very physical bottleneck
Nate opens with the big tension: hyperscalers are ready to spend another $1 trillion-plus on AI over the next 24 months, with Sergey Brin’s “I’d rather go bankrupt than lose the race on AI” setting the tone. But that spending spree runs straight into a boring, brutal reality: chips depend on helium, and a damaged plant in Qatar has exposed how fragile the physical substrate of AI really is.
Why helium is suddenly the star of the story
He walks through why helium is non-negotiable in fabs: it cools wafers during plasma etching, helps maintain uniform temperature, and is used to detect leaks in vacuum chambers because its particles are so small. He quotes Georgetown CSET researcher Jacob Feldgoise’s point that helium is “the only thermal conductor that makes it possible at that scale,” then adds the kicker: the more advanced the chip, the more helium it tends to consume.
The Ross Leffan shock: one plant, one strike, global consequences
Qatar produces about 2.4 billion standard cubic feet of helium per year, around a third of the world’s roughly 7 billion, and Ross Leffan has been the key source shipping east through the Strait of Hormuz. Nate says the plant has not just been paused but hit by missiles, with QatarEnergy reportedly admitting 14% of capacity is permanently damaged and some reconstruction could take up to five years — an eternity in AI time.
The 48-day countdown and who gets hit first
One vivid detail makes the whole thing feel immediate: helium in specialized ISO containers only lasts about 35 to 48 days before it starts to vaporize. That means stranded shipments to South Korea and Taiwan are effectively ticking clocks, and Nate connects that directly to SK hynix, Samsung, and TSMC — the companies behind HBM and logic chips that feed Nvidia GPUs, AMD accelerators, and Google TPUs.
It’s also an LNG story, which means an energy-cost story
Nate widens the frame: helium at Ross Leffan is inseparable from LNG because it is extracted as a byproduct of cryogenic gas processing. So the same disruption that tightens helium also raises East Asian energy costs, and for places like Taiwan — which imports 97% of its energy and holds only 11 days of gas reserves — that means more expensive fabs and eventually more expensive AI.
Europe and planners have a different problem: uncertainty
He says the issue is not that fabs instantly shut down; it’s that they may run slower, scale less, and pass through higher costs over time. Europe gets pulled in too because LNG prices rise globally, and Nate’s broader point is that planners are stuck because they cannot forecast the war timeline, the repair timeline, or how long complex industrial systems remain under missile threat.
The geopolitical twist: China could gain leverage
In the final third, Nate turns to Power of Siberia 2, the long-discussed Russia-to-China gas pipeline, arguing the crisis gives Beijing more reason to get the deal done. Pair that with China’s domestic helium push — including a Guangdong plant with 6N certification and ASML acceptance — and his warning is that China could end up with cheaper energy, better helium access, and a stronger native chip stack while South Korea and Taiwan remain exposed.
What this means for everyone from hyperscalers to laptop buyers
He lands on practical consequences: HBM was already sold out, DRAM was already up 70%, helium spot prices have doubled, and contract surcharges are up 30%. His advice is very direct and very Nate: if you need compute, buy it sooner this year, because whether you’re a hyperscaler, an IT procurement lead, or just buying a phone or laptop, he expects this to show up as delays and price bumps.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
