Summary
Dan Dreyfus, managing partner at Fortnite Capital, joins the All-In Podcast to make a pointed argument: the next binding constraint on the AI buildout isn’t memory bandwidth or chip supply — it’s copper. His case rests on the collision of several simultaneous capital cycles: data centers are now consuming an estimated $1 trillion per year in infrastructure spending, semiconductor fabs are on track to exceed $750 billion in investment, and grid modernization and defense spending are all competing for the same critical minerals at the same time.
The copper math is striking. Current global demand sits at roughly 30 million tons per year, and Dreyfus calculates that even at baseline GDP-growth demand — ignoring AI upside entirely — the world will need 700 million tons over the next 18 years. That figure equals everything mined in the previous 10,000 years combined. Meanwhile, fewer than a handful of tier-one copper mines are scheduled to come online before 2030, and new mines take 7 to 12 years to bring into production. Existing Chilean mines, some over a century old, are showing declining ore grades.
Dreyfus frames this as a classic supply-shock-meets-demand-shock commodity cycle — historically lasting 15 years with hundreds-of-percent upside — and argues it is still in its early innings. He also layers in a currency-debasement thesis, pointing to $40 trillion in U.S. federal debt growing at $2.5 trillion annually, as an additional tailwind for hard assets and commodities. For anyone tracking where AI infrastructure spending eventually runs into physical limits, this segment offers a concrete and specific perspective on the material supply chain beneath the compute stack.
📺 Source: All-In Podcast · Published June 10, 2026
🏷️ Format: Interview
