ATLASBrief← All briefs
Podcast · 2026-09-12

Odd Lots (Bloomberg): Robert Friedland on the World's Monumental Shortage of Copper

Robert Friedland argues the world cannot mine enough copper to sustain growth, let alone electrification — and that physical constraints are now binding on the AI buildout itself: gallium without which 'there is no NVIDIA,' 7-year gas-turbine waits, 8–10-year grinding-equipment waits, and hyperscalers telling him 'my God, we've got a problem now.'

EnergyData Centers & PowerGeopolitics & TradeAI Labs

Episode

Odd Lots (Bloomberg)
ShowOdd Lots (Bloomberg)
EpisodeRobert Friedland on the World's Monumental Shortage of Copper
GuestRobert Friedland
HostJoe Weisenthal, Tracy Alloway
Published2026-09-12T08:00:00Z
Duration70 min
Fidelity[transcript]
StanceBULLISH
ListenEpisode link ↗

Abstract

Robert Friedland, founder of copper miner Ivanhoe Mines, Ivanhoe Electric, and the iPulse startup, argues the industry cannot mine enough copper to sustain growth, let alone electrification. The conversation covers copper's record ~$14,875/ton print and the first production decline since 2017 (BNEF, per hosts), the 700Mt-in-18-years supply arithmetic, grade collapse at La Escondida, the sulfuric-acid price explosion and export bans, AI searches consuming ~30x Google-search energy, the weaponization of critical minerals, NPV-based valuation suppressing mining to under 1% of the S&P 500, and tariffs as a two-sided lever. No tradable vehicle is surfaced: this is a structural copper bull from a copper producer with an obvious long position — a positioning read, not an entry call.

The Theses

9 claims
1. Copper supply cannot meet baseline demand growth, let alone electrification, EVs, or AI data centers.
Evidence: Humanity has mined 700 million metric tons of copper since Mohenjo-daro and must mine that same amount in the next 18 years just to sustain 3% GDP growth for 8 billion people — before EVs, grid electrification, or data centers (Friedland).
2. Ore-grade collapse is turning the world's largest copper mines into capital sinks.
Evidence: La Escondida in Chile fell from ~2% grade when found to 0.8% today and heads to 0.4% within two years, with owners modeling $10–12B of investment just to make the mine 'go down in production' (Friedland).
3. Sulfuric acid is a hidden chokepoint for roughly a quarter of global copper output, and its price has exploded.
Evidence: Prices rose from $150 to over $1,000/ton in eight months; ~25% of copper recovery requires acid leaching; Russia and China have banned acid exports; Ivanhoe sells smelter-byproduct acid to fellow miners 'at higher and higher prices' (Friedland).
4. AI search makes copper a consumption story, not just a buildout story.
Evidence: A Google search uses the energy of a 100W bulb for ~12 seconds; an AI search uses 30x more — a 100W bulb for two to three minutes — and electrical demand for AI searches 'literally goes to infinity' (Friedland).
5. Critical raw materials are being weaponized as the world fragments into competing supply-chain blocs.
Evidence: Friedland has spent recent years with both the Biden team and the Trump administration as 'the weaponization of certain critical raw materials' becomes a national security interest; the acid-export bans show it already happening (Friedland).
6. Critical-materials dependency is now a hard constraint on the AI buildout's speed, not just its cost.
Evidence: US manufacturers must submit plans and blueprints to China's Military Commission to obtain critical metals, refusable 'for any reason or no reason'; 'Without gallium, there is no NVIDIA, no NVIDIA chips... We can't develop AI as fast as the Chinese' (Friedland).
7. The physical supply chain for the buildout is stressed at the equipment layer.
Evidence: Hyperscalers are approaching Friedland directly: 'my God, we've got a problem now'; ~7-year waits for gas turbines with force-majeure clauses; grinding-motor ring gears went from a 4.5-year wait to 8–10 years 'if ever'; at Ivanhoe's flooded Congo mine (~$15M/day lost) only Chinese builders could make the samarium-cobalt-magnet pumps, built in 30 days (Friedland).
8. NPV-based valuation structurally suppressed mining investment; the sector is now under 1% of the S&P 500, an all-time low.
Evidence: 'Mines cannot be modeled to NPV... the NPV model suppressed the valuation of mining companies'; China, seeing this, 'intelligently just bought the entire supply chain... bought everything that wasn't bolted down' (Friedland).
9. Tariffs are a two-sided lever for US copper self-sufficiency.
Evidence: A 30% tariff on $6 copper → $7.80 makes US mining viable against $100+/hour United Steelworkers labor and environmental/legal costs; the other side: domestic inflation, higher rates, midterm politics (Friedland).

Key Math

  • ~$14,875/ton copper, ~3% lower at recording — Record print: new all-time high earlier in the week; volatility already biting producers' investment decisions (hosts)
  • 700M metric tons ever mined vs 700Mt needed in the next 18 years — Supply arithmetic: the load-bearing demand claim, stated before electrification, EVs, or data centers (Friedland, interested producer)
  • La Escondida: 2% → 0.8% → 0.4% grade; $10–12B to shrink output — Grade collapse: the world's largest copper mine needs billions just to decline (Friedland)
  • $150 → >$1,000/ton sulfuric acid in eight months; ~25% of copper recovery acid-leached — Acid chokepoint: an unpriced input constraint compounded by Russia/China export bans (Friedland)
  • 30x: AI search vs Google search energy (100W bulb for 2–3 min vs 12 s) — AI energy multiplier: the consumption-side demand argument (Friedland)
  • ~7-year gas-turbine wait; ring-gear wait 4.5 yrs → 8–10 yrs 'if ever' — Equipment backlog: the buildout is constrained one level up the chain too (Friedland)
  • $15M/day lost at flooded Congo mine; only China could build the pumps, in 30 days — Single-source dependency: samarium-cobalt magnets as a worked example of the chokepoint (Friedland)
  • Miners <1% of S&P 500, all-time low — Sector underweight: the NPV-valuation suppression mechanism's endpoint (Friedland)
  • Ships: US ~8 vs China ~1,000 built last year — Sovereignty gap: the reindustrialization argument's starkest number (Friedland)
  • $1.5T proposed defense budget (+50%); Europe to 3% of GDP — Military copper intensity: 'all of these modern systems are copper-intensive' — governments buying copper as 'a proxy on money' (Friedland)

Variant Perception

Priced in: copper-as-beneficiary of electrification and data-center demand is already consensus, and the hosts themselves open with the tension that price volatility deters producer investment. What's genuinely new: (a) the Escondida grade-economics worked example — billions spent to produce less — as the concrete mechanism of the supply shortfall; (b) sulfuric acid as a second-order chokepoint, up ~6x in eight months with export bans, rarely present in the copper-bull conversation; (c) the consumption-side energy framing — AI search's 30x multiplier and the '20 critical metals in a server farm' list — material constraints the GPU narrative rarely prices; (d) 'Without gallium, there is no NVIDIA' — a direct line from critical-minerals policy to merchant-silicon supply; (e) China AI '99% as good as ours' and going open source — 'give it to American Enterprise for free' — while hyperscalers bet on 'astronomically large capital investments,' a demand-side bear note on the capex supercycle; (f) the hosts' own close: constraints exist 'at every step of the chain... one level up, higher up river, they have their own backlog,' and it is 'still not clear that the market will actually reward you for increasing production' given volatility — the natural bear. The discount is self-interest: a copper founder talking his book whose 700Mt arithmetic assumes no price-induced substitution or recycling response.

Positioning Read

Directional only
AI capex supercycle: strengthens on physical-constraint evidence (equipment backlogs, hyperscaler outreach), tempered by the China open-source 'free' framing as a demand-side bear note — mixed-positive.
Power & interconnect is the binding constraint: strengthens — equipment lead times, turbine waits, and uninterruptible-power physics (scandium fuel cells, rhenium turbines) layer onto the bottleneck story.
Energy demand: strengthens — the AI-search 30x multiplier, military-buildup copper intensity, and voter backlash on electricity prices reinforce the demand leg.
Policy / sovereign AI: strengthens — Bessent–China exchange, Xi's September 24 US visit, and 'the age of hoarding' as governments buy copper as 'a proxy on money.'
Buildout financing / inference economics / agent platforms / seat-based software / custom silicon / regulatory overhang: neutral — not addressed.

Frameworks

Just-in-time → just-in-case economy
'Our beautiful, integrated, just-in-time world economy breaks down into a just in case economy' — every country now securing its own supply chain, driving incipient supply-chain inflation.
The age of hoarding
Governments are buying copper metal 'as a proxy on money' — unreported — so 'nobody can calculate the supply-demand balance on any of these metals.'
Banana principle (build absolutely nothing anywhere near anybody)
Resolution Copper in Arizona has sought a permit for 35 years; the US cannot build mines, nuclear plants, or ships under current legal constraints.

Actionability

No tradable vehicle is surfaced — this is a structural copper bull from a copper producer with an obvious long position: a positioning read, not an entry call.

Fidelity Notes

Full episode transcript verified against the source: 1,178 captions through the 01:09:39 outro. This episode is part one of a multi-part conversation; Friedland returns for parts two and three.