Why this Soros alum is betting on copper and cables over AI chips
For much of the market, the AI revolution is a story told in silicon: eye‑popping capex from hyperscalers, blistering demand for advanced GPUs, and soaring semiconductor valuations. A Soros Fund Management alumnus looking through the classic macro lens sees something else: the real choke point is no longer compute, it’s power and plumbing. That is pushing capital away from chips and toward the metals and manufacturers that make electricity and data actually move—copper, high‑voltage cables, transformers, and the companies that install them.
The Soros playbook in one line: find the bottleneck, not the billboard. Right now, the bottleneck is the grid.
What’s changed: from compute scarcity to power scarcity
– AI demand is colliding with physics. Training clusters and inference at scale require immense, continuous power. Even the best chips are idle without megawatts behind them and fat pipes to move electrons and bits.
– Permitting and build times for transmission and substations run years, not quarters. You can spin up more HBM capacity or packaging lines in 12–24 months; you can’t conjure a 500 kV line or a 1,000 MVA transformer on that timeline.
– Evidence has piled up across markets: grid interconnection queues have swelled, data‑center projects have been gated by substation availability, and utilities warn about long lead times for high‑voltage equipment. In short, the constraint has migrated from compute to capacity.
Why copper sits at the center
– Ubiquity and irreplaceability. Copper is the conductor of choice from the power plant to the rack. Substitutes like aluminum work in some use cases, but where space, losses, and reliability matter, copper dominates.
– Inelastic supply. New copper supply is hard. Ore grades have trended lower for decades, major greenfield projects face permitting, water, and social hurdles, and timelines from discovery to production often exceed 7–10 years. Recent disruptions underscored geopolitical and operational risk.
– Stacked demand drivers. Electrification (EVs, renewables, heat pumps), re‑shoring of manufacturing, and now AI‑driven data‑center build‑outs layer on top of baseline demand. Even modest misses on the supply side can create multi‑year tightness.
– Favorable industry math. When markets tighten, copper’s price move tends to be nonlinear relative to small inventory changes, offering the kind of asymmetry macro traders prefer.
Cables, not wafers: an overlooked oligopoly with time on its side
– Scarcity where it matters. High‑voltage submarine and HVDC cables are complex to design, permit, test, and install. Capacity is constrained not just in factories but in a handful of specialized cable‑laying vessels and experienced crews. Order books for large projects have historically stretched multiple years.
– Built‑in pricing power. Many cable contracts include raw‑material pass‑throughs, reducing margin volatility, while scarcity supports better spreads on capacity and installation. Backlogs offer multi‑year earnings visibility—rare in industrials.
– Structural demand. Offshore wind export cables, interconnectors linking regions and countries, grid reinforcement for electrification, and data‑center clusters needing high‑capacity feeds all drive the same outcome: more copper‑rich, high‑voltage cable.
– Barriers to entry. Certification, reliability records, warranty risk, and the capital intensity of vessels and plants keep the field concentrated among a few global players, with regional champions on land cables and accessories.
Why prefer this over AI chips now
– Crowded trade and valuation risk. Leading AI chip names discount continued hypergrowth and dominant share. Even if the AI thesis is right, the market may be paying forward multiple years of perfection.
– Faster competitive cycles. Semiconductors face rapid product obsolescence, potential margin pressure as supply catches up, and policy risk around export controls. The “how much and how fast” of AI ROI also matters for sustainability of spend.
– Slower, stickier cycles in the grid. Transmission and substation projects, once funded and permitted, tend to complete. Backlogs and multi‑year execution dampen earnings volatility compared with the boom‑bust of some chip segments.
– Reflexivity at work. Every incremental gigawatt demanded by AI hardens the case for upgrades, which tightens the very markets—copper, cables, transformers—that feed that expansion.
Ways to express the view
This is not investment advice, but the thesis can be implemented along a spectrum of risk and cyclicality:
– Copper exposure
– Miners with high copper leverage and sound balance sheets.
– Royalty/streaming companies with copper exposure for lower operating risk.
– Futures or structured exposure for purer beta, with appropriate risk controls.
– Grid and electrification equipment
– High‑voltage cable manufacturers with subsea and HVDC capability.
– Transformer, switchgear, and grid‑automation suppliers.
– Transmission EPCs and specialized contractors that build lines and substations.
– Complementary angles
– Utilities with regulated transmission growth plans and constructive rate frameworks.
– Data‑center infrastructure REITs and developers with secured power and interconnects, where available at reasonable valuations.
– Relative value
– A barbell or pairs approach: long grid/copper beneficiaries against an underweight or hedge in the most crowded AI chip exposures, to isolate the bottleneck spread rather than make a binary call on AI.
What could go wrong
– Demand disappoints. If AI workloads prove less power‑intensive than expected due to efficiency gains, or if enterprise adoption lags, the near‑term pressure on grids could ease.
– Policy and permitting setbacks. Transmission buildouts can be delayed by local opposition or judicial challenges; reforms can stall.
– Supply surprises. A faster‑than‑expected recovery in copper supply via debottlenecking, restarts, or substitutions in some applications could cap prices. In cables, aggressive capacity additions or new entrants could compress margins over time.
– Macro shocks. A broad slowdown can defer capital projects, even if long‑term need remains intact.
Signals to watch
– Copper inventories and spreads across LME/COMEX/SHFE; treatment and refining charges (tight TC/RCs often signal concentrate scarcity).
– High‑voltage cable order announcements, lead times, and new vessel deliveries.
– Transformer backlog data and reported lead times by utilities and OEMs.
– Transmission permitting and policy reforms; interconnection queue trends.
– Hyperscaler disclosures on data‑center power procurement and on‑site generation plans.
The bottom line
A Soros‑style macro lens prioritizes bottlenecks, reflexivity, and asymmetry. Today, the AI narrative still grabs the headlines, but the returns may increasingly accrue to the quiet enablers: the copper pulled through conduits, the cables laid beneath seas and across plains, and the grid equipment that turns electrons into usable compute. Betting on the plumbing over the poster child is less glamorous than owning the hottest chip, but in this phase of the cycle, it may be where the most durable alpha hides.
