The Themes That Matter: Where Structural Capital Belongs Now

The first half of 2026 settled an argument. Semiconductors decisively outperformed the mega-cap platforms: the Magnificent Seven basket was roughly flat while broad semiconductor indices gained 70 to 90 percent, and memory names did better still. Goldman Sachs framed the rotation precisely: the market is rewarding the companies that earn from AI capex and questioning the companies that spend it. Hyperscalers are committing hundreds of billions to data centre buildout; the economics of that spend accrue first to the suppliers of chips, memory, capital equipment and power infrastructure.

The uncomfortable implication for passive investors is drift. A Nasdaq tracker bought three years ago as a platform bet is now, in attribution terms, substantially a semiconductor and AI infrastructure bet. Our own H1 attribution work confirmed it: mega-cap platforms barely contributed while semi-cap and infrastructure drove index returns. Passive exposure changed character without a single transaction. That demands periodic look-through, and it demands restraint when adding thematic vehicles that quietly duplicate what the index already holds.

Digital AI is priced. Physical AI, intelligence embodied in machines that manipulate the world, is earlier in its curve and structurally harder, which is precisely why the duration is longer. The proof points are accumulating: robotic systems now perform the majority of relevant surgical procedures in major hospitals, warehouse robot fleets have crossed the million-unit mark at the largest operators, and industrial deployments of humanoids have moved from demos to multi-year supply contracts with German manufacturers. SoftBank's Son calls physical AI the source of the next trillion-dollar company. He talks his book, but the direction is right.

Discipline matters more here than anywhere else in the portfolio, because thematic labelling in robotics is notoriously dishonest. Funds stuffed with Nvidia and Tesla are AI funds wearing a robotics costume. We demand revenue purity: surgical robotics businesses whose revenues are effectively all robotics, Japanese factory automation leaders, and the actuator and precision-gearing chokepoints where sixty to seventy percent of a humanoid's manufacturing cost sits. Humanoids themselves remain a call option, not a core: China accounts for the overwhelming majority of installations at roughly half Western cost, batteries and reliability remain unsolved, and Western unit economics work only in narrow contexts. We size that optionality accordingly, via a single UCITS vehicle, and no larger.

Stanley Druckenmiller's copper thesis has aged into consensus, which does not make it wrong. The International Copper Study Group has flipped its 2026 balance from a forecast surplus of over 200,000 tonnes to a deficit of roughly 150,000 tonnes; UBS sees a deficit exceeding 400,000 tonnes as mine disruptions in Chile, Peru and Indonesia collide with demand. Copper traded through 12,000 dollars for the first time. And the long-run arithmetic is worse: S&P Global projects demand rising fifty percent to 42 million tonnes by 2040 while production peaks around 2030, leaving a potential ten-million-tonne shortfall even with recycling doubling.

The new demand vector is the one the mining industry never planned for. Data centres require four to six tonnes of copper per megawatt of installed capacity, and AI buildout is stacking on top of grid expansion, EVs and defence. Supply cannot answer quickly: a new copper mine takes a decade or more from discovery to production. This is the cleanest structural deficit in the commodity complex.

Uranium spent 2025 consolidating and then answered the doubters: spot broke through 100 dollars per pound in January 2026 before settling near 90, long-term contract prices reached the mid-80s, and enriched uranium hit 190 dollars per separative work unit against 56 three years ago. The supply side keeps under-delivering, with Kazatomprom, the largest producer, warning output is below plan and more Kazakh material flowing east to China and Russia rather than west.

What changed the demand curve permanently is the technology sector. Microsoft is restarting Three Mile Island, Meta has committed to a 1.2 gigawatt reactor campus with Oklo, and Amazon and Google are funding small modular reactor programmes measured in gigawatts. On reasonable deployment scenarios, technology-driven demand alone could absorb a low-teens percentage of global supply. These buyers are price-insensitive relative to regulated utilities; they need firm carbon-free power and can pay for it. Western policy has reversed from managed decline to strategic asset. That combination, inelastic new demand meeting structurally constrained supply, is the definition of the setups Druckenmiller hunts.

The numbers are no longer projections; they are budget law. NATO's Hague commitment targets 5 percent of GDP by 2035, 3.5 percent of it core defence. European allies raised spending nearly 20 percent in real terms in 2025 to 864 billion dollars regionally. Germany alone moves from 95 billion euros in 2025 to a planned 117 billion in 2026 and 162 billion by 2029, having rewritten its constitutional debt brake to do it. The EU's ReArm framework mobilises up to 800 billion euros, with a 150 billion euro joint procurement instrument now disbursing. This is a multi-year, treaty-anchored order book, the closest thing to contracted revenue growth that public equity markets offer.

Two disciplines govern our expression. First, Europe-pure over global: the marginal euro of spending is mandated to be spent in Europe, so diluting the theme with US primes gives away the point. Second, execution scepticism. The sector's 2026 test is conversion: procurement delays, labour shortages, fragmented national programmes and cancelled flagship contracts have already produced sharp single-name drawdowns. Order books are a proxy for growth; production, revenue and margin are the proof. Valuations now demand the proof.

Ray Dalio's framing is the right one: late-stage debt supercycles end with currencies absorbing the adjustment, and gold is the asset that sits outside every sovereign's promise. The official sector agrees with its actions. Central banks have accumulated roughly 1,000 tonnes annually for four consecutive years, double the prior decade's pace, and bought a net 244 tonnes in the first quarter of 2026 alone. Gold now represents a larger share of global central bank reserves than US Treasuries for the first time since 1996. The catalyst is remembered, not forecast: the 2022 freezing of Russian reserves taught every non-aligned central bank that dollar assets carry counterparty risk. That lesson does not expire.

Price has followed. Gold set a record quarterly average near 4,900 dollars in Q1 2026 after peaking above 5,400 in January, and the major houses' year-end targets sit between 5,400 and 6,000. We are not price-target investors in gold; we hold it as portfolio insurance against fiscal dominance, and we trim into euphoria rather than add to it, as we did earlier this year. Buffett's critique that gold produces nothing is correct and beside the point. Insurance produces nothing until the day it produces everything.

China refines roughly 90 percent of the world's rare earths and manufactures 94 percent of sintered permanent magnets, the components inside EV motors, wind turbines, robotics actuators and precision-guided munitions. Beijing has now demonstrated, repeatedly, that it will use that position: the 2025 licensing regime, the January 2026 catalogue expansion, dual-use bans aimed at Japan, and the June 2026 listing of the leading US producers on its export control list. Ex-China prices for controlled materials have spiked as much as sixfold. The West's response is real money, a ten-billion-dollar US strategic reserve programme, price floor mechanisms under discussion, and a 54-nation minerals framework, but rebuilding independent processing is a decade-long project at minimum.

The investment logic follows George Soros's reflexivity: every act of Chinese coercion strengthens Western policy support for the handful of credible ex-China producers, which raises their strategic value, which invites further coercion. The scarce asset is not ore in the ground; it is operating separation and processing capacity outside China with heavy rare earth capability. That list is very short, and it is where we are positioned.

Japan is the rare developed-market theme where the catalyst is institutional rather than cyclical. The Tokyo Stock Exchange's campaign against sub-book valuations has produced record cross-shareholding sales, record shareholder proposals, and buybacks on track for a record 20 trillion yen in the fiscal year to March 2026. Corporate cash-to-assets still sits near 21 percent against 8 percent in the US, which is the opportunity: the summer 2026 Corporate Governance Code revision explicitly targets that hoard, pushing the reform story from improvement into capital release. Foreign investors have noticed, with net buying in 2025 running at many multiples of the prior year and accelerating after the Takaichi government's pro-growth mandate.

Buffett's endorsement of the trading houses remains the cleanest single-name read on the theme: diversified cash-generative conglomerates, cheap on global comparables, run by managements that have learned to return capital, with embedded exposure to the commodity and energy security themes above. Our expression pairs a broad Japan fund core with optional overlay in the houses themselves. The currency is part of the position: unhedged yen exposure is an implicit bet on the normalisation that the reflation thesis requires.

Every portfolio needs a theme that does not correlate with the AI capex cycle. Healthcare is ours. The demographic demand curve is the most reliable forecast in investing: ageing populations in every developed market and most of China guarantee volume growth for decades. Layered on top is the largest new drug market in a generation. GLP-1 metabolic medicine carried Eli Lilly to the first trillion-dollar pharmaceutical market capitalisation, oral formulations launched this year are expanding the market rather than cannibalising injectables, and US Medicare has begun piloting coverage at consumer price points. Consensus forecasts put the class at a multiple of its current size by the early 2030s.

The sector spent 2024 and 2025 derated on policy fear and GLP-1 disruption anxiety, which is exactly the Marks setup: a quality sector priced for its risks rather than its durability. We hold it through a broad developed-market healthcare vehicle rather than picking the GLP-1 winner, because the honest answer on Lilly versus Novo versus the oral challengers is that we do not have an edge, and the index owns them all.

The most consequential companies of this cycle, in frontier AI and space, remain private. Investors want the exposure and the market has manufactured vehicles to sell it to them, several trading at premiums to net asset value of 100 to 300 percent. Paying three pounds for one pound of assets is not a technology thesis; it is a greater-fool thesis, and we treat such vehicles as structurally uninvestable regardless of how good the underlying portfolio is. The disciplined route is a listed investment trust with genuine private-market access trading at or below asset value. Premium-to-NAV discipline is not pedantry. It is the difference between owning the future and paying for it twice.