The Hosking Partners strategy rose +3.5% net in the third quarter, taking year-to-date (YTD) net returns to +19.3%, outperforming the MSCI ACWI Index by 1.9% in the quarter and +6.3% YTD. In 2026, the strategy has now outperformed the benchmark over the trailing 1, 3, 5, and 10-year periods, in the face of ever narrower, AI-dominated indices, coupled with rapidly rising bond yields – the US 10-year ended September at its highest level since 2007. It is especially gratifying that our highly diversified, cyclically sensitive, small-to-mid-cap biased strategy outperformed against these headwinds, reflecting alpha generation in long-standing, uncorrelated, idiosyncratic investment ideas.
Market review
Debt issuance by the AI hyperscalers is now, as acknowledged by the US Federal Reserve, so large that it is pushing up government debt costs, the so-called ‘crowding out’ effect but in reverse. Meanwhile AI-infrastructure spend is sucking up labour and capital across the US economy. Last September, we published a Hosking Post called The AI Paradox, which highlighted a regime change of hyperscalers turning from market-dominant, capital-light butterflies into capital-intensive caterpillars. One year later, we would make the following observations on this cycle, which now defies all hyperbole.
First, the level of capital expenditure has gone on to beat most prior expectations. Just one year ago, total capital deployed for AI data centre capex was estimated by Jensen Huang to be $3-4 trillion by 2030. That number has doubled to an eye-popping $7 trillion, according to estimates from Goldman, McKinsey and JP Morgan. In % of GDP terms, we have not seen anything like this since the late 19th Century US railway boom.
Second, economy-sized levels of debt are being issued to fund this capital deployment. AI-related debt issuance has risen from c1% in 2024 to over 20% of all US investment grade issuance. In 2026 alone, over $500 billion of investment grade debt has been issued to fund this build, a number expected to rise by several hundred billion in 2027. JP Morgan estimates that an additional $4.1 trillion of AI debt will have to be raised through 2030.
Third, credit investors will now determine the future evolution of the industry, and financiers are having to find ever more innovative ways to sell on this debt. The proliferation of special purpose vehicles (SPVs) and off-balance sheet leases and commitments marks a new chapter of complexity and opacity. It may or may not be co-incidental that many of the underlying indicators of AI ecosystem health are weakening – for instance, falling prices of tokens and weakness of frontier models vs ‘good-enough’ performance of open-weight competitors.
Where does this leave us? 12 months ago, we applied our simple, capital cycle framework to pose an AI conundrum:
“A major question remains as to whether this is a winner-take-all (or most) industry, or more of a commodity infrastructure. If the LLM market remains competitive and relatively commoditised, with switching between providers relatively simple, it would be logical to assume that returns fall toward marginal costs.”
What is now becoming clear is that the data centre is a factory. Compute appears to be a commodity. In industries which are competitive and capital heavy, four decades of experience teaches us that a powerful load-the-plant imperative will drive behaviours. To cover high fixed costs, pricing – in this instance, tokens of compute – will fall toward pressuring returns for all players.
A provocative analogy that would have major ramifications for global equities would be that AI-data centres go the way of the solar industry – a rapidly adopted and revolutionary technology that is commoditised and dominated by China. With a supposed $3.5 trillion combined IPO valuation, this conundrum will be tested by the progress of OpenAI and Anthropic towards their flotations in the upcoming months.
Performance review
Within the strategy, the top 10 stock contributors were not US, mega-cap, AI-theme beneficiaries. The top contributors came from nations as diverse as Norway, Greece, South Africa and the UK, with sectors spanning energy, financials, materials and consumer discretionary, and a median market capitalisation of $4.7bn. Many of these holdings have been in the strategy for several years, earning multi-bagger status, and challenging the standard interpretation of Hendrik Bessembinder’s study[1] that one should focus on predicting and holding only the small handful of ‘winning’ stocks.
The continued disruption to energy flows in the Middle East provided a tailwind for two sub-baskets within our 12% energy exposure: oil tankers and oil refiners. The oil tanker basket contributed c80bps (returning +50%), with strong performance from Hafnia, DHT and International Seaways. Our energy shipping basket has returned 440% since we first wrote about it in a 2017 Hosking Post. The initial thesis was that of a classic, contrarian capital cycle. The sector was perceived as cyclical and low quality, but the withdrawal of capital had created a supply-constrained industry which has played through with rising day rates in this key chokepoint for global energy distribution. This forgotten sector has become much more relevant post the Russian invasion of Ukraine and the continued geopolitical disruption since, a factor that is prompting new equity issuance and a rising order book. As a result, we have reduced our exposure, albeit slowly.
Our basket of oil refiners added c70bps (returning 57%) over the quarter with strong returns from Motor Oil and Marathon Petroleum. Whilst the benefit of energy disruption aligns with our oil tanker companies, we see longer term upside as the refining cycle is still immature. The ability to bring on new refining supply, particularly in western nations with restrictive planning and environmental policies, is limited. Yet demand for product remains high. Indeed, as per the Energy Information Administration (EIA), US demand for gasoline continues to rise to all-time highs, despite continual predictions to the contrary.
A number of idiosyncratic holdings within the UK were notable contributors in the quarter. Chief among these was Saga, whose 25% return on invested capital reported in H1 2026 results (on an annualised basis) demonstrates the increasingly capital-light, high return nature of the business. This 25% ROIC is currently comparable to US tech darling businesses Meta, Microsoft and Amazon. Looking forward, these US mega cap tech stocks are becoming more capital intensive. With Saga’s cruise ships effectively fully booked, the company is stretching the 75-year-old Saga brand into financial services, gathering an astonishing £2.1 billion in deposits in nine months through a JV partnership with the bank NatWest – an insight into the potential for further capital-light growth. Despite an 11x increase in the share price from the 2022 lows, we see further upside, as we outlined in a Hosking Post earlier this spring. Elsewhere in the UK, ASOS management is delivering on a turnaround plan that has been many years in the making, and we are cautiously optimistic of its Saga-like upside.
Japan was the top performing region, adding c150bps (returning +11%) in the quarter, driven by solid performance across our strategy of c80 Japanese stocks, with top contributors spanning sectors as diverse as financial services (Aichi Financial Group) to communication services (Recruit Holdings) to industrials (Komori Corporation) to materials (Toyo Seikan Group Holdings).
Our underweight in US mega-cap tech was a relative headwind with five stocks – Microsoft, Nvidia, Apple, Meta and Palantir – adding c$3 trillion of equity market cap over the quarter and detracting c260bps of relative performance.
Investment activity
We continued to sell down our memory semiconductor positions in the quarter. As discussed in our recent Hosking Post, we are supportive of the long-term industry evolution but cautious of super-normal profitability. We recognise the challenge of calling a precise top and continue to cautiously take profits on the back of share price rallies. In Japan, we continue to rotate out of our companies that have performed well and recycle capital into deep value, often smaller-cap companies that stand to benefit from an improving Japanese economy. The 44% rise in oil prices over the quarter allows us to trim areas within our 12% energy exposure, especially where prices discount long-term energy disruption – a situation that may or may not transpire, depending on unforecastable geopolitics. We used the correction in precious metals to evolve our gold and platinum group metal (PGM) baskets. We also added to several individual stock opportunities that have seen interest and valuation compression in the midst of the AI-frenzy and rising interest rates. For example, high-quality corporate office providers in New York (Vornado) and London (Great Portland Estates) appear well placed in this regard given restricted supply.
Conclusion
Reviewing the quarter, there are three over-arching components to our returns, all of which are likely to be amplified should the great AI-cycle rollover. The first is successful stock picking, something that is likely to be highly rewarded in any market broadening. Second, and harder to quantify, is the structural advantage afforded by Hosking Partners’ go-anywhere, global generalist approach. By not being siloed into a particular type of investing (e.g., ‘quality’ or ‘value’) or more prosaically a narrow sector or geography, we have freedom to compare and exploit opportunities anywhere. The ability to act on the contrast between, say, a lowly valued, quasi-monopolistic Greek oil refiner and an expensive, highly levered US enterprise infrastructure stock is rare and valuable. Third, the tolerance of a large number of shares increases our ability to invest in areas where there is no or little competing interest. This ability to take on uncorrelated, idiosyncratic ‘risk’ is at the core of many of the substantial contributors to our performance that have influenced returns this quarter and we believe will continue to do so over the coming years.
[1] Bessembinder found that a small minority of stocks account for the majority of net US stock market wealth creation over time. We believe the winners are difficult to predict and are as likely to emerge from overlooked corners of the market as from today’s consensus favourites.
