Insights from the US

AI is becoming cheaper and more capable, but its economic benefits will accrue mainly to companies with pricing power, proprietary data, customer relationships or exposure to constrained physical supply.

• Emma Fisher attended 40 sessions at Goldman Sachs’ Communacopia conference in San Francisco in September 2026, where the main themes were token-cost optimisation, open-source models and the spread of always-on AI agents.
• Pinterest said open-source models reduced its cost to less than 8% of a comparable frontier model, while SemiAnalysis reported that one-year rental contracts for Nvidia H100 GPUs rose almost 65% between October 2025 and August 2026; Epoch AI estimated that the cost of achieving GPT-4 performance on PhD-level science questions fell 40-fold per year.
• The AI build-out has reached roughly 2% of US GDP in annual capex by the largest US technology companies, with financing increasingly shifting from hyperscalers’ free cash flow to debt; default protection on Oracle and CoreWeave widened sharply.
• AI productivity will not automatically raise margins because companies may pass savings to customers, as banks did after adopting ATMs, core banking, internet and mobile technology; Magellan therefore favours businesses such as James Hardie, The Lottery Corporation and Aristocrat, which have distinctive pricing power.
• AI threatens existing moats based on knowledge and switching costs: Car Group is responding with its Nexgate dealer platform and proprietary data, while Magellan exited Xero because agents could automate bookkeeping and reduce switching costs.
• The physical economy remains a constraint on AI expansion: mines take around 15 years to develop, Australian ammonium-nitrate supply has been flat since 2017, and tightening supply should support EBIT per tonne for Orica and Dyno Nobel alongside commodity beneficiaries such as BHP, Rio, Santos and Ampol.

Takeaway: Investors should focus less on whether AI is a bubble than on which companies can retain its productivity gains, defend their customer relationships and benefit from scarce physical inputs, while monitoring disruption to software and digital intermediaries.