The compute supply. A handful of designers and foundries control the GPUs, accelerators and high-bandwidth memory the industry depends on — NVIDIA holds an estimated 70–95% of training silicon at ~75% gross margins, with TSMC and HBM makers forming the structural bottleneck. As workloads shift from training to inference, we add custom-ASIC exposure growing ~28% a year.

The compute operators. Hyperscalers — Azure, AWS, Google Cloud and Oracle — are funding the >$500B buildout and marking raw GPUs up as rented inference. We hold the operators turning today's capital expenditure into durable, contracted cloud revenue.

The compute monetisers. Foundation models from the frontier AI labs — OpenAI, Anthropic and Google DeepMind — plus the copilots and vertical software built on top of them, turning raw compute into recurring enterprise revenue. The youngest and fastest-moving layer — we size it for asymmetry and manage it actively.

The compute owners — before they list. We take direct pre-IPO and secondary-market positions in the frontier private AI companies, where the largest value is still being created off the public market. In 2026 these became the biggest private assets in the world — Anthropic raised near a $965B valuation and OpenAI filed to list above $850B — and we secure that late-stage compounding before it reaches an exchange.