Follow the Checkbook: Nvidia's $150 Billion Tells You Exactly Where the AI Bottlenecks Are

Jensen Huang published a short blog post in March 2026 that most investors read as a technology manifesto. We read it as a capital allocation map. His five-layer framework, energy, chips, infrastructure, models, and applications describe the full stack required to industrialize intelligence. But there is a layer he did not include, because Nvidia does not sell copper. We call it Layer 0. And it may be the most important investment in the stack.

The performance data is slowly making the case. Over the past six months, the layers closest to software are down double digits. The layers closest to physical scarcity are up 55–99%. Institutional capital is shifting from the top of the cake to the bottom, and most benchmarks have not yet moved.

In this issue of What Matters, we map all six layers, track every investment Nvidia has made across the stack, eleven deals totalling over $150 billion in the last eight months alone, analyse what those investments reveal about where the real bottlenecks are forming, and identify the specific assets and companies that connect the AI buildout directly back to the commodity supercycle, the monetary debasement story, and the privatisation of the frontier that underpin everything we write.

The trade everyone is making is the top of the cake. The trade that matters is the base.

Continue reading below.

 The 6-layers - investors have focused on L1, L2, and L3 (1-year change)