The silicon supply chain

Semiconductors are the substrate of the entire AI economy, and the capital cycle behind them is historic: leading-edge fabs now cost tens of billions each, and every major economy is subsidizing construction inside its borders. The chain that designs, builds, and equips those fabs is the deepest moat system in technology.

Historical context: chipmaking consolidated for decades until each critical layer — lithography, design software, advanced logic manufacturing — narrowed to a handful of firms or fewer. AI demand is now stress-testing that concentrated structure with the largest order wave it has ever seen.

The structural driver is capability plus geography: AI models demand leading-edge logic and memory in exploding volume, while governments treat domestic capacity as national security — layering subsidy-funded construction on top of commercial demand in a way no prior cycle had.

The technology frontier keeps adding investable layers: advanced packaging that stitches chiplets together, high-bandwidth memory stacked beside accelerators, backside power delivery, and the metrology and inspection tools each new node requires more of. Complexity is the equipment industry's revenue model.

Adoption economics are visible in capex plans: foundry, memory, and logic leaders publish spending that flows directly to equipment and materials suppliers with multi-quarter lag — one of the more forecastable revenue chains in technology for those who track it.

The beneficiaries span design-software vendors, equipment makers from deposition to test, materials suppliers (photoresists, gases, substrates), and the packaging and assembly tier — plus the specialty foundries serving analog, power, and radio-frequency chips every AI system also contains.

The value chain rewards different risk appetites at different layers: equipment oligopolies compound steadily, materials suppliers ride volume, and small-cap specialists — a niche metrology tool, a substrate innovator — carry the torque when their layer becomes the bottleneck.

The overlooked layer is exactly there: sub-scale equipment and materials names, test-and-burn-in specialists, and the power-semiconductor makers riding data-center electrical content. Many trade on trailing cyclicality while their AI-driven backlog builds.

Competitive dynamics are geopolitical as much as commercial: export controls partition markets, subsidy races relocate capacity, and every restriction reshuffles which suppliers can sell where. Policy literacy is now part of semiconductor analysis.

Risks: this is a deeply cyclical industry — inventory corrections are violent, capex plans get cut fast, and multiple compression arrives before the downturn is visible in earnings. Export-control changes can strand revenue overnight, and technology transitions can leapfrog incumbent toolmakers.

What to watch: equipment book-to-bill and backlog language, memory pricing, foundry utilization, subsidy disbursements converting to orders, and export-control revisions. The research follows the chain from capex announcement to component order — and the small caps positioned at the tightest links.