AI Compute Supply Chain.
Where to invest in the AI compute stack: the five layers the market hasn't priced.
When NVIDIA reports earnings, the public market reads it through a small set of names: NVDA itself, TSMC, ASML, and the four US hyperscalers buying the chips. That set is correct — and incomplete. Every Blackwell GPU is the output of an eight-layer production pipeline that runs from silicon wafers up to the megawatts that power the data centres. The market has rerated three of those layers aggressively over the past year — optical interconnect names up 1000%+, SK Hynix up 900%, advanced chip test up 200-300%. The other five — substrates, fabrication equipment, foundry, compute silicon, power — carry the next leg. This project names the firm that owns the binding bottleneck at each layer, scores them on a four-dimensional criticality scale, anchors them to forward EV/Sales, and publishes the concentrated names we hold against the gap.
What we've actually found.
Three of the eight layers are already played out — forward leverage is execution, not multiple expansion.
Optical interconnect (Lumentum, Innolight up 1100%+), HBM at SK Hynix (up 900%), and advanced chip test (Advantest, Lasertec up 200-300%) have already done the rerate. From here, the question is delivery against the price the market has already paid.
The five mispriced layers carry the next leg.
Substrates, fabrication equipment, foundry, compute silicon, and power infrastructure are where the criticality-to-pricing gap still sits open. Each one has a named firm that owns the binding bottleneck, scored on the same four-dimensional scale as the layers above it.
NVDA is up only 13% YTD 2026 — the smallest gain of any 2026-active large-cap AI name in the cohort.
Catch-up capital is rotating to HBM second sources (Samsung +160% YTD, Micron +120%), power infrastructure (GE Vernova +71%, Vertiv +115%), and the optical second leg (Innolight +90% YTD). The headline name is not where 2026 leverage sits.
The eight-layer interactive scoreboard, the firms that own each binding bottleneck, four-dimensional criticality scoring, live forward EV/Sales across the cohort, the methodology, and the editorial guardrails all live inside the v0 founding paper.
Published pieces.
Each card is a standalone read — a substantive finding or a release. Source audits, propagation fixes, and engineering hygiene live in the Changelog below as they accumulate.
The five layers of AI compute the market still mispriced — and the names we hold.
Twelve firms at the eight-layer cohort's clearest residuals between bottleneck criticality and forward-multiple pricing. Each one is layer-tagged, ticker-anchored, with live YTD / 1-year pricing, what moved, what didn't, and the structural limitation on the position.
Substrates, fabrication equipment, advanced packaging, optical second-leg, and power infrastructure are the five layers carrying the next leg. The names here let a portfolio manager position against that map without having to rebuild the cohort or the criticality scoring.
Introducing the AI Compute Supply Chain bottleneck scoreboard.
The founding paper. The eight-layer interactive scoreboard, the firm that owns each binding bottleneck, four-dimensional criticality scoring per name, live forward EV/Sales across the cohort, and the methodology shown openly.
Most supply-chain coverage names one firm at a time or one process step at a time. This is the first public scoreboard that partitions the whole stack, scores every layer on the same criteria, and refreshes pricing live from public filings.
Methodology, propagation, audits.
Smaller fixes — source audits, page/code reconciliation, engineering — that aren't publishable on their own but are recorded here for traceability.
No changelog entries yet — the Supply Chain project has been a clean development arc of substantive additions. Source audits and pricing-reconciliation passes will land here as they accumulate.