Lab · AI Handoff Board
The AI trade so far has been a construction story — chips, fabs, power, racks. The next phase begins when running AI agents becomes cheaper than the work they replace: the moment agentic AI moves from possible to economical. When that happens, value doesn't jump — it hands off, stage by stage, from the companies building the AI factory to the companies operating it to the companies living off its output. This board measures each handoff as a ratio of our own indices — nine ratios, one scoreboard, updated daily. Closes through 2026-07-17.
How to read it: every ratio starts at 1.00 on 2026-06-30 (the v2 reconstitution of all three indices). Above 1.00 — the numerator side has led since inception; the weekly change shows what moved last. Each card states which direction the agentic-economical thesis predicts — the board is the referee, not the advocate. Ratios read cleanly on multi-week horizons, not single sessions.
Rubin — the supply chain registers the shift first
Who checks the machines’ work?
Verification / Design thesis ↑
Testing & Metrology (S9) against EDA & Chip IP (S1) — the thesis-validation spread from the Kimi K3 report. If AI makes chip design abundant, verification becomes the scarce step.
Testing & Metrology (S9) ÷ EDA & Chip IP (S1)
Is chip design commoditizing?
Design / Physical thesis ↓
The Design layer against a Fabrication + Assembly & Test composite — incepted July 17, 2026. If design work commoditizes while atoms stay hard, this grinds lower over quarters.
Design (L1) ÷ Fabrication + Assembly & Test (L2+L3)
Running fabs, or building fabs?
Consumables / Tools thesis ↑
Gases, chemicals & fab subsystems (revenue scales with wafer starts — the fab’s OpEx) against litho + deposition-etch equipment (revenue scales with expansion — CapEx). The build-out maturing into operation, measured inside Rubin itself.
Gases & Chemicals + Fab Subsystems (S18+S17) ÷ Litho + Deposition & Etch (S4+S5)
Cheaper tokens move the marginal dollar from building capacity to running agents
Building AI, or running AI?
Operate / Build thesis ↑
The master ratio: Agentic Infrastructure (the AI operating layer) against Rubin (the build-out). When inference gets cheap enough that agents are economical, the marginal dollar shifts from building the factory to running it.
Agentic Infrastructure ÷ Rubin Build-Out
Do the AI clouds out-earn their suppliers?
Operators / Suppliers thesis ↑
Compute Operators (the neoclouds that buy Rubin-generation silicon and sell agent-hours) against the whole Rubin supply chain. The literal handoff point: utilization economics against equipment economics.
Compute Operators (T1) ÷ Rubin Build-Out
Agentic Infrastructure — agents move from demos to production
Are agents actually running?
Execution / Substrate thesis ↑
Runtime, data & coordination (where agent workloads live) against the compute substrate they run on. When agents move from demos to production, the execution layer grows relative to raw capacity.
Execution (L2) ÷ Substrate (L1)
Is trust the new bottleneck?
Trust / Execution thesis ↑
Identity, security & observability against the execution layer — the operating world’s own S9/S1. If agentic work becomes abundant, trust becomes the scarce step, and governance re-rates.
Govern & Secure (L3) ÷ Execution (L2)
Value lands in the P&L of the businesses that use agents
Do the users capture the value?
Use / Operate thesis ↑
Agentic Winners (the businesses that use agents) against Agentic Infrastructure (the layer they pay). The last handoff to confirm — and the most meaningful when it does: agents producing measurable business outcomes.
Agentic Winners ÷ Agentic Infrastructure
Broader than the megacaps?
Beyond the Gateways thesis ↑
Control Plane + Application Leaders against the Megacap Gateway tier inside Winners. Early regime, the gateways absorb everything; genuine adoption broadens into the pivoting software cohort. The breadth check that keeps the board honest.
Control Plane + Application Leaders (B+C) ÷ Megacap Gateway (A)
The full chain in one line
From building AI to living off it?
Use / Build thesis ↑
The full chain in one line: Agentic Winners against Rubin Build-Out. Has the economy moved from building the AI factory to living off its output? The slowest ratio on the board — and the verdict.
Agentic Winners ÷ Rubin Build-Out
All legs are equal-weight (EW) sub-indices of the three Closelook indices — cap-weight would turn the board into a megacap survey. Composites average their members, each normalized to 1.00 at 2026-06-30; that date is the honest inception (the v2 reconstitution of Rubin 125, Agentic Infrastructure 34 and Winners 40 — earlier history is pro-forma backfill). Three caveats travel with every reading. Geography: Rubin's physical side is Asia-heavy while the operating and user layers are US-heavy, so cross-index ratios breathe with the yen, the Kospi and regional tape — not only with the thesis. Single names: a ratio move can be one constituent's week; readings should be checked against their top driver before they are trusted. Youth: the board's first weeks coincided with a semiconductor distribution phase — early cross-index readings largely inhaled that correction. The instrument is built before the event it is meant to measure, which is the point: Rubin-generation systems ramp in the second half of 2026, and the board will be watching with a baseline already in place.
The first two ratios were incepted in the Kimi K3 Rubin-map report — this board extends them across all three indices. Sub-index charts live on the Rubin, Agentic Infrastructure and Winners terminals.
Investment-diary framing: this is how we measure our own thesis — decision support, not investment advice.