Year 4. Picks-and-shovels exhausted. Vertical AI underexplored.
Accelerating · 2022-
early
accelerating
peak
declining
§ The wedge — what we think vs consensus
Pending author input.
Contrarian read not yet authored for this shift. The wedge section will name the consensus position, our differing read, and the structural reason for the divergence.
§ Thesis
What's actually shifting.
The AI Boom is transitioning from infrastructure-build (foundation models, GPU buildout, hyperscaler capex) into vertical-application + agent-economy phase. Capital allocation has rotated: the next 3-5 years are not about who trains the biggest model but about which vertical AI applications produce 10× productivity gains AND can be defended against incumbents. The 'picks-and-shovels' trade is largely priced; the next leg is application-layer differentiation, agent-economy infrastructure, and the regulatory-arbitrage geography that hosts compute.
§ Stage history
How it got here.
2022-Q4
early
ChatGPT public release. Thesis crystallizes for generalist users.
Where the categorical reads land in particular names.
Specific named positions not yet authored. This section will carry tickers / companies / asset-class names with thesis, risk, and sizing notes — the difference between a category read and a position read.
§ Signal tracking
What would tell you the shift is accelerating — or stalling.
Watch for (acceleration)
Agent-economy traction — measurable revenue from autonomous workflows, not demos
Enterprise AI ROI clearing 3× sustained for non-pilot deployments
Vertical-AI exits at $1B+ valuations with revenue-attached, not growth-attached, multiples
AI-native startups crossing $100M ARR in <3 years from launch
Foundation-model price compression to commodity tier with sustained demand still bid
Anti-watch-for (stalling / reversal)
Foundation-model commoditization that destroys training-economics moats faster than vertical wedges form
Sustained energy or compute constraints capping deployment scale
AI regulation pushing deployment compliance cost above $10M / quarter for mid-cap operators
Quantitative watch metrics not yet authored. This section will carry specific named metrics with their threshold levels and current values — the at-a-glance dashboard that turns a description into a tracker.
Key differenceAI is on an AGI trajectory; Internet was application-layer commoditization. Foundation-model labs may not bust the same way pure-internet plays did. Watch for capacity over-build vs revenue-ramp gap (the Internet-boom signature).
Key differenceAI substitutes cognitive + physical labor; Industrial Revolution was mechanical only. Distribution of gains may follow the same pattern (capital first, labor later) but on a faster timeline given communications speed.
§ Related Lab findings
Where the mechanism is rigorously tested.
No Lab finding has been authored on this shift yet. The shift is tracked here as macro frame; rigorous mechanism testing comes when a finding is registered against the corpus.
§ Cross-shift interactions
Where this shift compounds or conflicts with another.
AI labor substitution accelerates as working-age populations contract; the demographic pressure is the demand-side floor under productivity-AI applications.
Bifurcated AI ecosystems form along chip-control + training-data nationalization lines; AI Boom outcomes become path-dependent on which ecosystem an operator is in.
Hyperscaler power demand from AI training/inference is now a primary driver of energy-transition urgency — particularly nuclear renaissance and grid investment.
Programmable money (stablecoins / tokenized rails) is the financial substrate AI agents need to transact. Tokenization is the AI agent-economy enabler.
§ Track record
Prior calls + outcomes for this shift.
No prior calls logged for this shift yet. The track record builds over time as predictions resolve. It’s the credibility ledger — visible past calls and their outcomes, same way the Lab corpus tracks pre-registered predictions.