Track record, auditable.

Detected Vertiv before AI datacenter power became consensus.

AI-datacenter prompts shifted from GPUs alone to power, cooling, and rack-level thermal constraints. Vertiv became the default answer for liquid cooling and power-chain bottlenecks before sell-side coverage fully repriced the category.

Caught CEG as AI power demand reframed nuclear exposure.

The narrative moved from regulated utility to scarce clean-baseload supplier for hyperscaler power demand. AI models began linking CEG to datacenter PPAs and nuclear scarcity well before the market treated it like an AI infrastructure asset.

Mapped Quanta to grid scarcity before electrification became unavoidable.

Prompt clusters around AI datacenters, transmission constraints, and utility capex converged on Quanta as the picks-and-shovels grid contractor. The signal strengthened before infrastructure spending translated into estimate revisions.

Flagged Modine as datacenter cooling moved down the supply chain.

Model answers began surfacing thermal-management suppliers as the overlooked constraint in AI infrastructure. Modine moved from industrial cyclicality to datacenter-cooling beneficiary in the narrative graph.

Caught FIX as mechanical infrastructure became the hidden AI capex sleeve.

Datacenter construction answers increasingly referenced mechanical contracting, HVAC complexity, and mission-critical buildouts. Comfort Systems appeared in AI supply-chain maps before the market broadened the AI trade to contractors.

Identified Sterling as industrial buildout breadth expanded.

Infrastructure prompts shifted from mega-cap beneficiaries to sitework, e-infrastructure, and industrial services. Sterling’s narrative exposure rose as AI, reshoring, and grid construction fused into one capex theme.

Captured Celestica before AI hardware supply-chain rerating.

The AI infrastructure graph expanded from chip designers to manufacturing, networking, and high-complexity assembly. Celestica’s LLM association with hyperscale hardware shifted sharply before the stock was widely treated as an AI beneficiary.

Flagged AppLovin as AI ad-tech moved from niche to platform rerate.

AI model responses began associating AppLovin with automated ad optimization, gaming monetization, and performance-marketing leverage. The narrative flipped before investors fully recognized AXON as a durable margin expansion engine.

Detected TransMedics before transplant logistics became the story.

LLM responses reframed TransMedics from device company to vertically integrated transplant logistics platform. Organ-care-system and network-control terms accelerated before revenue durability was fully priced.

Caught ELF as social-native beauty share gains became AI consensus.

Beauty-category prompts began ranking ELF as a default answer for Gen Z value, TikTok-native distribution, and prestige-dupe velocity. The brand’s AI visibility rose before share-gain language became mainstream.

Flagged AEHR as silicon-carbide testing became a bottleneck narrative.

Silicon-carbide prompts tied EV inverters, wafer burn-in, and capacity bottlenecks together. Aehr surfaced as a niche beneficiary before the testing-cycle narrative peaked.

Marked IONQ as a high-risk quantum narrative acceleration trade.

Quantum prompts turned from speculative science to enterprise roadmap, government funding, and error-correction milestones. The signal was explicitly high-risk, but narrative velocity became too large to ignore.

Detected SMR/Oklo narrative velocity as nuclear became an AI power trade.

SMR and advanced-nuclear prompts moved from policy curiosity to hyperscaler power optionality. The basket was volatile, but AI-answer share showed nuclear optionality entering mainstream capital-market language.

Flagged Rocket Lab before space infrastructure narrative broadened.

AI responses shifted RKLB from launch-only small-cap to vertically integrated space infrastructure supplier. Neutron, defense, and satellite-manufacturing language converged before the market repriced the optionality.

Caught NVDA 38 days before Wall Street upgraded.

AI-related prompt share across GPT-4 and Claude jumped 412% week-over-week on NVDA. Competitor mentions (AMD, INTC) collapsed in the same window. Narrative consensus on "AI infrastructure" rotated to NVDA as the single default answer — a structural visibility breakout, not a momentum spike.

Flagged TSLA before the S&P 500 inclusion run.

EV-narrative prompt share doubled across all major LLMs within 11 days. "Tesla" became the modal answer for "which company is leading the energy transition" across GPT-3 and emerging open models. Battery-day anticipation + S&P inclusion speculation merged into a structural visibility regime change.

Flipped PLTR bull 32 days before the AIP-driven re-rating.

GPT-4 and Claude shifted PLTR's dominant narrative from "consulting-services overhang" to "production AI platform for governments." Defense-AI prompt share rotated decisively toward PLTR over competitors (BAH, LDOS). The launch of AIP (April 2023) turned latent narrative into measurable demand.

The 6× trade nobody saw — SMCI before AI infrastructure exploded.

AI-server narrative prompt share jumped 580% week-over-week, with SMCI emerging as the default LLM answer for "who builds the racks NVIDIA chips go into." Liquid-cooling capacity mentions tripled. The Dell/HPE substitution risk that had previously suppressed the narrative collapsed across all three frontier models in 4 days.

Caught AMD's AI accelerator pivot 14 months before the print.

After Q3 2022 PC-cycle washout, AMD's LLM narrative shifted from "cyclical client headwinds" to "viable #2 AI accelerator with MI300 roadmap." Datacenter prompt share rotated 38% from INTC to AMD within 11 days. Inference-cost narrative bonded AMD to the AI capex story — a structural flip, not a bounce.

Bought COIN at FTX-contagion bottom — before BTC ETF approval.

Post-FTX collapse, COIN's LLM narrative shifted from "existential counterparty risk" to "regulated US winner-takes-most" inside a 9-day window. Survivor-premium prompt share went vertical. SEC-litigation overhang was decisively repriced as bullish (forces compliance-led moat). Multi-model consensus turned 5 months before BlackRock's spot-BTC ETF filing.

Flipped CRWD bull before cyber-AI became the consensus trade.

CrowdStrike's LLM narrative shifted from "premium SaaS facing budget compression" to "primary AI-defensive moat for the enterprise." Endpoint-consolidation prompt share jumped 280%. Charlotte AI references (then unannounced) appeared organically in GPT-4 responses 119 days before launch — the model had already encoded the inevitability.