Educational publication. Not investment advice. Neither John Gillespie nor InsightfulAgents.AI LLC is a registered investment adviser.
The AI capex cycle is moving an enormous amount of capital. The narrative companies you read about everywhere are the ones building the data centers, training the models, designing the chips. Underneath all of them sit five businesses whose cash flows are upstream of the headlines. Power, water, copper, natural gas, uranium. The boring inputs.
That is the consensus pitch: own the picks-and-shovels of the AI build-out, not the model-trainers. So I ran the five names most often surfaced in that pitch through OCCAM'S FORENSIC JURY™, the four-lens forensic methodology I use to second-opinion every position. The methodology disagreed with the consensus read on three of the five.
The video below walks through which three, what each lens caught, and why the methodology output should make a long-term holder pause rather than panic-trade.
If the embed does not load, watch the video directly on YouTube at youtu.be/9jR0YEt6oNs.
Why these five
The five names are the candidates most often pitched as AI-survivors in the cash-flow lane. Each one sits upstream of the data-center build cycle. Each one has been in business for decades. Each one pays a dividend that has been compounded by reinvested cash flow, not by issuing new equity.
That is the lane the Forensic Jury runs against. The methodology weights cash flow above GAAP earnings, balance sheet strength above revenue growth, and dividend trajectory above current yield. The questions are: does the business actually convert sales to cash, can it self-fund the next capex cycle without diluting shareholders, and is the dividend record a function of operating quality or financial engineering.
The five stocks under review
Cameco Corporation. The AI build-out needs baseload power that is not weather-dependent. Nuclear is the only carbon-free baseload source at scale. CCJ owns the bottleneck across mining, conversion, and enrichment. The thesis is that as the data-center power draw forecast keeps revising upward, contracted uranium prices and enrichment fees benefit before the spot market does.
NextEra Energy. Two-legged business: the largest renewable-generation portfolio in the U.S. plus a regulated Florida utility. Both legs get tailwind from AI data-center demand. The regulated side gives the cash-flow predictability investors pay for. The renewable side adds the growth narrative. The methodology question is whether the growth-side multiple is being underwritten by the utility-side cash flow, or the other way around.
Enterprise Products Partners. A dividend king with 25-plus consecutive years of distribution increases. Toll-road economics on U.S. shale: EPD does not own the commodity, it transports and processes it. The AI data-center power draw runs heavily through natural gas, both directly and as a backup for renewables. EPD's footprint sits upstream of that demand.
Energy Transfer. The largest U.S. midstream company by scale. Multiple direct natural-gas-to-data-center supply contracts already announced. The methodology read on ET requires looking at debt the way pipeline operators think about debt: as project capital. A pipeline company carries debt to build new projects; when those projects go live, the debt resolves through contracted cash flow.
Rio Tinto. An AI data center uses several times the copper of a conventional facility: interconnects, cooling loops, transmission build-out. RIO is one of the lowest-cost integrated producers in the world across copper, aluminum, and iron ore. The methodology question on cyclicals is always whether you are buying near a trough or near a peak. The video walks through how the four lenses split on that question.
What the methodology is
The Forensic Jury runs each ticker through four independent AI personas with different priors: an Auditor focused on filing quality and disclosure integrity, an Architect focused on capital allocation discipline, a Storyteller focused on narrative versus operating reality, and a Sentinel focused on market signal and third-party data divergence. Each persona produces an independent verdict with confidence and reasoning. The methodology weighted-merges the verdicts but preserves dissent on the public record. When three lenses agree and one dissents, the dissent is the most important output of the run.
The five cards above are setup. The video is where the methodology output lands.
Where this fits
This page is the canonical mirror for the YouTube video published 2026-05-20. The video is also archived at the Internet Archive snapshot at archive.ph/yPdil and the source video and metadata are preserved in the InsightfulAgents.AI compliance archive. The canonical mirror exists so that the methodology output is searchable, citable, and time-stamped on the open web, not only inside the YouTube platform.