The AI Extraction Investigation: An Invitation

WHAT WE'RE INVESTIGATING, AND HOW YOU CAN HELP A plain-language guide to the AI Extraction Investigation October 2026 (Frozen against the investigation ledger as of October 5, 2026. Every claim below matches what the records establish -- no further.) WHAT THIS IS AI companies, hyperscalers, infrastructure providers, and their financiers are committing enormous sums to AI infrastructure -- data centers, chips, power plants, and the financing around them. Our governing question is simple: where does the dollar enter the system, where does it travel, and where does the risk ultimately exit? We're not starting with an answer. We're following the money through public records: SEC filings, government documents, contracts, utility proceedings. When the records establish something, we'll say so. When they don't, we'll say "unresolved." And if the evidence disproves the thesis, we'll follow it there. A dead end is a finding, not a failure. WHAT WE'VE FOUND SO FAR 1. Many of the giant numbers you've seen in headlines aren't cash. They're future commitments, leases, guarantees, and contract ceilings. A "$21 billion deal" can mean "$21 billion IF we deliver working computers for six years." We separate every number by what it actually is. 2. The financing is tangled. Companies across the AI ecosystem invest in one another, buy from one another, provide financing, and sometimes provide guarantees or other credit support. The Bank for International Settlements -- the central banks' bank -- published research in October 2026 finding that roughly 55% of the investment value measured in its dataset involved AI firms investing in other AI firms, and identified potential opacity and financial-stability risks from these relationships. 3. Customers are helping finance the buildout. Oracle disclosed that $75 billion of its large AI contracts involved customer prepayments or customer-supplied hardware -- contributions that substantially reduce the capital Oracle itself has to raise. 4. The money trails we've tried to follow hit the same wall. We attempted to trace AI infrastructure spending through to actual end customers -- the businesses and people whose payments ultimately support it. Our CoreWeave-to-Microsoft chain, for example, terminates at "unresolved": Microsoft's public disclosures do not identify the end customers associated with the relevant CoreWeave capacity. That isn't proof that demand isn't real. It means the public record doesn't currently let us establish where the demand ultimately comes from. 5. Your state government is buying AI too. Massachusetts signed a contract giving up to 40,000 executive-branch workers access to OpenAI's ChatGPT tools. The revenue department built a "Child Support Payment Predicter" (its official name) that won a state innovation award -- and public records we've located do not yet explain what the model predicts, what variables it uses, or what downstream action, if any, follows a prediction. 6. Surveillance infrastructure is being networked. Police departments share gunshot sensors and camera networks across cities. In the Pioneer Valley, a police regional hub pulls in ShotSpotter feeds from neighboring cities and a camera network reported at over 500 cameras -- including private cameras volunteered by businesses -- with school camera integration underway. 7. AI is often deployed to act on people, rather than alongside them. In the deployments we've examined, AI is more often used to make decisions about people than to give people greater power over decisions that affect them. We have found far fewer systems designed to give the person on the other side an AI advocate of their own. If you know of one, we want to hear about it -- the asymmetry is a thesis we're testing, not a settled fact. WHAT WE DON'T KNOW (THIS IS WHERE YOU COME IN) - We haven't yet found a complete, auditable money trail from an AI infrastructure investment through to a named end customer whose own use is established in the public record. If you know of one, that's the most valuable thing you could bring us -- prove us wrong. - We don't know who the child support predictor flags or what happens next. Public records requests are in progress. - We don't know what the AI guarantees are priced on -- in particular, what assumptions they make about what used computer chips will be worth in five years. If those assumptions are wrong, someone absorbs the loss. - We haven't mapped the workers: who builds the data centers, who labels the data, who does the human work behind "AI." - We haven't done the honest case FOR the buildout. If this is all legitimate infrastructure investment, we want to know. We actively try to prove ourselves wrong. HOW YOU CAN HELP You don't need to be an expert. You need curiosity and a browser. - Pick one open question above and dig. Everything we use is public. - If you work somewhere that buys AI services, tell us who from and what for. That's demand-side evidence we can't get from SEC filings alone. - If you've been scored, flagged, or decided by an algorithm -- in child support, in a hospital, by police tech -- your story is data. The pattern we keep finding is: a person enters a system, the system scores them, the score follows them. Documentation is the first step. - If you find something, send the document -- not just the claim. Tell us what it says, where it came from, when it was published, and what it does NOT establish. That's the standard everything here is held to. THE RULES - Records, not verdicts. We don't accuse; we document. - A relationship isn't automatically a circularity finding. Two companies can invest in each other, buy from each other, or have a commercial relationship without proving the underlying demand is circular. We trace the source of payment until the evidence establishes genuine end use -- or until the trail becomes unresolved. - We separate what we know from what we infer from what we don't know. Every number carries its source and its limits. - We try to falsify our own thesis. If the evidence says the system is sound, we'll report that. - No one speaks for anyone else. Your story is yours. This investigation is part of the Memory Ark -- a permanent public record built by ordinary people documenting how institutions actually work. The method is free, the tools are public, and the work belongs to everyone who does it. Contact: [Ricky to fill in]