Original analysis.
Applied frameworks.

NVIDIA

Structural analysis of NVIDIA’s position in the AI compute capital structure: the fourth house that sells the machines, collects in days at a 75% margin, holds an $82 billion equity portfolio in its own buyers, and drew $25 billion from the same bond market its customers depend on. How the circle is recorded: the same two-billion-dollar CoreWeave investment, re-expressed as a price concession, moves the segment growth rate seven points and gross profit two billion dollars on identical cash. The key is equity, what it opens is debt: three private placements, advance rates from 90 to 70 percent, a margin hurdle above NVIDIA’s own, and the inverted seat where the supplier holds common stock, ninth of nine, while the credit was extended by independent arrangers whose terms moved against the borrower across five months; and now the required share: the two-input identity ((1 − a) / m) that converts an advance rate and a gross margin into the share of a program a placement has to capture to return its cash, run across four facilities and rising from about twenty-seven cents to about forty, the three-group sorting of the $63,440m equity book (machinery in place, machinery incomplete, relationship runs the other way), the $105 billion Pike County residual value guaranties filed under Item 2.03, and the historical analogues from Lucent and the 2007 repo haircuts showing the sequence of what moves first. Quality of cash, circular financing, revenue quality, and the twelve structural elements. Grounded in the 10-Q, the 10-K, the 13F-HR, the proxy, the credit agreements, the Form 8-K, and the accounting standards that govern what each discloses.

11 pieces · May–August 2026

Quoted by name in MarketWatch’s feature coverage of NVIDIA’s circular financing (June 2026).

SpaceX

Structural analysis of SpaceX from S-1 through the first public quarter, capital structure, the $235B cash gap widened to about $400B with every added dollar going to the AI build, the Terafab fab gone quiet, the $25B minimum-cash commitment whose size reached the public through Fitch rather than the filings, the fixed-baseline framework (a company changing its mind is the event a fixed baseline exists to catch), and the state of the merger: reading the Tesla-SpaceX entanglement through two filings, finding the combination already happening through instruments that require no vote, and what a formal deal would make invisible.

19+ pieces · May–August 2026

Anthropic

Structural disclosure analysis of Anthropic’s pre-filing landscape — 999 counterparty filings naming a company that has filed nothing, the ~$35 billion SPV financing the Google-TPU build-out, and the judgment calls the S-1 will make about what the record already holds. Reading across Amazon, Google, Broadcom, SpaceX, CoreWeave, and Apollo filings to reconstruct a picture whose disclosure spreads across companies because the arrangements spread across companies.

2 pieces · July 2026

Alphabet

The largest quarterly profit in corporate history, $112.2 billion, was mostly a mark: $99 billion of unrealized gains on two bets, one the company cannot sell and one no one can price. The operating quarter underneath was excellent and stood without it. Quality-of-cash framework applied to the record, the reconciliation, the two marks, and what the next week’s filings from Microsoft, Amazon, and Alphabet itself will test.

1 piece · July 2026

Apple

Structural analysis of Apple as the fifth house — the toll-taker at the edge of the buildout whose cleanest revenue is another house’s money. Introduces materiality as a disclosure tool, formally defines what a house is, reads the Apple-Google circle through the quality-of-cash lens, and now reads the quarterly filing itself: $117 billion of operating cash flow against $6.8 billion of capex, a gross intangible asset line born across three quarterly filings, a Services record, a CEO transition, and a ROIC that outruns the builders by reading the balance sheet they chose not to build.

2 pieces · July 2026

CoreWeave

Structural analysis of CoreWeave’s unit economics, revenue structure, capital requirements, and the quality of cash held together: three statements presenting three businesses, the convergence test, the residual question, the $1.3 billion customer-liabilities reclassification, Amy Hood’s “short-lived assets” testimony, and the four-corners ROIC test applied to a pure-play with no cover, examined through the quality-of-cash framework. The synthesis piece, written before Q2, sets the questions the filing has to answer; one filing read across three disclosure genres, nine exhibits carrying only the filing’s own words; one credit agreement read as a participant ledger, the nine-step waterfall, the price of the last seat, and the structural inversion; and now the capstone: three facilities in one hundred and thirty days, the return on invested capital at negative 0.74 percent against the supplier’s 58.6 on the same construction, the friction that lifts the hurdle above 10 percent, the four parameters of circular financing, and the finding that the corpus holds value to the participants independently of its own return; and now the activation measure: 58 percent of committed capacity that the balance sheet can turn on from cash after scheduled principal, a three-test elimination (contribution, leverage, cure) finding common equity as the only instrument that passes all three, three lending models in thirteen months from asset-based to advance-rate-only with the fee gross-up removed, the lock priced at about $87m whether it opens or not, a nine-step waterfall paying eight participants before the residual, and two draw-period deadlines whose calendar the draws themselves shorten.

8 pieces · June–August 2026

OpenAI

Structural analysis of OpenAI’s operating economics, the quality-of-cash ratio as a single-number diagnostic (cash to committed uses, about five cents per dollar), the $122B Series 7 round, the RPO chain connecting Oracle, NVIDIA, and SoftBank to OpenAI’s listing, and the timing question behind the commitments, grounded in audited financials and primary filings.

4 pieces · June–August 2026

Oracle

Structural analysis of Oracle’s fiscal 2026 annual report — two businesses in one column, $638 billion in backlog, $54.4 billion raised in a single year, and the quality-of-cash framework run through the full statement of cash flows.

3 pieces · June–July 2026

AI & Market Structure

Cross-company analysis of AI platform economics, market structure shifts, and the capital structure of the compute buildout, who borrows, what each is made to promise, what the indentures reveal about the houses and the players, a four-rung grading of every dollar in the buildout by what must still happen for it to become cash, a commitment-disclosure reading across six companies’ filings, a single ratio (cash to committed uses) run first on OpenAI and Apple with the ledger and the donut-and-bubble comparison, the counterfactual where the form of a single two-billion-dollar arrangement moves the segment growth rate seven points without moving a dollar, one filing read across three disclosure genres, one credit agreement read as a participant ledger, and now nine parties at a former uranium enrichment site in Pike County, Ohio, four accounts of one commitment describing three different instruments, the filed guarantee book at $3.5 billion against a reported ceiling thirty times that, the Broadcom precedent from one quarter ago, the test that August 26 settles, and the activation measure: what percentage of committed capacity a leveraged balance sheet can turn on from cash after scheduled principal, a three-test elimination finding common equity as the only key that fits every lock, three lending models across four agreements in thirteen months, and a nine-step waterfall paying eight participants before the residual, the required share identity that measures what fraction of every committed dollar the guarantor must recover from its own customers, three groups of equity sorted by what each book can actually pay, $105 billion in guaranties filed at one address in Pike County, Ohio, and the historical sequence from retained interests to committed-versus-drawn to advance rate to required share. The bubble, if one is here, arrives at one address at a time. The supplier addition reads the ladder from both sides at once, finding six positions toward customers and four toward suppliers at a single company, the two rungs whose reflections carry no name also carrying no filed instrument, the Lucent and Cisco distinction (what was advanced decides what the failure leaves: credit leaves a creditor, a commitment to buy leaves a write-down), and the vendor lending rung empty on both faces while the arrangement the phrase describes occupies ten positions across six rungs.