Published research.
Thematic, empirical, and valuation work — built from primary filings and first-party data, free to read.
Published notes
Twenty Times Cheaper — Ten Catalysts That Decide Whether It Happens
Launch got twenty times cheaper once and built exactly one profitable business. The next step rests on three milestones nobody has flown, and the only demand big enough to justify it carries a thermal bill no one has priced. Ten dated tripwires say which 2036 arrives. Follows our SpaceX pre-print.
Read the note →·PDFThe Last Moat Is the Coupon
Strategy says its last business is distribution — wrapping bitcoin in paper the mandated money can buy. We went looking for that buyer in the SEC's own records and in what management said on 17 August. What we found instead: a coupon that only goes up, and a bill paid by the common. Follows The Solvency Map.
Read the note →·PDFThe Visible Hand
In forty days Intel's register was rebuilt by SoftBank, the US government and NVIDIA — and then the President's posts started moving the stock. Sixteen filings disclose the state's hand; none will disclose his. A companion to The Fifth Seat.
Read the note →·PDFThe Fifth Seat
NVIDIA's $500 billion compute-financing machine, read seat by seat. The CDS everyone is quoting is a nine-month-old instrument pricing hedge demand, not solvency — the filings show the "seller" grew its private-investment book 13× while arranging other people's capital. The August 10 platforms add a seat no prior wave had — the originator, who earns on the risk without holding it — and the end of the pipe is the annuity holder, long GPU residual value and take-or-pay credit. Five audit points pre-registered for the Q2 10-Q. A follow-up to our Losing Bets ledger.
Read the note →·PDFThe Solvency Map Is Not the Credit
Strategy published an interactive chart of the bitcoin prices at which its converts and preferreds go undercollateralized. We reproduced its floors to the dollar and reverse-engineered its formulas — then found the market pricing the same preferreds 5–63× wider than the model grades itself. The floors are real; the spreads are not. A follow-up to our June valuation and Q2 addendum.
Read the note →·PDFThe Losing Bets That Built the World
Two hundred years of technology buildouts keep one ledger: the railways, the grid, radio, fiber — the technology won, the builders' capital paid for it, and the surplus migrated permanently to the people who used the cheap capacity. Five waves with primary-source figures, the counter-ledger of the builders who won and the structural features they shared, the four seats at the table — and the AI buildout mapped seat by seat, with the surplus migration measured live. The historical capstone of our AI-infrastructure series.
Read the note →·PDFThe Drained Reservoir
Strategy's USD Reserve fell 61% between January and May with no balance disclosed in between — and STRC's break from $100 dates from the week the number resurfaced. The anatomy of a broken and repaired peg: the June 29 framework and the refill carried the preferred from $74.57 back toward par, while the buyback's measured price effect rests on a single session, tested against the sibling preferreds that got none.
Read the note →·PDFThe Price Nobody Charged
Twenty-one consumer companies reported inside one window, and nearly all sold more than a year ago — yet the businesses that physically serve the customer earned no more for it. Reading all twenty-one filings in full turned up a three-way split we did not expect, and a reading of the American household balance sheet that no official statistic can see. We pre-register a call on Tuesday’s Household Debt and Credit report.
Read the note →·PDFThe Tripwires: A Six-Week Re-Score
Six weeks after our first quantum assessment set three pre-registered tripwires, we go back and score them. The logical-qubit frontier moved 48→70 (claimed), the quantum stocks fell 10–26% through the field’s best-ever news cycle, and Bitcoin’s fix got funded and specified — but not activated. One tripwire is warm; four new ones close the note.
Read the note →·PDFThe First Print, Scored — SpaceX (SPCX)
We published five gauges the night before SpaceX's first-ever earnings report so the scoring could not be revised afterwards. Three cleared, two missed — and the guide we were told did not exist turns out to be the most aggressive number in the quarter: $100 billion of annualised revenue by December, against $23 billion today. Fair value rises 18% to $78.51; the AVOID stands. And 911.5 million shares come free to trade on 6 August, into a float of 646 million.
Read the note →·PDFThe Fade Line, Measured — Palantir (PLTR) After the Print
The sequel scores the print: every pre-registered threshold cleared, fair value restated $112 → $149 under rules written before the answer — and by the open the market had repriced through it. At $154 and 59× sales, richer than before the print, the burden of proof now rides with the bulls.
Read the note →·PDFThe First Print: What $111 Still Has to Believe — SpaceX (SPCX)
SpaceX reports tomorrow — the first earnings print in its 24-year history — down 31% from its first close toward the $66.50 fair value of our June initiation. What the print must show to justify even $111, five gauges pre-registered the night before, a first-print straddle with no base rate for anyone, and the Tesla-merger arithmetic the descent is repricing weekly.
Read the note →·PDF·The original valuation →The Fade Line — What Palantir's (PLTR) Price Actually Demands
Palantir reports tonight at 56× sales. We run the valuation backwards: the price demands ~31% revenue growth for a decade, and the whole bull–bear fight compresses to one number — the fade rate. Breakeven ×0.78; weighted fair value $112. Falsifiable within hours.
Read the note →·PDFThe First Invoice
Two weeks ago we said Big Tech's depreciation bill was coming. Q2 delivered the first instalment, so we scored our own note in public: four for four on accelerating depreciation, a capex assumption that missed by 1%, and Amazon's free cash flow turning negative. The clocks never moved. Scores The Depreciation Question.
Read the note →·PDFMicrosoft (MSFT) — The $116 Billion Question
It spent $115.9bn on equipment in a year, kept Big Tech's widest server depreciation range, and earned $17.95 a share. On a three-year clock that becomes $16.96 — and on the day it filed “no change,” it stretched a different clock instead. Fair value ≈$403 vs $451. Also under Tech Valuation.
Read the note →·PDF·Track record →Amazon (AMZN) — The Honest Machine
The only company in Big Tech that ever admitted its servers wear out faster — and it paid double its own estimate for saying so. This quarter AWS grew its fastest in eighteen; trailing free cash flow went negative anyway. Fair value ≈$142 vs $272. Also under Tech Valuation.
Read the note →·PDF·Track record →Apple (AAPL) — The Control Group
Four of the Mag 7 spend $1–2 on AI hardware per dollar earned. Apple spends seven cents, and its capex is falling. We ran our depreciation clock over it and the re-mark rounds to zero: the cleanest earnings in the index — attached to its most expensive price. Fair value ≈$208 vs $309. Also under Tech Valuation.
Read the note →·PDF·Track record →Meta (META) — Paying Twice
Every server is paid for twice: cash today, earnings tomorrow. In the June quarter the first payment hit $31.1bn against $15.8bn of net income — and the second, on Big Tech's longest server clock, has barely begun. Fair value ≈$370 vs $539. Also under Tech Valuation.
Read the note →·PDF·Track record →Alphabet (GOOGL) — The Convention Price
Reported P/E of 17. Strip one accounting gain and it is 33. Three Street conventions — stock comp added back, a softer discount rate, a higher terminal growth — reproduce the market price to within $1.06. You are paying the convention price. Fair value ≈$256 vs $318. Also under Tech Valuation.
Read the note →·PDF·Track record →Strategy (MSTR) Q2 2026: The Drain, Made Visible
Our June note called the risk a slow funding drain rather than a liquidation. Q2 ran it in real time: the dollar reserve fell to half a year of coverage, bitcoin was sold at a $203M realized loss to pay a preferred dividend, and the flagship preferred broke to $70. We scorecard our own eight claims — including one we got wrong — and audit two claims from the call. Fair value ≈$94, against a stock that now trades at exactly its net asset value. Companion to What Are You Really Buying? Also under Tech Valuation.
Read the note →·PDF·The original valuation →·Track record →The MSTR Era vs the ETF Era
Three and a half years as the market’s levered bitcoin proxy moved Strategy two index points against simply holding the coin. The premium everyone credits to that monopoly showed up only after the spot ETFs listed — and nine-tenths of it is already gone. Strip out the leverage and six years of alpha comes to −0.1% a year. Companion to What Are You Really Buying?
Read the note →·PDF·The valuation →Coinbase (COIN) Q2 2026: The Scissors Open Wider
Coinbase just posted record market share, record USDC balances, and a new $100M product line — alongside a revenue miss, a third straight GAAP loss, and a worse July. Trigger check on our July valuation: nothing tripped, fair value unchanged at ≈$102. Companion to The Right Company at the Wrong Price. Also under Tech Valuation.
Read the note →·PDF·The original valuation →The CLARITY Act: Sell the News, or the Catalyst?
Every crypto law so far has moved Bitcoin roughly 0%. Yet a 0.25% advisor-platform allocation would need to buy 44% of every coin on every exchange — and the shelves sit at an 8.5-year low. Three scenarios for passage, each with a checklist you can score in real time. Companion to CLARITY’s Endgame.
Read the note →·PDF·The odds: CLARITY’s Endgame →The Staircase Turn
The Fed walked rates down three steps in late 2025. Inflation’s three-month pace has climbed every month since — from 2.7% to 6.3%. Ahead of Wednesday’s decision, the market prices a 28% chance the next move is up, and the cut contracts are dead at a fifth of a penny. Written before the meeting, on purpose — and scored on 2 August in an addendum on the same page: the hold landed with three dissents for a hike, and the inflation streak broke the next morning.
Read the note →CLARITY’s Endgame: Recess, the September Window, and the Midterm Freeze
In June we put ~13% on a before-August signing of the crypto market-structure bill (the original note) against a market at 18% — that contract is dying at under 1¢. The sequel restages the question: ~2% by the recess, ~18% in the three-week September window, ~35% for 2026 at all.
Read the note →·PDF·What passage means for BTC →Tesla (TSLA) — Four Times the Finish Line
Eighty-six percent of Tesla's $1,088bn enterprise value is the price of robotaxi and Optimus. That price embeds roughly four times the company's own board-set milestone of one million robotaxis — at one-seventh the valuation the award attaches to it. Unsupervised vehicles counted today: about 21. Also under Tech Valuation.
Read the note →·PDF·Track record →The AI Infrastructure Map
One profit migration, two forces, four notes. The premium is leaving the asset-light hyperscalers for the makers of hardware and power — and it splits as it lands. The map that connects the series, and the place to start: the chips are the trade; the power is the investment.
Read the guide →·PDFGE Vernova (GEV) — The Sold-Out Decade
The purest winner of the AI power buildout beat Q2 and raised guidance — and fell 8.7% anyway, because at ~25× its own 2028 target the good news was already paid for. Backlog $176bn, turbines sold out to 2030. Fair value ≈$672 vs $985 — the multiple, not the business, is the risk. Also under Tech Valuation.
Read the note →·PDF·Track record →NVIDIA (NVDA) — Priced for the Clock Not Breaking
The best business in the AI complex at $4.9tn: a reverse-DCF says $203 requires ~$1tn of revenue by FY2031, and NVIDIA's revenue is, by identity, other companies' capex. Fair value ≈$163 vs $203 — great business, full price. Also under Tech Valuation.
Read the note →·PDF·Track record →The AI Capex Payback Clock
Microsoft, Alphabet, Amazon and Meta will spend roughly $700bn on AI infrastructure this year. On a three-year hardware clock that capital must earn about $221bn a year to clear its cost — yet crediting every dollar of AI profit, the 2026 vintage still falls ~$137bn short. It clears only if the spending stops growing.
Read the note →·PDFThe Power Wall
The premium in AI is migrating from the hyperscalers to the hardware makers — and it splits as it lands. Silicon is a shortage rent that prices away in two years; power is a bottleneck the buildout can't outrun — a five-year grid-connection wait, turbines sold out to 2030. The chips are the trade; the power is the investment.
Read the note →·PDFStructural, or Just Winter?
Two arguments say Bitcoin's trouble is permanent this time: the gamblers left for prediction markets, and the mining megawatts pay 10–20× more serving AI. Tested against flows, filings, and 20 GW of tenant-less power, one is mostly rejected — and the other is true about the wrong thing.
Read the note →·PDFThe AI Buildout, in Four Notes
Four notes, one question: when the AI buildout is financed like this, what is left for the investor? The map of the series — two lenses, two tests, and the two undisclosed numbers the whole complex is priced on. Start here.
Read the guide →·PDFThe Neocloud Fuse
CoreWeave borrows against its assets at Treasuries +200bp and against itself at roughly +534. That wedge is the market's verdict on neocloud credit — and the story of how a $1.4 trillion headline shrinks to a fraction once you sort contracted from announced.
Read the note →·PDFOracle (ORCL) — The Round Trip
The stock that round-tripped: a 64% fall took Oracle from pricing its nameplate backlog to pricing our base case. Legacy Oracle is worth ~$72 alone; at $124 you pay ~$52 for the AI bet — full value for an outcome that must go right. Fair value ≈$114. Also under Tech Valuation.
Read the note →·PDF·Track record →CoreWeave (CRWV) — The Thin Floor
The thesis in one security. Its revenue is more contracted than its peers' — take-or-pay, prepaid, Microsoft-anchored — but the signed book covers a seventh of the price, and the rest is a levered option on how long a GPU stays useful. Fair value ≈$60 vs $73. Also under Tech Valuation.
Read the note →·PDF·Track record →The Depreciation Question
The buyers of AI compute book their servers over five to six years; the silicon is superseded every two. Restated on a three-year clock, six income statements shed 4–17% of reported earnings — and roughly $221bn of recognition gets deferred into 2026–28. Every input from the companies' own filings. Scored out-of-sample in The First Invoice (31 July).
Read the note →·PDFThree Machines, One Asset: The Regulated Crypto Complex
Coinbase, Robinhood, and Block share a sector ETF and almost nothing else — coupled to Bitcoin at ~90%, ~13%, and ~4% of their economics. Separating crypto identity from crypto economics ranks the group in three directions at once: rich, full, and roughly fair.
Read the note →·PDFRobinhood (HOOD): The Price Is a Growth Commitment
The price is a growth commitment. Robinhood broke its crypto dependency mid-winter — but at ~62× normalized earnings, $110 banks a 20%-a-year compounding machine for four uninterrupted years. Weighted fair value ≈$87 vs a $109.86 market price. Also under Tech Valuation.
Read the note →·PDF·Track record →Block (XYZ): The Most Misread Name in Crypto
The most misread name in crypto. Reported revenue overstates the business and a tripled loss line is mostly accounting timing — underneath sits a mid-teens gross-profit compounder. Weighted fair value ≈$83 vs $78.72, conditional on one credit signal. Also under Tech Valuation.
Read the note →·PDF·Track record →How Big Does the Bid Get? Sizing the Institutional Bitcoin TAM
Institutions already hold ~$143B of Bitcoin — one-tenth of one percent of a $144-trillion pool. A channel-by-channel build across treasuries, ETFs, pensions, insurers, banks, sovereign funds, stablecoin issuers, and central banks puts the base case at ~$556B and $180–250k by 2030. A second lens — Bitcoin as a store of value rivaling gold, bonds, and real estate — lifts the band to $500k–$1.3M, with full gold parity alone computing to ~$1.4M a coin. And all of it collides with a shrinking float: only ~2.4M liquid coins against demand running ~9× new issuance.
Read the note →·PDFThe Private Credit Reckoning
The $3-trillion private-credit boom is having its first real stress test. How it most likely unwinds, and how bad it could realistically get — with the danger running through the plumbing (banks, insurers, retail wrappers, and the Japanese carry trade) more than the funds themselves.
Read the note →·PDFIs the Bottom Forming? The On-Chain Floor vs. the Macro Ceiling
Whales reportedly absorbed a record ~270,000 BTC into the June low while US spot ETFs bled their worst month ever — the sharpest ownership-transfer divergence of the ETF era. What that signal is actually worth, backtested: conviction buyers build floors, flow buyers build turns.
Read the note →·PDFCircle (CRCL): The Price of Distribution
The price of distribution. Circle earns the interest on $73B of USDC — then pays nearly two-thirds of it away to the partners who distribute it, a share climbing toward zero. Weighted fair value ≈$50 against a $64.62 market price. Also under Tech Valuation.
Read the note →·PDF·Track record →Coinbase (COIN): The Right Company at the Wrong Price
The right company at the wrong price. Coinbase is two businesses — a hyper-cyclical trading franchise and a quietly compounding "financial plumbing" annuity — and on through-cycle earnings, fair value ≈$102 vs $165, returning less than simply holding bitcoin in every scenario. Also under Tech Valuation. Q2 2026 results: see the addendum.
Read the note →·PDF·Q2'26 addendum →·Track record →The Debt Endgame
How the United States most likely resolves a debt it cannot grow, default, or hike its way out of — through financial repression and the stealth inflation tax that quietly shrinks the debt while widening the divide. Who walks through the last open door, and who pays.
Read the note →·PDFStrategy (MSTR): What Are You Really Buying?
What are you really buying? After a brutal de-rating the famous premium is gone — MSTR trades at ~1.0× net asset value, no longer the "$2.50 for $1 of bitcoin" the bears shorted. Sum-of-the-parts fair value ~$86 leaves a ~1.7× levered, convex bet on bitcoin with a permanent preferred coupon. Also under Tech Valuation. Whether that leverage has ever paid: see The MSTR Era vs the ETF Era.
Read the note →·PDF·Valuation model (XLSX)·MSTR era vs ETF era →·Track record →How Close Is Quantum Computing to Being Useful?
Where quantum computing actually stands, in plain English. Error correction crossed threshold in 2024–25, yet nothing economically useful has run. Experts now put a code-breaking machine at roughly one-in-three within a decade — which is why banks, governments, and Bitcoin are already migrating off today's encryption.
Read the note →·PDFWill the CLARITY Act Be Signed Into Law Before August?
Will the CLARITY Act be signed into law before August? A probability estimate triangulated across a prediction market, a conditional-gate model, and an outside-view reference class: central estimate ~13% by 31 July, versus roughly even-money anywhere in 2026.
Read the note →·PDFWhat Actually Drives Bitcoin
A walk through every factor that actually leads Bitcoin — global liquidity, the manufacturing cycle, momentum, the growth×inflation regime, the power-law floor — and an honest scorecard on the canonical assumptions. The read: a high-beta risk asset that follows liquidity and its own trend.
Read the note →·PDFThe TAM of AI Using Bitcoin
Sizing Bitcoin as the monetary/settlement layer of the AI economy, 2026–2030 — agentic machine-to-machine payments, Bitcoin-native rails (L402, Lightning), AI-driven BTC demand, and the miner-to-AI-compute convergence.
Read the note →·PDFDisclosures. TON618 Capital. For information purposes only — general, impersonal information of regular circulation, not tailored to any individual and not issued in connection with compensation from any client. Not an offer to sell or a solicitation of an offer to buy any security, fund interest, or digital asset, and not investment advice or a recommendation regarding any instrument. TON618 Capital has no clients and distributes all research free of charge, and on that basis publishes in reliance on the publisher's exclusion from the definition of "investment adviser" under the Investment Advisers Act of 1940 (§202(a)(11)(D); cf. Lowe v. SEC, 472 U.S. 181 (1985)). TON618 Capital is not registered as an investment adviser or broker-dealer in any capacity; it is a Bitcoin fund and may hold or transact in instruments it discusses, with material conflicts disclosed in the individual notes.
Artificial intelligence is used in the creation of this research; all methodology and data integrity are reviewed and approved by the Fund's Chief Investment Officer, Keyth Beck. CFA® and Chartered Financial Analyst® are registered trademarks owned by CFA Institute; any reference describes the analytical methodology applied and does not imply authorship or review by a CFA charterholder, nor any affiliation with CFA Institute. Past performance is not indicative of future results. Digital assets are highly volatile and may result in total loss of capital. Feedback: keyth@ton618capital.com.