The Drain, Made Visible

Q2 2026 ran our June scenario in real time: the reserve drained, bitcoin was sold at a loss to pay the dividend, and the flagship preferred broke to $70. We scorecard our own note, update the numbers, and audit two claims from the call.

TON618 Capital Research · Prices as of 31 Jul 2026 close (MSTR $93.28 · BTC ~$62,875) · Download PDF · June 2026 valuation note

The quarter that audited the note

Strategy reported its second quarter after the close on 30 July: a net loss to common of $8.62 billion, or −$24.45 per share, against a consensus of +$16.85. Nearly all of it is one line — an $8.32 billion unrealized mark on the bitcoin stack, taken at the quarter-end price of roughly $58,700. Our June note argued that for this company GAAP earnings are noise and the funding structure is the signal. Q2 was the quarter that tested the claim, because both moved violently at once.

The stack itself crossed a line it had never sustainably crossed before: underwater against cost. As of 26 July the company holds 843,775 BTC with an original cost basis of $63.7B — an average of $75,476 per coin against a market near $62,875. The accumulation machine spent six years buying; the last eighteen months of buying happened above the current price.

Source: Strategy 8-K Ex-99.1 (30 Jul 2026), holdings & cost basis as of 26 Jul 2026; BTC spot 31 Jul 2026 (Coinbase). Chart: TON618 Capital.

And the funding side ran, in miniature and in fast-forward, exactly the sequence the June note called the real risk — not a margin call, but a drain. What follows is the scorecard: every material claim from the June note, judged against what the quarter actually did.

The scorecard — June's claims vs. Q2's facts

Verdicts on our own note, same standard we apply to everyone else's. Sources: 8-K Ex-99.1, Q2 balance sheet, and the 30 July earnings webinar.

The drain, quarter by quarter

The third row of the docket is the one that carries the note's title, and it deserves the full tracing. Trace the reserve. It opened the year at $2.25B (2.7 years of coverage), was spent down to $871M — half a year of coverage — retiring converts, and then, with the account nearly dry and a dividend due, the company sold 3,588 BTC at an average of ~$60,000 against a ~$116,000 cost basis: a $203M realized loss, taken to pay a preferred coupon. By late July the reserve had been rebuilt to $3.75B, or 2.1 years of coverage — but rebuilt with equity: roughly $1.28B of common sold through the ATM in July, ~13M new shares, issued at roughly 1.0× NAV.

Source: 30 Jul 2026 earnings webinar (management figures; Jun 30 point computed from the Q2 balance sheet cash + short-term investments against the $1.76B/yr obligation). Chart: TON618 Capital.

Management, to its credit, said the quiet part out loud: “we learned the hard way” that the reserve was too thin, and a formal policy now targets 2–3 years of coverage with a 1-year floor. The candor is genuine and we note it. So is the arithmetic: the reserve was rebuilt by issuing equity at roughly NAV — dilution as the price of credibility — and the coupon it defends grew 13% in a single quarter.

Meanwhile the funding mix narrowed. Of the $8.4B raised through the ATM in Q2, every dollar came from exactly two instruments: $2.9B of MSTR common and $5.5B of STRC. The other four preferreds — STRK, STRF, STRD, STRE — raised zero. And STRC itself now trades near $89.50 against a $99–100 objective, which is why the company has begun buying it back at a 13% discount while still issuing it at par through the ATM. When a company is simultaneously the largest buyer and the largest seller of its own funding instrument, the instrument is no longer just a funding channel. It is a price target.

The $5 billion that isn’t — and the 3.3% a year that is

Within a day of the call, the wire consensus hardened into a single number: Strategy authorizes up to $5 billion in bitcoin sales. The company’s own documents say something more specific, and the difference is worth a section. Slide 28 of the earnings deck states the BTC Monetization Program in three lines: sell bitcoin to (1) fund the USD Reserve up to $1.25B; (2) fund preferred dividends and interest — currently $1.76B annually — or replenish the Reserve after paying them; (3) fund the repurchase programs, currently capped at $2.00B. Sum the three buckets and you get $5.01B — the headline. But no company document states $5B as a sale authorization, and the buckets are not alike: the first is capped and one-time, the third is discretionary, and the second is per year, forever.

The “$5 billion” that was actually spoken on the call refers to something else entirely. In the CEO’s prepared remarks (caption track, 00:21:34): fund the USD Reserve up to $1.25B — “that would take our US dollar reserve up to five billion dollars if we were to do that fully using Bitcoin, which would take us up close to three years” of coverage. The $5B is the destination size of the cash reserve ($3.75B on hand plus the capped $1.25B top-up), not a quantity of bitcoin to be sold. Two identical dollar figures, one a reserve target and one a press-computed sale ceiling, met in a headline; the wires ran the wrong one.

Sizing the real program: the capped reserve top-up is ~20,000 BTC, 2.4% of the stack. Management’s own slide 32 answers the market-impact question — a maximum weekly sale of $19M against ~$26B of average daily bitcoin liquidity, about 0.08% — and on that point they are right: this is not a supply story for bitcoin. It is a composition story for MSTR, and the bucket that matters is the uncapped one. Funding the $1.76B coupon from coin consumes 3.3% of the stack every year at today’s price — about 2.7% net of software gross profit. That drain sets a break-even on the issuance machine, because bitcoin-per-share only grows when accretion from selling stock above NAV outruns it:

Source: TON618 Capital calculations. Accretion from issuing a fraction n of the share count at m× NAV = (1+n·m)/(1+n) − 1; break-even solves accretion = net drain of 2.7%/yr (coupon $1.76B less ~19% software gross-profit coverage, on a $53B stack at $62,875). Deck figures: slides 26, 28, 29, 32 of the 30 Jul 2026 earnings presentation.

At the 2× premiums of 2024 the coupon was a rounding error; at today’s ~1.0× it is a ratchet. Below NAV, no volume of issuance can hold bitcoin-per-share flat — the metric the company brands as BTC Yield turns structurally negative, at roughly −3% a year, for as long as the premium stays shut. One genuinely accretive valve exists inside the program: repurchasing STRC at its current ~13% discount retires a $100 perpetual claim for ~$87 and permanently removes that slice of the coupon — the only bucket that shrinks the drain rather than servicing it. And the sales themselves carry a quiet tax logic the headlines skip: with $18.5B of unrealized losses on the stack, coin sold underwater harvests deductible losses — slide 29 books a $59M potential tax asset against the $203M realized loss already taken.

One dollar, four jobs

That coupon is the drain's fixed flow — $1.76B out, every year, in every market. Here is the question no analyst asked on the call: what refills it? Every remedy management described draws on the same scarce resource:

Schematic: TON618 Capital, from the capital flows described in the 30 Jul 2026 earnings webinar and 8-K.

Returning STRC to par is funded by buybacks. Buybacks are funded by bitcoin sales or by the ATM. The ATM is accretive only while MSTR trades above net asset value — and management's own deck valued it at $2.0B, or ~5.5% of net reserves, at 27 July prices — and at today's post-print price it rounds to zero on our numbers. If the premium closes, the ATM stops feeding the reserve; the reserve stops covering the dividend; the dividend gets paid by selling bitcoin; and selling bitcoin shrinks the collateral behind the credit whose discount started the loop. That is not a hypothetical: it is the observed sequence of May and June 2026, run once and arrested by $9.7B of new issuance.

The June note called this structure “a reflexive premium machine whose accretion only works above NAV.” The addendum's update is simply that the machine has now been observed running in reverse, briefly, in production. The margin between forward motion and reverse is the mNAV premium — five and a half points at the call, and at today's price, on our numbers, roughly nothing.

Two exhibits from the call

Q2 audited our note; fair is fair — we audited the call. We ran the full 2h35m webinar through a structured argument scan. Management was, in the main, more candid than the genre requires — the reserve failure, the realized loss, and the STRC crash were all volunteered with specifics, and the STRC-at-$70 explanation (third-party credit lines withdrawn) actually undercuts management's control narrative, which is the opposite of spin. Two claims, however, do not survive scrutiny, and both are load-bearing for the equity story.

The valuation, restated

Same method as June — software on comps, bitcoin at NAV, senior claims at face — new inputs throughout. Two structural changes since the June note: $1.5B of converts retired at a discount (claims shifted from dated debt to perpetual preferred, roughly one-for-one), and ~28M new common shares, most of them sold to rebuild the reserve.

The call quoted preferred at both $14.4B and $15.4B; the balance sheet resolves it — $14.44B is carrying value, $15.46B is liquidation preference. We use liquidation preference, as in June.

Share count held constant at ~385M; further accretion/dilution ignored. Directional, not point forecasts; probabilities omitted because the expected value is a bitcoin view.

The threshold map moved in an interesting way. The strike is unchanged: bitcoin alone covers all $22.2B of senior claims down to ~$26k, same as June, because total claims are almost exactly what they were — only the mix rotated. But the common's wipe-out price fell from ~$23k to ~$20k, because the reserve tripled: there is simply more non-BTC value under the equity than there was in May. That cushion was bought with ~8% more shares outstanding. Both facts are true at once, and they are the quarter in one sentence: the structure is safer, and the per-share claim on it is smaller.

Carried forward

• The 10-Q (due ~early Aug) — verify every 8-K figure; watch the preferred rollforward and the July ATM share count.
• STRC vs par by ~8 September — management's own 70-trading-day benchmark. We scored their analogy weak (IPO seasoning ≠ post-crash repair); the date is now on the record either way.
• mNAV through 1.0× — the loop's switch. Sustained sub-NAV trading turns the ATM dilutive and the flywheel into a ratchet.
• September 2027 put on the 2028 converts ($183 conversion) — management expects equitization; at today's price that requires a double.
• BTC Yield trajectory — 4.5% YTD vs 22.8% in 2025; the deceleration is the accretion engine slowing in public.
• MSCI / index question — deferred in January, unresolved; a live overhang on a thinner premium.

The June note inferred the drain from the structure. Q2 made it visible: the reserve ran to half a year, bitcoin was sold at a loss to pay a coupon, and the flagship credit broke to $70 before $9.7B of new paper arrested the slide. The balance we carry forward is the same one we opened with — a geared claim on bitcoin, priced at about its net asset value — and the question the next two quarters will answer is whether the premium that refills the reserve comes back, or whether the loop runs again.