The Number Is $97.36
The number is $97.36.
That is the average price Strategy paid for 1,810,885 shares of STRC โ its Series C variable-rate perpetual preferred stock โ between Aug. 31 and Sept. 7. Cash out the door: $176.3 million. Par value retired: approximately $181.1 million.
The gap is roughly $4.8 million. That gap is the trade.
STRC is engineered to hold $100. Not marketed as. Engineered. The instrument carries a monthly reset and a variable coupon, and the entire purpose of that construction is that the price converges on par and stays there. A $100 instrument that pays a floating rate is, in theory, a $100 bill that pays you to hold it.
Strategy bought it for $97.36.
Glitch detected. Source traced.
And that is the least interesting line in the filing.
Between Aug. 31 and Sept. 7, Strategy repurchased zero STRF. Zero STRK. Zero STRD. Zero MSTR. It sold zero shares through its at-the-market program. It bought zero bitcoin and sold zero bitcoin. Holdings sat flat at 845,050 BTC, acquired for $63.73 billion, an average of $75,412 per coin.
Cash on hand: $5.10 billion in the USD Reserve โ the pool earmarked to back preferred dividends and debt interest โ plus $1.44 billion of general-purpose USD cash.
The board, in the same 8-K filed Tuesday, approved doubling the aggregate authorization under the Digital Credit Securities Repurchase Program from $1 billion to $2 billion. As of Sept. 7, $1.19 billion remained available.
The week's most important disclosure is not what Strategy bought. It is what Strategy did not buy.
Four engines drive this machine. This week, three of them were idle, and the one that ran was pointed inward. That is the entire story, and almost nobody is reading it that way.
Context: What Strategy Actually Is Now
Start with the category error that most coverage keeps making. Strategy is no longer "a software company that holds bitcoin." That description was accurate in 2020. It has been false for at least two years.
Strategy is a capital structure with a software business attached to it. The product is not MSTR. The product is the capital stack โ the layered claims on 845,050 bitcoin, each tranche sold to a different buyer with a different mandate, a different risk budget, and a different reason to show up.
That stack is worth reading line by line, because the filing only makes sense if you know who is standing where in the waterfall.
The preferred series
The Strike and Stretch preferred series are the load-bearing members of the structure.
STRC โ Series C Perpetual Stretch Preferred. Variable rate, monthly reset, par-anchored at $100. This is the instrument in the filing. It is designed to behave like a floating-rate note: the coupon resets so that the price stays near par, and the holder receives an income stream that tracks short rates without taking the duration risk of a fixed coupon.
STRF โ Series F Perpetual Strike Preferred. Fixed coupon, roughly 10%. Sits senior to common in the liquidation waterfall.
STRD โ Series D Perpetual Stretch Preferred. Fixed coupon, roughly 10%. Same seniority tier.
STRK โ Series A Perpetual Strike Preferred. Fixed coupon, roughly 8%, and here is the part most people skip: the Strike series carry conversion mechanics. Where a series is convertible into common, the fully diluted share count changes, and so does every bitcoin-per-share calculation anyone publishes.
I am reconstructing this taxonomy from public filings and my own notes rather than from a press release. The labels matter less than the mechanics. What matters is that preferred sits in a specific place: below debt, above common, with no maturity date and a cumulative dividend claim.
Why "engineered to hold $100" is not marketing
Par value is not a price target. It is a contractual anchor with three distinct functions.
First, it defines the liquidation preference. In a wind-down, the preferred holder gets $100 per share before common sees anything. That is a real claim on real assets.
Second, it defines the dividend base. A coupon of X percent is X percent of par, not X percent of the market price. A share trading at $97.36 still accrues dividends on $100.
Third, it defines the redemption reference. When an issuer calls or repurchases the instrument, par is the number it negotiates against.
This is why the $97.36 print is strange. A variable-rate perpetual with a monthly reset and a par anchor should not trade at a discount for long. The reset exists precisely to eliminate the price deviation. If the coupon moves with the market, the price has nowhere to go.
The deviation is the signal.
The June 29 overhaul and the weekly heartbeat
This buyback program traces to the financing overhaul Strategy announced on June 29. That package authorized up to $1.25 billion in bitcoin sales alongside separate $1 billion buyback authorizations for preferred and for common stock.
Read that sentence again, slowly, because it contains a disclosure that almost nobody parsed at the time.
The company authorized itself to sell bitcoin. Not to buy it. To sell it.
A treasury operation whose stated purpose is accumulating a scarce asset built in a $1.25 billion liquidation authorization. That is a contingency plan, and contingency plans are confessions. I have written about this pattern since 2017, when I spent forty-eight hours straight debugging an Ethereum pre-sale script before Mainnet launch and found an integer overflow that would have drained a slice of early funds. The bug was not the story. The story was that the team had never modeled the failure mode, and the fact that the code contained no handling for it told me more about the project than the whitepaper did.
Strategy models its failure modes. That is not bearish. It is the opposite of bearish. It is the most professional thing in any of these filings, and I say that as someone who has spent most of her career dismissing press releases in favor of bytecode.
The cadence itself is a mechanism
These 8-Ks now arrive on a weekly rhythm. They function as a treasury dashboard: BTC held, cost basis, ATM activity, preferred repurchases, reserve balances. In a market where most disclosures are quarterly and most corporate treasuries are opaque, Strategy has turned its balance sheet into a live instrument.
There is a cost to that transparency. Every week, the market gets a fresh read on exactly which engine is running and which one has stalled. There is no room to smooth a bad week into a good quarter.
What the market got this week was a dashboard with three dead gauges.
Core: The Arithmetic Nobody Ran
Here is the calculation that should have led every piece of coverage, and did not.
The discount capture
Strategy spent $176.3 million in cash and extinguished approximately $181.1 million of par value. That is a $4.8 million instantaneous gain โ 2.7% on capital deployed, captured at the moment of settlement, with no market risk attached.
But accretion to what is the question that separates analysis from arithmetic.
Accretion to common equity is the standard answer, and it is nearly meaningless here. If MSTR's equity value sits in the tens of billions, $4.8 million is a rounding error. Presenting it as a shareholder victory is the kind of framing that gets repeated because it sounds precise.
The real accretion is on the obligation side. Retiring $181.1 million of par value removes the forward dividend attached to that par. If the variable reset holds near 9%, that is roughly $16.3 million per year of cash obligation that no longer exists.
Against $176.3 million of cash spent, that is a cash-on-cash return of about 9.24% โ locked, contractual, and independent of what bitcoin does next.
Retiring preferred at $97.36 with a 9% coupon yields roughly 9.24% on cost. Buying bitcoin with the same $176.3 million has an expected return that is, in the strict sense, unknowable.
That is the trade. Strategy swapped an unknowable return for a locked synthetic yield, plus a $2.64 discount per share, plus a reduction in the claims standing ahead of common.
Every part of that is defensible. None of it is exciting. Which is exactly why the market read it as noise.
The model
I run these numbers in Python because I do not trust myself to hold a capital structure in my head, and neither should you. Here is the reconstruction.
# STRC buyback accretion model
# Inputs from the 8-K filed Tuesdaycash_spent = 176_300_000 shares_bought = 1_810_885 par = 100.00
avg_price = cash_spent / shares_bought # 97.36 par_retired = shares_bought * par # 181_088_500 discount_capture = par_retired - cash_spent # ~4.79M
assumed_reset_rate = 0.090 dividend_removed = par_retired * assumed_reset_rate # ~16.3M cash_on_cash = dividend_removed / cash_spent # ~0.0924
remaining_auth = 1_190_000_000 weeks_at_pace = remaining_auth / cash_spent # ~6.75
print(avg_price, discount_capture, cash_on_cash, weeks_at_pace) # 97.36 4788500 0.0924 6.75 ```
Two outputs matter.
The first is the 9.24% cash-on-cash. The second is the 6.75.
The pace to exhaustion
At the pace set between Aug. 31 and Sept. 7, the remaining $1.19 billion of authorization is exhausted in approximately 6.75 weeks.
That puts the drawdown somewhere in the second or third week of October, assuming the discount persists and the pace holds. Both assumptions are load-bearing. The pace has been running hot โ the program has deployed roughly $810 million of the $2 billion already โ and a discount that persists is a discount that someone keeps having to buy.
The market should be watching the drawdown rate on this authorization the way it watches an unlock schedule.
Because an exhausted authorization without a renewal is a specific kind of event. It means the marginal buyer of STRC has stepped away, and no replacement has been named. The peg then stands on its own, or it does not.
The dividend wall, and why it is not a wall
Here is where the reserve matters. The filing reports $5.10 billion in the USD Reserve, explicitly earmarked to back preferred dividends and debt interest.
I built a coverage model with labeled assumptions. These are my reconstructions, not company guidance, and anyone publishing a more precise number is guessing harder than I am.
reserve = 5_100_000_000
assumed_pref_notional = 5_500_000_000
assumed_blended_pref_rate = 0.0920
assumed_conv_debt = 8_200_000_000
assumed_conv_rate = 0.012pref_dividends = assumed_pref_notional assumed_blended_pref_rate # ~506M debt_interest = assumed_conv_debt assumed_conv_rate # ~98M annual_obligation = pref_dividends + debt_interest # ~604M
coverage_years = reserve / annual_obligation # ~8.4 print(round(coverage_years, 1)) # 8.4 ```
Roughly eight years of cash obligations, prefunded and ring-fenced.
Sit with that for a moment, because it inverts the popular narrative. The standard bear case on Strategy's preferred stack is a dividend wall: a cash obligation the company cannot meet if bitcoin turns and the ATM window closes.
On my reconstruction, the wall is prefunded well past any plausible stress horizon. This is not a liquidity story. It is a credibility story.
And the buyback is what happens when credibility, not liquidity, is the binding constraint.
That distinction changes the interpretation entirely. A company buying back its own paper to conserve cash is defending solvency. A company buying back its own paper with eight years of prefunded dividends sitting in reserve is defending a price. Those are different problems, and they call for different judgments.
The line that did not move
845,050 BTC. $63.73 billion cost. $75,412 average.
The position is in unrealized gain at any spot price above the average. The sensitivity is mechanical and worth stating plainly: every $10,000 move in the bitcoin price moves the value of the treasury by approximately $8.45 billion.
That number is the reason the rest of the structure exists.
But the position did not grow. Zero coins bought. Zero coins sold.
That is the anomaly. In a bull market, this machine has exactly one purpose: convert equity premium into bitcoin. When the premium is wide, the ATM runs and the treasury compounds. When the treasury is not compounding, one of two things is true โ either the premium has closed, or management has chosen not to test it.
Either reading is information. Neither reading is in the filing.
On bitcoin-per-share, the week was approximately flat. The numerator did not move. The common share count did not move, because retired preferred does not shrink the common denominator unless a convertible series is involved. What did change is the size of the claims standing ahead of common: $181.1 million of par, removed permanently.
Against a fully diluted share count in the hundreds of millions, that is less than a dollar per share of improvement in the residual claim. Real. Small. Not the point.
The ATM line, and the premium question
Zero shares sold under the at-the-market program.
Exchange volume anomaly flagged.
In a bull market, ATM issuance is the accretive engine. The math is unforgiving: selling common at a premium to net asset value increases bitcoin per share for existing holders. Selling below NAV destroys it. The threshold is not NAV itself. It is NAV plus fees, plus market impact, plus the slippage of working an offering into a live book.
When shares stop being sold, the premium has either compressed below that threshold or management has paused deliberately.
Here is the part that should bother anyone paying attention. Strategy funded the repurchase entirely from USD cash rather than from ATM proceeds.
Consider what the alternative would have looked like. Issue common at a premium, use the proceeds to retire preferred at a discount, and you capture both spreads โ the equity premium and the credit discount โ in a single pair of trades. That is the arbitrage the structure was built for. It is the reason the two authorizations were approved together on June 29.
They did not do it. They reached into the cash drawer instead.
A company spending cash to buy its own liabilities rather than issuing equity to buy more bitcoin is running a defensive posture. Calling that posture bullish requires ignoring the direction of the cash.
The reset formula is an oracle
This is the section I would want a reader to remember.
STRC is a variable-rate instrument. The rate is not set by a committee meeting in a boardroom. It is determined by a formula that takes reference-rate inputs and produces a coupon on a monthly schedule.
That formula is an oracle.
I have spent the better part of a decade arguing that oracle feed latency is the most underpriced failure mode in decentralized finance. The argument has always been that price feeds lag, that the lag is exploitable, and that a system whose solvency depends on a feed it does not control is a system with a hidden counterparty.
The same problem now sits inside a corporate capital structure.
If STRC's reset formula references a rate that lags the market, then the coupon is wrong for some window of every month. If the coupon is wrong, the fair value diverges from par. If the fair value diverges from par, the instrument trades at a discount โ not because the credit is impaired, but because the pricing mechanism is imprecise.
A persistent sub-par print on a par-anchored floater is either a credit opinion or an oracle-latency artifact. In practice, it is usually both, and separating them is the entire analytical job.
I flagged this structure in 2020 during the Compound forensics, when I identified a flash-loan vector in the interest-rate model three hours before exchanges halted trading. The flaw was not in the reserves. It was in the rate calculation. The lesson has not changed: the place to look is never the asset. It is the formula that claims to price the asset.
Contrarian: The Parts of the Filing That Imply a Different Company
1. A peg that requires a buyer is not a peg
STRC is engineered to hold $100. This week, Strategy paid $97.36 for 1,810,885 of them.
There is a version of this story in which the company is harvesting a discount created by dislocated sellers, and everything works out. That version is possible.
There is another version, and it is the one I keep coming back to. An instrument that requires its issuer to buy it in order to hold par is not holding par. It is being bid. And a bid has a size. The size is $1.19 billion, and at the current pace it runs out in under seven weeks.
Liquidity draining. Logic broken.
Now the reflexivity loop. It runs like this: short rates fall, the monthly reset produces a lower coupon, income-oriented holders who bought STRC for the carry rotate into something with a better spread, the price falls further below par, Strategy buys more, the reserve draws faster, and less cash is available for anything else. Each turn of the loop makes the next turn more likely.
I want to be precise about the comparison this invites, because pattern-matching on shape alone will lead readers to the wrong conclusion.
In 2022, I spent three months dissecting the TerraUSD mechanism and wrote 15,000 words arguing that the collapse was inevitable because the game-theoretic incentives were broken from the first block. The peg there was defended by minting and burning a sister asset with no external cash flow. There was no reserve. There was no revenue. There was only reflexivity.
STRC is not that. There is a real treasury behind it. There is dividend coverage measured in years. There is a business generating cash. The magnitude of the fragility is not comparable.
The shape of the loop is comparable. And shape-matching is how a market that survived 2022 will misjudge 2026. A structure can be fundamentally sound and still exhibit reflexive behavior in a specific instrument. Both things can be true. Analysis that insists on one or the other is not analysis.
2. The billion dollars that was never spent
The June 29 overhaul authorized $1 billion for preferred repurchases and $1 billion for common repurchases. The preferred side has been drawn down by roughly $810 million. The common side is untouched.
Why would a board authorize a common buyback and then never use a dollar of it?
Because repurchasing common at a premium to net asset value destroys bitcoin per share. If MSTR trades above the value of the bitcoin it holds, buying MSTR is buying bitcoin at a markup on behalf of shareholders who already own it. It is value-destructive by construction.
So the common authorization only becomes rational below NAV.
An unused authorization is a disclosed contingency plan. The board has modeled a scenario in which MSTR trades below the value of the bitcoin it holds โ and has pre-committed capital to act in that scenario.
That is not a prediction. It is a disclosed possibility, and it is the most interesting sentence in the June 29 package. It tells you where management believes intrinsic value sits. It tells you they consider a sub-NAV print plausible enough to warrant a standing authorization. And it tells you that the $1 billion sitting idle is not a war chest for offense. It is a floor defense that has not been triggered.
Watch for it to be triggered. If it ever is, that is not a bullish signal. That is the contingency becoming the base case.
3. The sister series were left alone, and the reason is arithmetic
Zero STRF. Zero STRK. Zero STRD.
The obvious reading is that the program is a support operation for the whole preferred stack, and STRC simply needed it most. That reading is lazy.
Here is the deduction. If short rates have been falling โ the same move that compresses STRC's reset coupon โ then fixed-coupon preferreds become more valuable, not less. A 10% fixed coupon in a 6% world trades above par. A 10% fixed coupon in a 5% world trades further above par.
If STRF, STRK, and STRD are trading at or above par, then repurchasing them destroys value. Every dollar spent buys back less than a dollar of par. The program would be burning shareholder money to retire expensive paper.
A program that buys only the one instrument trading below par is not a rescue. It is an arbitrage. And the fact that management is running it selectively tells you they understand NAV math better than most of the people covering them.
So the correct inference from the untouched series is not "the other preferreds are fine." It is "the other preferreds are priced above par, and that is the company's own revealed opinion."
4. "Digital Credit" is doing more work than any number in the filing
Read the name of the program. Digital Credit Securities Repurchase Program.
Preferred equity is not credit.
Credit sits above preferred in the waterfall. Credit has a maturity date, a covenant package, a defined claim on specific cash flows, and a legal recovery process. Preferred has none of that in the same form. It has a perpetual term, a cumulative dividend, and equity tail risk โ the risk that in a genuine stress event, the preferred holder is treated as a residual claimant wearing a bond's clothing.
The naming is not accidental. Calling this paper "credit" recruits a buyer base that applies credit frameworks: spread to benchmark, coverage ratios, rating-agency logic, duration matching. Those buyers bring mandates, which bring size, which brings a lower cost of capital for the issuer.
Those frameworks are also the wrong instruments for the job. Coverage ratios assume a stable cash flow. The cash flow here is a function of a volatile bearer asset and the market's willingness to keep bidding the structure. Duration matching assumes a defined maturity. There is no maturity.
I have watched this maneuver before, in a different register. When a major payments company issued its own stablecoin, the strategic logic was not primarily about payments rails. It was about perimeter. Better to become a regulated partner than to wait to be regulated. The label determines who shows up, and who shows up determines the rules.
Categories are load-bearing. When a product is renamed into a category, it inherits that category's buyers and that category's blind spots โ and the blind spots arrive first, because nobody re-reads a term sheet for an instrument they already think they understand.
5. NFT metadata mismatch found
Here is the structural problem underneath all of it.
STRC looks like a token with a peg. It has a par value, a reset schedule, a coupon, and an exchange listing. Every surface feature reads as mechanical and deterministic.
The rules governing it do not live on-chain. They live in a certificate of designations โ a legal document, off-chain, interpretable, and subject to the discretion of a board and the judgment of a court in a scenario that has not occurred yet.
I spent two weeks in 2021 reverse-engineering the Bored Ape metadata retrieval process and found the same gap in a different costume: an asset that presented as on-chain scarcity depending on an off-chain server for the attributes that gave it value. The scarcity was real. The properties that made the scarcity legible were not.
Same structure here. The par anchor is real. The mechanism that makes the par anchor credible โ the redemption terms, the dividend enforcement, the treatment of deferred payments, the conversion mechanics on the Strike series โ lives in a document almost nobody holding the instrument has read.
That is not a claim of fraud. It is a claim of opacity, and opacity is a risk factor that does not appear in any yield calculation.
6. Three dead gauges and one directional one
Pull the week together.
No bitcoin bought. No bitcoin sold.
No ATM issuance.
No common repurchase.
One engine running: $176.3 million of cash converted into the retirement of $181.1 million of the company's own preferred claims.
Strategy spent the week buying bitcoin exposure from its own liability holders instead of from the market.
That is a coherent strategy when a specific instrument is dislocated. It is a warning when it is the only thing the machine is doing.
7. Where the buyer base actually comes from
One more piece, and it connects to modeling work I did in 2024.
That year I built a Python tool to track real-time institutional flows into the spot bitcoin ETF complex. The thing that fell out of the data was not a crypto correlation. It was a tradfi correlation: institutional flows into crypto products behaved like institutional flows into any other risk asset, rotating on traditional volatility regimes.
The buyers of variable-rate preferred are the same population. Rate-sensitive, mandate-constrained, rotating between preferreds, closed-end funds, and short-duration credit on a spread basis.
Strategy has imported the term premium of the preferred market into the bitcoin treasury.
That is the actual innovation in this capital structure, and it is also the actual exposure. The STRC discount is not primarily a crypto event. It is a credit-spread event that happens to be denominated in a bitcoin company's balance sheet. When the spread widens, it widens for reasons that have nothing to do with chain data, and no amount of on-chain analysis will explain it.
Which is why the surface-level coverage โ "Strategy bought back $176 million of preferred" โ is not just incomplete. It is looking in the wrong market.
8. The capacity problem, borrowed from a different discipline
One analogy, because it is exact enough to be useful.
Rollup data availability works the same way as a dividend reserve. In quiet blocks, capacity looks infinite and cheap. Blob space absorbed the first wave of rollup demand and everyone concluded the scaling problem was solved. My position on that has not changed: post-Dencun blob data saturates within two years, and every rollup's gas cost doubles again when it does. Capacity that looks ample at the current burn rate is not capacity. It is a measurement of the current burn rate.
A reserve that covers eight years of dividends at today's obligations is a strong position. It is not a statement about what the obligations become if the reset rate falls, the discount widens, the buyback pace accelerates, and the reserve is the only thing standing between the instrument and an open market bid that does not exist.
Every capacity number is a snapshot. Snapshots are not guarantees.
Takeaway: Three Prints to Watch, and One Question
The next few weeks will settle most of this, and the filing gives you three specific numbers to track.
First, the implied price. When the next 8-K reports a repurchase, divide cash spent by shares retired. If the average comes in below $97.36, the discount is widening and Strategy is chasing it. If it comes in above $99, the peg is doing its job without help. That single number is a cleaner read on the structure than any of the narrative around it.
Second, the ATM line. The first week that shows shares sold is the week the premium returned. It will also be the week bitcoin accumulation resumes, because those two things are the same event. Until then, the machine is not compounding โ it is maintaining.
Third, the drawdown. $1.19 billion remaining, roughly 6.75 weeks at the observed pace. If the authorization is exhausted without a renewal, the market learns something it has not been told: whether STRC holds par when the issuer stops buying. If it is renewed before exhaustion, the market learns something else โ that the peg has become a standing operating expense.
And one question, which is the one I would put to the board if anyone asked.
The instrument is engineered to hold $100. This week the issuer paid $97.36 to make that true. The math works. The reserve covers it. The trade is accretive. Every component is defensible.
So at what price does it hold without him?
Coda: The Quiet Market Is the Signal
A technical strategist, in a podcast released alongside this news cycle, described current bitcoin conditions as a quiet market and framed that quiet as an opportunity.
The quiet is not weather. It is a mechanical consequence.
The marginal buyer of bitcoin spent 2025 being a corporate treasury operation financed by an equity premium and a preferred stack. This week that operation bought zero bitcoin, sold zero shares, and spent its cash purchasing its own liabilities at a discount to par.
When the largest structured bid in the market turns inward, the market goes quiet. The quiet is not the setup for the next move. It is the receipt from the last one.
Watch the ATM line. When Strategy starts selling shares again, the quiet breaks โ and everyone who mistook the quiet for a baseline will discover they were reading a symptom the whole time.
