On October 16, a single number entered Japan's disclosure chain and then the global crypto press: 11.11%. Capital Research, the filing said, now held 143,301,084 Metaplanet shares, up from 125,455,820 on September 8, when the same holder sat at 9.32%. Every headline that followed read identically. Institutional conviction. Smart money rotating into the largest Bitcoin treasury company listed in Tokyo. The narrative of 2025 has a shape now โ capital leaving ETF wrappers for equity vehicles that can lever, borrow, and buy โ and this looked like its proof point.
I read it three times, then I did the division.
143,301,084 divided by 0.1111 gives you a voting-right denominator of roughly 1.2898 billion shares. 125,455,820 divided by 0.0932 gives roughly 1.3461 billion. The denominator fell about 56.25 million shares over five weeks, a 4.18% contraction. But Metaplanet repurchased 73,981,400 shares in that same window โ all of them at ยฅ261, ยฅ19.3 billion of capital, executed as a single block on the Tokyo Stock Exchange's ToSTNeT-3 off-auction system.
Seventy-four million shares left the count. The count only fell by fifty-six. That gap โ roughly 17.7 million shares โ is the most interesting thing in this story, and almost nobody wrote about it. Alpha hidden in the noise.
You need to understand what kind of animal Metaplanet is before the arithmetic means anything. It is not a protocol. There is no consensus layer, no sequencer, no token contract to decompile. It is a listed Japanese operating company that decided, in 2024, to become a Bitcoin accumulation vehicle โ the Strategy playbook, transplanted to the Tokyo exchange. Its balance sheet is the product. Its equity is the funding instrument. Its KPI is not revenue in any traditional sense but BTC per share.
That distinction changes how you audit it. When I audited ICO whitepapers in a Bangkok Telegram group in 2017, I went through fifteen projects and flagged eight, and the tells were never in the marketing copy. They were in the supply table โ the unlock cliff, the team allocation, the difference between total supply and circulating supply that the deck blurred on purpose. A treasury company inverts that problem but keeps the same math. There is no fixed supply here. Shares can be issued at will, and the entire value proposition is that management issues them only when doing so is accretive to the per-share Bitcoin backing. The supply schedule is discretionary. The adversary is now a board of directors rather than a vesting contract.
The mechanism is usually called the flywheel. When the stock trades above net asset value โ mNAV greater than 1 โ the rational move is to sell equity, convert the proceeds into BTC, and raise the Bitcoin backing per share for everyone. When the stock trades at or below NAV, the flywheel reverses: issuing shares now destroys per-share value, and the only rational capital action is to retire them. Metaplanet has said, in its own disclosure language, that it ties buyback policy to capital efficiency and to Bitcoin held per share. That is the company telling you which regime it thinks it is in.
The tool matters too, and it is where most coverage goes wrong. ToSTNeT-3 is a mature off-auction matching mechanism on the Tokyo exchange. It is built for large single-shot crossings โ you negotiate a block, you print it, you are done. It is not an on-market buyback program. It does not put a continuous bid under the stock. It does not defend a price level for weeks. It restructures a cap table in one afternoon. Confusing the two is the most common analytical error I see in treasury-company writing, and it is the error at the heart of this one. A single crossing tells you about capital structure. A continuous program tells you about price support. They are not the same signal.
Now the arithmetic, laid out cleanly, because the whole story lives in the decomposition.
Between September 8 and October 16, Capital Research's stake moved from 9.32% to 11.11% โ a gain of 1.79 percentage points. The holder bought 17,845,264 net new shares. Using the October 16 denominator of 1,289,838,727, that buying alone accounts for about 1.38 percentage points. The remaining 0.41 points came from somewhere else entirely: the denominator shrank, so the same holder's existing shares became a larger fraction of a smaller whole.
Roughly 77% of the headline increase was buying. Roughly 23% of it was a buyback. Put differently, a quarter of what the press called accumulation was actually subtraction โ subtraction performed by the company, not by the investor. Code doesn't lie, but narratives do.
This is the methodological point I want tattooed on every reader who trades treasury stocks. A percentage is a ratio. You cannot read a ratio and infer its numerator. When a company retires shares into treasury, those shares lose voting rights, and the denominator collapses. Every remaining holder's percentage rises automatically, with zero action taken. The reverse is equally true and equally misunderstood: issue shares, and every holder's percentage falls without anyone selling a single unit. I watched this exact confusion hollow out retail portfolios in the 2020 DeFi summer, when "TVL up 40%" headlines were driven by token emissions rather than deposits. Same disease, different asset class. The number that hurts you is rarely the one that is wrong. It is the one that is right but measured on the wrong denominator.
Let me be precise about the accounting, because that is where the sleight of hand lives. When a Japanese listed company repurchases its own shares, those shares move into treasury stock. They remain issued โ they still exist on the register โ but they carry no voting rights and no dividend claim. For the purposes of a voting-right denominator, they vanish. For the purposes of a share-count headline, they often do not, because the two numbers are reported separately and cited selectively. The buyback did not destroy 73.98 million shares of economic claim; it parked them. Management can cancel them, resell them, or hand them to employees. Until cancellation, the "reduction" is provisional. Anyone modeling permanent per-share accretion from a treasury-stock repurchase is modeling a decision management has not yet made.
Now the hole. The buyback removed 73.98 million shares. The denominator fell only about 56.25 million. The residual is roughly 17.7 million shares that had to come from somewhere โ and the only somewhere is net new issuance. Even if you generously allow for rounding error in the two disclosed percentages, each carried to two decimals, so maybe plus or minus 1.3 million shares combined, you cannot close a 17.7 million share gap. Something was issued while something was being retired.
The most plausible instrument is a moving-strike warrant, the kind of facility that lets a company drip shares into the market as the price rises. If that is what happened, the implication is uncomfortable for the bull case: Metaplanet was buying and selling its own equity simultaneously, and the net share count fell far less than the buyback announcement implies. The per-share Bitcoin improvement from the repurchase is real, but it is smaller than the gross figure suggests, because the gross figure ignores the offsetting issuance. Gross numbers are marketing. Net numbers are accounting. Only one of them pays.
I want to be careful here. This is an inference from two disclosed percentages, not a confession from the company. I have not seen the monthly share-count filing, and neither, apparently, has most of the coverage. But this is precisely the check I learned to run in 2022, when I spent six months certifying Thai fintech professionals on AML and disclosure chains after Terra collapsed. The lesson from that period was brutal and simple: a disclosure chain is not an audit. The company explicitly stated it did not independently verify beneficial ownership. The 11.11% figure originated with the investor's change report, flowed to the company, flowed to the media, and arrived in your feed looking like a verified fact. Trust is the new currency, and this particular currency was minted by a third party who had every incentive to make the number look decisive.
There is also a date that should stop you cold. The disclosed buyback authorization period runs from October 14, 2026 to October 13, 2027. Read that again. If accurate, the authorization was not even in force when the announcement was published โ the start date sits a year in the future. That contradicts the standard practice of executing within twelve months of a board resolution. The benign explanation is a typo: the period should read 2025 to 2026. The less benign explanation is that the article was actually published in October 2026, in which case every timestamp in this analysis shifts forward a year and the snapshot is stale. I flag it rather than resolve it, because resolving it requires the primary source โ Metaplanet's own IR page โ and any analyst who quotes the secondary report without checking that page is doing astrology, not auditing.
Then there is the KPI itself. The company's stated metric is Bitcoin per share, not total Bitcoin. That is the correct metric, and it is also the metric most easily gamed, because it has a denominator. Total BTC is a fact. BTC per share is a ratio, and ratios can be improved by shrinking the denominator without acquiring a single satoshi. A buyback that retires 5% of the float raises BTC per share by roughly 5.3% with zero new Bitcoin purchased. That is not fraud โ it is arithmetic โ but it means the KPI rewards exactly the operation we are examining, and it means a rising BTC-per-share print is not evidence of accumulation. It is evidence of a capital structure decision. The market routinely conflates the two, and that conflation is where retail money goes to die.
The Strategy comparison is instructive and usually drawn wrong. MSTR's model runs on at-the-money equity issuance and convertible debt โ instruments that let it sell volatility and cheap duration to buy a hard asset. Metaplanet's disclosed toolkit is the mirror image: issuance plus buyback. Both are legitimate. But they imply opposite views of the premium. One monetizes a rich multiple; the other repairs a cheap one. When you see a treasury company doing both in the same quarter โ and the 17.7 million share gap suggests exactly that โ you are looking at a management team hedging its own narrative, buying back when it can and issuing when it must. That is not a scandal. It is a firm admitting it does not control the premium.
One more number worth pinning. The buyback cleared at ยฅ261. ToSTNeT-3 crossings typically reference the prevailing market price, so ยฅ261 is a reasonable proxy for where the stock traded. Multiply by the implied denominator of 1.2898 billion and you get a market capitalization of roughly ยฅ336.6 billion โ call it $2.24 billion at ยฅ150 to the dollar. That is your anchor. But here is what I cannot compute, and neither can you from this material: the company's actual BTC holdings and its debt balance are absent. Without those, mNAV is uncomputable. Bitcoin per share is uncomputable. The entire valuation framework that makes a treasury company legible to an analyst is missing. A market cap without a NAV is a number without a denominator โ it tells you the price of a claim, not the value of what is claimed.
What I can compute is the overhang. The total authorization is up to 200 million shares or ยฅ60 billion. Roughly ยฅ19.3 billion has been used. That leaves about ยฅ40.7 billion โ enough, at ยฅ261, to retire another 156 million shares, roughly 12% of the current count. If management executes that, every remaining holder's percentage rises by double digits without anyone buying anything. That is the single largest future source of passive percentage increases in this cap table, and it has nothing to do with Capital Research's conviction. The noise will look like signal again, and it will look even more convincing the second time.
Run the logic in reverse and the trap closes. Suppose Metaplanet executes the remaining ยฅ40.7 billion of authorization. The count drops another 12%. Capital Research's percentage climbs toward 12.6% with zero purchases. The headline writes itself: institutional stake rises again. Nobody will mention that the investor did nothing. Now suppose the opposite โ the company issues shares to fund a large BTC purchase, the accretive move in a premium regime. The count rises, the percentage falls, and the headline reads institutional stake diluted. Also nobody did anything. In both cases the signal is manufactured entirely by the denominator, and in both cases the press will attribute it to the investor. This is not hypothetical. This is the mechanical future of every treasury-company cap table on earth, and the market has no vocabulary for it yet.
Here is the verification checklist I would run before trusting any of this, and it takes about twenty minutes. Pull Metaplanet's monthly share-count and treasury-stock disclosure from IR, not from the press release โ the gap between issued and outstanding is where the 17.7 million lives. Reconcile the buyback authorization dates against the board resolution, because the 2026 start is either a typo or a red flag. Find the BTC holdings and the debt balance, because without them mNAV is a guess wearing a spreadsheet. Check whether the repurchased shares were cancelled or merely parked, because parked shares can come back. Read the moving-strike warrant terms if one exists, since that is the mechanism that would explain simultaneous issuance. None of these steps are exotic. All of them are absent from the coverage.
Here is where I part company with the consensus reading.
The reflexive interpretation of a buyback is bullish. Management is confident, the stock is cheap, capital is being returned. I think that reading is backwards for a treasury company, and the evidence is in the tool selection. A company that believed its equity was undervalued relative to its Bitcoin would not be retiring shares. It would be issuing them โ selling expensive paper to buy cheap BTC, the exact flywheel that made Strategy's model work. The decision to buy back is a decision that the premium is gone. It is management implicitly conceding that mNAV has compressed toward or below 1, which is precisely when the flywheel inverts and retirement becomes the only accretive move.
That is not bearish in a crude sense. It is rational capital allocation, and I respect it. But it is not the same thing as conviction, and the market priced it as conviction. There is a difference between a board that is buying because it sees value and a board that is buying because it has run out of accretive ways to sell. From the outside they look identical. From the cap table they look nothing alike.
The second contrarian point is about the disclosure itself. Everyone treated 11.11% as a signal about the investor. The more informative signal is that the company did not verify it. In a regime where treasury companies are the primary on-ramp for institutional Bitcoin exposure, the integrity of the disclosure chain becomes a systemic variable, not a footnote. I have lost real money โ 15% on impermanent loss in 2020 because I trusted a yield number without modeling the pool โ and the lesson generalized: the number that hurts you is never the one that is wrong. It is the one that is right but measured on the wrong denominator. The headline measured a numerator and reported it as a position. That is the blind spot, and it is the same blind spot that has separated retail from institutions in every cycle since 2017.
Watch the monthly share-count filings, not the percentage headlines. The next time a stake in a treasury company rises, run the denominator first โ if it shrank, subtract the buyback before you credit the buyer. And keep your eye on the ยฅ40.7 billion still authorized, because the quietest way for a percentage to climb is for the company to make the pie smaller while you are watching the slice.
The real question for 2026 is not whether Metaplanet can buy more Bitcoin. It is whether the flywheel still turns when the premium is gone โ and whether a disclosure chain that nobody audits can survive being trusted at face value.