There is an argument that has kept the same shape in crypto commentary for years. Debt rises, the interest bill rises with it, bonds stop feeling risk-free. Once enough managers of large portfolios see the link, allocation tips and the price jumps. Then comes the number: the threshold is 25 percent.

The number is real. It is in Science, it is cleanly measured. It measures the wrong kind of tip.

The 25 percent counts people in a game with no right answer. What moves large portfolios is not understanding. It is the answer to a different question: what happens to me if I am the only one still at zero?

The 25 percent comes from a game with no right answer

Centola, Becker, Brackbill and Baronchelli, Science 2018. Ten online groups of 18 to 30 people, 194 in total. They name a face and get paid for agreeing. Then the researchers seed a growing minority pushing a different name.

Below a threshold, the minority fails. Around a quarter, the group flips.

Two things vanish in the retellings.

Resolution. In the smallest group, 25 percent is five people. The paper's own band is 21 percent fail, 25 percent succeed. In one trial, one person made the difference.

Incentive. The only prize was coordination. No truth, no switching cost, no 0.4 percent sleeve, no career.

That is how a convention spreads. Not how an allocation spreads.

The literature gives a range, not a number

Everall and colleagues, Earth System Dynamics 2025, 59 classified tipping events: 95 percent of critical masses sit below 40 percent of the population, the possible range runs from 10 to 43 percent. The authors write of potential, not inevitability.

Otto and colleagues, PNAS 2020, technology adoption: 17 to 20 percent of market or population share. Chenoweth's 3.5 percent is non-violent mass movements, the mechanism is mobilisation plus elite defection, not the diffusion of an idea.

Tipping is a pattern. The height is context. One number is one paper.

The pattern carries to capital. The number does not.

Three breaks.

The unit. Every threshold above is a share of people. Prices respond to dollars. JPMorgan's 2026 survey covers 333 family offices in 30 countries with very unequal wealth. Eleven percent of heads can be one percent of the capital, or thirty. One sovereign fund at a one percent weight moves more than a quarter of all family offices at the same weight.

The shape of the decision. The naming game is A or B. Allocation is 0, 0.4, 2, 5 percent. A two-state model says nothing about the step from 0.4 to 2, and that is the step that matters.

The truth condition. A convention has no right answer. An allocation gets marked afterwards, and being early and right is rewarded, not punished.

The pattern survives the transfer. The 25 does not.

The pressure is real. The switch it is not.

The Congressional Budget Office reports $963 billion in net interest from October 2025 through July 2026, about $3.18 billion a day. The ten-year Treasury yield stood at 4.79 percent on 1 September 2026, the thirty-year at 5.27. Japan's ten-year crossed three percent the same day for the first time since 1996, while Japan holds $1.14 trillion of Treasuries as the largest foreign holder.

US debt: $36 trillion in November 2024, $38 trillion in October 2025, $39 trillion in March 2026, $39.9 trillion on 7 August 2026. Quoting one of those without a date is quoting a random draw from two years.

That can raise the cost of staying at zero. It does not make 25 percent of managers the switch.

The surveys count heads, not capital

JPMorgan, fieldwork May to July 2025:

Bitwise and VettaFi: 32 percent of advisors had crypto in client accounts in 2025, up from 22. BNY, with the opposite headline: 74 percent of ultra-high-net-worth offices are investing in or exploring crypto, a 21 percent increase on the year in BNY's words.

"Has exposure" and "looking at it" are different questions. Both sides quote the one that fits.

Headcount can rise while capital barely moves. That is adoption as survey, not allocation as weight.

Holdings need a date too. US spot ETFs held 1.245 million Bitcoin on 2 September 2026, Strategy alone 843,775 in July 2026. Without the date, you are comparing snapshots from different months.

The threshold is not epistemic. It is reputational.

Coordination games describe asset managers very well. Just not at the point of understanding.

Scharfstein and Stein, 1990: being wrong alone costs your reputation. Being wrong with everyone does not. That is the experiment's logic. Insight is optional.

As long as "I am the only one at zero" has no cost, zero is the safe choice, and understanding changes nothing. Once that question shows up in committee, zero is the risky choice. Then allocation can move fast without anyone changing their mind.

Price is the downstream print of that capital. It is not the tipping variable.

Defaults tip. Convictions do not.

Hold Bitcoin because it sits in a model portfolio and you never made a Bitcoin call. You have nothing to defend.

BlackRock, February 2025: one to two percent Bitcoin in two model portfolios, out of a universe of about $150 billion. June 2026: the same range formally called reasonable for multi-asset portfolios. 30 March 2026: the US Department of Labor proposes a safe harbour for fiduciaries adding alternatives to 401(k) menus, and itself expects the channel to run mostly through target-date funds, as small sleeves.

That is how an allocation threshold falls without a conviction moving.

The strongest objection

"Adoption is rising. 32 percent against 22 is real movement."

It is. It counts advisors holding anything, not how much. The typical recommendation sits in low single digits, the average family office weight at 0.4 percent. Breadth without depth is exactly what a capital-weighted threshold does not cross.

The reverse cut: "89 percent is a 2025 snapshot, surveys lag." Possible. Then they are equally useless as proof that a tip is near. You do not get to treat them as leading when they help and stale when they do not.

What to watch

Surveys about understanding count heads and opinions. The wrong meter for a price story, and the wrong meter for this one.

Watch the share of the Bitcoin stock sitting in decoupled channels: model portfolios, target-date funds, default menus. Places where nobody made a Bitcoin decision and nobody has to defend one.

The open question, before the next 25 percent slide: which allocation threshold, fixed in advance rather than found afterwards, was ever crossed without the default changing first?

FAQ

Is the 25 percent study bad? No. It answers its own question convincingly. It measures the spread of a convention in groups of 18 to 30, not the spread of an insight among capital allocators. The problem arises in the quoting.

Why does the difference between people and dollars matter so much? Because prices respond to capital. With very unequal wealth, a headcount share says almost nothing about the capital share. One percent of allocators can move more money than half of them.

Does the 3.5 percent rule apply here too? No. It comes from research on non-violent mass movements and rests on mobilisation and elite defection. That is a different mechanism from the spread of an investment opinion.

Do these numbers say Bitcoin has failed with institutions? No. They say breadth and depth have come apart. More and more addresses hold something, and the weights stay small. Both observations are true at once and get quoted separately depending on interest.

Why is there no single figure for the national debt? Because the mark sits anywhere between $36 trillion (November 2024) and $39.9 trillion (August 2026) depending on the date. What is solidly citable is the CBO interest figure: $963 billion over ten months.

What would refute the reading in this piece? A clear jump in adoption without the share of decoupled holdings having risen first. Then the threshold would be epistemic after all, and not reputational.