Two Tuesdays ago I did what I always do when a headline number lands in my feed: I rebuilt the arithmetic behind it before I read a single interpretation of it. The DeFi Education Fund, working with Ipsos, had published a survey of 1,545 American adults on financial technology and crypto. The number that traveled fastest was 13% — the share who said they support blockchain, crypto, or DeFi as a policy matter.
The number that stopped me was not 13%. It was the one sitting two columns away in the same table: 74%. Three-quarters of those same respondents said the financial system needs to change, and that they want more control over their own money. And 5% said they highly trust decentralized platforms.
Seventy-four, five, thirteen. Three figures from a single questionnaire, and the distance between the first and the second is sixty-nine percentage points. That gap is the most important data point in the entire release, and almost nobody quoted it. Instead the discourse split into the usual two camps — one insisting that 13% proves the industry has failed, the other insisting that 13% is early and therefore fine. Both camps are reading the wrong number. Both are also skipping the part where you check whether the number is even stable before you build an argument on top of it.
I have spent eleven years watching this industry confuse the thing it builds with the thing people need. The story isn't in the token, it's in the trust. This survey, whatever its flaws, is one of the cleanest instruments we have ever been handed for measuring the distance between those two things.
To understand why a document like this matters, you have to understand what it is for. The DeFi Education Fund is an advocacy organization, the kind of American 501(c)(4)-style entity that exists to move policy, not to produce neutral knowledge. Ipsos is one of the three largest market research groups in the world, listed in Paris, with a sampling methodology mature enough that its name functions as a credibility anchor. The pairing is not a scandal. It is simply how policy shops operate: you rent methodological legitimacy from a firm that has it, so that your numbers can survive a congressional staffer's skepticism.
The important consequence is that this is commissioned research. The sponsor has meaningful influence over question order, wording, and option framing. That does not make the data false. It makes the data directional rather than precise, and it means the framing deserves as much scrutiny as the findings.
What makes this release genuinely unusual is that it measures something the industry almost never measures: how people who are not in the room perceive us. Every adoption narrative of the past decade was validated internally. In 2017 we counted wallets created. In 2021 we counted NFT mints. In 2024 we counted ETF inflows. In 2026 we count the transaction volume of autonomous agents. All of these are inside metrics — they describe what the converted are doing. None of them describe what the unconverted believe. This survey is an outside metric, and outside metrics are the scarcest asset in crypto research.
I learned that lesson the hard way in 2020. I was finishing a cybersecurity degree in Vienna and moderating the Discord for Ampleforth, an elastic supply protocol with more than five thousand daily actives. The rebasing mechanism was elegant. It was also, to a normal person encountering it for the first time, indistinguishable from magic that stole your balance overnight. I started turning the supply mechanics into visual guides — what happened, why the number in your wallet moved, why nobody had taken anything from you. Support tickets fell by roughly 40%. Nothing about the protocol changed. What changed was that someone named the fear out loud before explaining the math.
That experience shaped how I read every dataset since. Which brings me back to the arithmetic, because the arithmetic is where this release is most vulnerable and most interesting.
Before you interpret a number, you check whether it can carry the weight. At n=1,545 and p=0.5, the margin of error at 95% confidence is approximately ±2.5 percentage points. Hold that against the headline growth statistic: prior-year holders at 18%, current holders at 22%. That is a four-point move — roughly 1.6 times the margin of error. On a full-sample basis it sits at the edge of statistical significance. If it is a post-weighted subsample estimate, it very plausibly is not significant at all.
The release describes this as having "risen." That is a hardening verb applied to a soft signal, and it is the most common rhetorical move in industry-funded research. I have done it myself in reports I am not proud of.
There is a second arithmetic oddity worth flagging. The number 13% appears twice in this release with two completely different meanings: once as the share who support crypto as a policy matter, once as the share who used crypto in the past year. Those are not the same construct, they are not the same question, and juxtaposing them in a short press document invites conflation. It could be coincidence. It could also indicate that support and usage were placed in the same question matrix, in which case respondents may have anchored one answer against the other.
And there is a provenance problem. The survey fielding window is given as September 8–16, with a release date of October 28, and the comparison baseline is drawn from the prior year. The timeline is internally self-consistent. It is not, however, something I was able to cross-verify against the primary PDF before writing this, and the 2025 baseline source, sample, and questionnaire are not disclosed in the release. If the 18% figure came from a different instrument with different wording and different weighting, then 18% to 22% is not a time series. It is two unrelated snapshots placed side by side. That is the single largest methodological question in the document, and it determines whether the growth story exists at all.
So I put a discount on every number before I interpreted any of them. What survives the discount is the shape.
The eight-point inversion is the finding the industry should be losing sleep over. Thirteen percent used crypto in the past year. Five percent highly trust decentralized platforms. Usage exceeds trust by eight points. That is a structurally unstable configuration, and it contradicts a foundational assumption in most DeFi models: that TVL represents sticky capital.
Capital that does not trust its rails behaves like a tourist, not a resident. It arrives when yields are attractive and leaves at the first sign of protocol risk, governance drama, or regulatory noise. You cannot build a lending market on tourists. You can build a casino, and the industry has, repeatedly, but a casino has different capital dynamics than a bank.
I watched this exact pattern up close during the Ampleforth period. Users were actively using the protocol while emotionally bracing for it to hurt them. The technical performance was fine. The affective position was defensive, and defensive users do not compound. When I finally started naming that defensiveness in our community updates rather than arguing against it, retention improved — not because the product got better, but because people stopped feeling gaslit by optimistic announcements.
Sunshine and liquidity treat trust deficits the same way. They price them, quietly, every single day, and the price is called yield compression.
The four-point inversion is subtler and it has a name I have started using: use without understanding. Nine percent report high familiarity with crypto. Thirteen percent used it in the past year. Usage exceeds understanding by four points, which implies a meaningful population that touches crypto without thinking of itself as touching crypto — and without consenting to be counted in the industry's adoption metrics.
These are the people who bought a spot Bitcoin ETF inside a brokerage account and still call it their broker position. They are the people who hold a token inside Robinhood and never opened the withdrawal screen. They are the people who sent a USDC transfer through PayPal and experienced it as a payments upgrade, not as a cryptographic settlement layer.
This cuts in two directions at once. It means 13% overstates the industry's reach, because passive holding gets counted as usage. It also means 13% understates the industry's actual footprint, because embedded scenarios are invisible to survey instruments that ask whether you have "used cryptocurrency." The real number is probably neither higher nor lower but differently shaped than the headline suggests.
The strategic implication, though, is unambiguous. When only 9% of the adult population has real familiarity, technical differentiation is invisible at the point of contact. I spent part of last quarter auditing a prototype built on Uniswap V4 hooks. The developer surface is genuinely powerful — you can encode fee logic, oracle behavior, and liquidity conditions directly into the pool lifecycle. It is also genuinely terrifying, in the sense that a hook is an unaudited extension point sitting in front of user funds.
But here is the thing that matters for a survey like this. A hook is not something a 74%-of-people user can feel. Neither is a modular rollup proof system. Neither is restaking. We built an extraordinary amount of programmable surface area over the past three years, and the population that can perceive the difference between our architectures remains in the single digits. That is not a marketing failure that better marketing can fix. That is a translation failure built into the product vocabulary.
The twenty-two-point label discount is the most actionable number in the document, and it is buried. Thirty-five percent of respondents support policy positions framed as financial technology. Thirteen percent support the same general territory framed as blockchain, crypto, or DeFi. The ratio is 37% — crypto support runs at roughly one-third the level of fintech support, a 2.7x gap.
That is almost certainly a framing effect rather than a rejection of substance. In the public mind, "fintech" is a better banking app, and a better banking app is unambiguously good. "Crypto" carries FTX, Celsius, Terra, and money laundering, in that order, and the association fires before any evaluation happens. Same person, same day, different word on the page, twenty-two-point swing.
The competitive read here is uncomfortable. Our real competitor is not another chain. It is the CeFi-adjacent product the public already files under fintech, plus traditional finance's own digital improvements. In that arena, the industry's most cherished vocabulary — decentralized, permissionless, censorship-resistant — is not a selling point. In public context, those words function as comprehension cost. Every one of them is a tax on attention, and the user is paying with the thing we have the least of.
The cheapest growth lever available to this industry is not a new mechanism design. It is renaming. Nobody wants to run that meeting, because renaming feels like surrender to the people who got into this space for ideological reasons. But the data does not care about how the meeting feels.
The 40.9% dormancy rate is the number that should terrify operators, and it is the number almost nobody is running against. Twenty-two percent report having held crypto at some point. Thirteen percent used it in the past year. That is nine points of attrition — 40.9% of everyone who ever held, now gone.
Low penetration is a time problem. Early technologies look like this. You wait, the compounding happens, the curve bends. High dormancy is a habit problem. A user who bought and stopped did not fail to understand the product. They tried it, formed a judgment, and left. Their cost to reacquire is far higher than a first-time user's cost to acquire, because you are not fighting ignorance, you are fighting a conclusion.
In the winter of 2022, after Terra, I felt my own version of that conclusion forming. ESFJ wiring does not do well with extended isolation, and I was conducting research in an ecosystem that had just collectively been humiliated. So I organized a weekly support circle in Vienna — ten small-group sessions for junior analysts, maybe fifty people in orbit. What kept those people in the industry was never the yield math. It was the fact that someone else admitted, out loud, that they felt naive. Shame was the exit vector. Naming it was the retention mechanism.
Narratives precede utility — but trust precedes narrative. That is the emotional signature of a dormant holder, and it is the segment with the highest reactivation ROI in the entire market. Almost no budget targets them. The entire acquisition apparatus is pointed at people who have never touched crypto, which is the most expensive audience in existence, while nine points of warm, already-educated, already-experienced users sit dormant and unaddressed.
The on-chain triangulation confirms what the survey is saying, which is why I take the shape seriously even while discounting the digits. Across the last eighteen months the aggregate picture has been consistent: total value locked rising, stablecoin supply rising, ETF net inflows at records, and active address counts flat to mildly declining.
Those two directions cannot both be describing user growth. They describe capital density growth. Existing participants are adding size; new participants are not arriving. That divergence is the on-chain shadow of a thirteen percent that refuses to move, and it is the strongest available corroboration that the survey is directionally right even if its exact figures are fragile.
My Layer 2 view sits uncomfortably inside this. There are dozens of rollups now, each with its own bridge, its own liquidity incentives, its own sequencer, and they are all fighting over the same finite set of users. That is not scaling. That is slicing already-scarce liquidity into ever-thinner fragments and calling the fragments progress. The 13% flat line is the demand-side evidence for a supply-side fragmentation problem — and no amount of additional throughput fixes it, because throughput was never the binding constraint.
We spent five years optimizing the constraint that wasn't binding. Gas went to fractions of a cent. Wallet onboarding improved dramatically. And usage did not respond, because friction was never the reason a person who does not care about decentralized finance was not using decentralized finance.
The same logic quietly condemns another vertical I have written about before. Dynamic NFTs and programmable royalties are technically interesting and commercially irrelevant to the artist who needs stable buyers rather than a more sophisticated stack. A retail-emotion-dependent category facing flat retail penetration is not facing a design challenge. It is facing a demand ceiling, and the ceiling is the point.
Following the transmission chain tells you who feels this first. Centralized exchanges are most exposed in the medium term, because retail trading volume is still the revenue base and a static 13% means the incremental user simply never arrives. The pivot to institutions, derivatives, listing fees, and market making is not opportunistic. It is forced. DeFi protocols face yield compression for the same reason: the same limited capital chasing the same returns, with no new liquidity entering the pool.
Layer 2s face the fragmentation tax more harshly than the base layer, because they compete on incentives that must be continuously paid, and incentive-funded TVL has the shortest half-life of any capital in the market. NFT and gaming verticals, which depend most heavily on retail sentiment and new entrants, face the clearest ceiling.
Stablecoins are the exception, and the exception is structural. Payment and remittance products are the only crypto category that can wear the fintech label without paying the twenty-two-point discount, because the public already categorizes a dollar transfer as banking-adjacent convenience. Nothing about that requires the user to understand a private key or a consensus mechanism. The shortest path from 5% trust to mainstream adoption runs through the words "send money," not through the words "decentralized finance."
And there is an underappreciated beneficiary: B2B infrastructure. When 9% of the public has real familiarity, intermediaries remain permanently necessary. Banks, brokers, and payment companies become the required translation layer, and their infrastructure procurement is a durable demand signal that does not depend on consumer education succeeding. Selling shovels to institutions has a materially higher and more predictable return than buying attention from a public that is not currently paying any.
Here is where I part company with both readings of this release. The lazy interpretation is that 13% proves the industry failed and the public rejected it. The contrarian interpretation, and the one I think survives contact with the full dataset, is that this is a distribution failure sitting on top of validated demand.
Seventy-four percent of Americans say they want more control over their money. That is not a weak signal. That is the exact psychological precondition for self-custody, sitting there, unprompted, in a commissioned survey. And 5% trust the platforms that offer it. A sixty-nine-point gap between wanting the outcome and trusting the provider is not a demand problem. It is a bridge problem. Nobody has built the bridge in a language people recognize as theirs.
My second disagreement is with my own side's reflex. The immediate industry response to data like this is "we need more education." I do not believe that, and the survey gives me a reason. The dominant affect in these numbers is not hostility. It is apathy. Thirty-five percent support fintech, thirteen percent support crypto, and the remaining majority is not angry about crypto — it simply does not consider crypto relevant to its life. Hostility means the topic has weight. Apathy means the topic does not exist.
Education does not cure irrelevance. Relevance does. For a decade, every marketing dollar this industry spent answered the question "why are we better?" and almost none answered the question "why should you care?" Those are different questions, and only one of them is being asked by the person on the other side of the screen.
My third disagreement is with the data itself, and I want to be explicit rather than diplomatic about it. I do not fully trust this release. The margin of error swallows part of the growth story, the baseline provenance is undocumented, the key terms are undefined, the sponsor is an advocacy organization with an interest in a particular narrative, and the favorable metric was disclosed while the unfavorable time series was not. Low evidence strength, high transmission strength — that is the exact failure mode of our current information environment, and it is precisely what will happen to this survey once it becomes a headline.
And yet the shape holds. Dormancy at roughly four in ten ever-holders, a use-without-trust inversion, a use-without-understanding inversion, and a twenty-two-point label discount — those four structural facts do not depend on whether the true penetration rate is 11% or 15%.
The window for the next move opens after the American stablecoin framework clarifies, and I would put it at six to eighteen months past that point. Two products will define that window. The first is a consumer-grade self-custody wallet with real fiat rails that explains itself in the language of safety rather than sovereignty — "your money cannot be frozen, your platform cannot lend it out." The second is the institutional translation layer that lets a bank answer a customer's custody question without teaching them about key derivation.
Neither of those requires a breakthrough in cryptography. Both require something harder: a sentence. The industry is not waiting on technology. It is waiting on a sentence that a seventy-four-percent person hears and recognizes as being about them.
Until someone writes that sentence, we will keep publishing surveys that measure the same thirteen percent from slightly different angles, and we will keep mistaking the measurement for the progress.