The 72-Point Gap: What Fireblocks' 2026 Infrastructure Survey Actually Measures

Two Numbers, Same Document

88%. 16%. Both figures come from the same PDF, published by the same vendor, in the same quarter.

The first number is on the cover. The second is buried in the body. The first says financial institutions have committed — or expect to commit — budget to digital asset infrastructure. The second says only a sliver of them have moved that infrastructure into production.

Read them separately and you have a press release. Read them together and you have a funnel. The funnel is where the actual information lives, and the 72-point variance between stated intent and operational deployment is the largest single gap I have recorded in an institutional adoption survey this cycle.

I opened the Fireblocks 2026 report expecting a data table. I got a headline, a definition problem, and a methodology that fits on a napkin.

That is the anomaly worth writing about. Not the fact that institutions are interested in digital assets — that has been settled since 2020. The anomaly is that a company selling the exact category it is surveying published its own adoption curve, defined neither the sample nor the word "production," and then let the friendly number travel.

"Too good to be true" is usually a rounding error. This time it is an editorial decision.

Context: Who Is Publishing, and Why It Matters

Fireblocks is not an anonymous Telegram account. Founded in 2018, it builds multi-party computation (MPC) custody infrastructure and a transfer network connecting institutions to exchanges, custodians, and blockchains. Its client base runs from mid-tier funds to large banks. By any reasonable market map, it sits in the top tier of institutional digital asset custody alongside Coinbase Custody and BitGo, with Anchorage filling out the tier.

That pedigree is exactly why the survey needs to be read with a specific lens. This is vendor research. The defining property of vendor research is not that the data is fabricated. It is that the data is selected. The publisher controls the sample, writes the questions, defines the terminology, decides the release window, and chooses which figure goes on the cover. Every one of those levers is available to optimize toward a conclusion.

The conclusion here is unambiguous: budgets are rising, institutions are moving, and the category — Fireblocks' category — is expanding. I have no reason to doubt the direction of travel. I have substantial reason to doubt the magnitude, and the magnitude is the entire commercial purpose of the document.

What the survey actually measures is procurement intent across a self-selected respondent pool. What it does not measure is the thing every downstream reader will assume it measures: capital deployed, assets custodied, transactions settled, revenue booked. Those are operating metrics. Intent is a sentiment reading.

The distance between a sentiment reading and an operating metric is where I do my work. It is also where this article lives.

One more framing note before the math. The category being surveyed — digital asset infrastructure — spans three distinct businesses that get bundled together in every vendor deck: custody, settlement, and tokenization middleware. Each has a different buyer, a different regulatory gate, and a different conversion timeline. A single blended percentage across all three is, structurally, a number that describes nothing precisely. It is an average of dissimilar things, which is the most common way to manufacture an impressive figure.

Core: Follow the State Transitions

When I audit a contract, I ignore the comments and trace the state machine. Same discipline applies here. Let me lay out the transitions a respondent would actually move through.

• State 0 — institution expresses budget intent: 88%
• State 1 — institution runs a proof-of-concept: not disclosed
• State 2 — institution reaches production: 16%

The most important row is missing. Without a proof-of-concept figure, the 16% is ambiguous in a way that produces two opposite headlines.

Scenario A: 90% of the pool ran a POC and 16% reached production. That is an 18% pilot-to-live conversion. Ordinary. Enterprise software converts somewhere in that range. Nothing to see.

Scenario B: 25% of the pool ran a POC and 16% reached production. That is a 64% conversion rate, which would be exceptional — it would mean institutions that touch the technology almost always ship it.

These are different worlds and the survey publishes one number that lives in both. I have reviewed enough deployment pipelines to know which scenario is likelier, and it is not the flattering one. In 2017, three weeks before LendingBot's mainnet launch, I found a reentrancy vulnerability in their withdrawal logic and submitted a patch. The code shipped clean. The organization did not. Key custody procedure, incident response, operational controls — all aspirational at the time. Ship-readiness is a stack, and the bottom of the stack is never the cryptography. It is policy, licensing, and audit.

That is why the 16% is the only hard figure in the document. It is the only number that survived contact with an auditor or a regulator.

Now the 88%, and why it is soft.

Budget intent is the weakest form of commitment in enterprise procurement. It is compiled before vendor selection, before the business case clears risk committee, frequently before anyone has confirmed which legal entity would hold the assets. A committed budget can be frozen, deferred, or quietly reallocated to a competing priority in the next fiscal quarter, with no public trace. The published figure merges two states — "committed" and "expect to commit" — that have materially different completion rates. Committed budget survives at least one internal review. Expected budget survives nothing yet.

If I were pricing the true probability that the 88% converts to deployed infrastructure, I would weight committed intent near 0.6 and expected intent near 0.3, then treat the 16% as the empirical anchor. The observed conversion is what it is. The intent number is a leading indicator with a long and noisy lag.

The Comparison That Actually Tests the Claim

I have run this same test before, in a cleaner dataset. In 2024, after the spot Bitcoin ETF approvals, I built a dashboard tracking daily net creations for BlackRock's IBIT and Fidelity's FBTC and correlated them against BTC price. On multiple days I found a decoupling: price rising on negative net flows. That told me the marginal buyer was not the institution. It was retail momentum masking institutional distribution.

The generalizable rule: when a survey and an operating metric disagree, the operating metric wins. The operating metric for this category is Assets Under Custody — AUC. Not one vendor in the survey disclosed it. Not Fireblocks either. A vendor publishing a demand survey while withholding the metric that would validate it is communicating something about the metric.

Let me be precise about what AUC would show if the 88% were converting. Coinbase Custody, BitGo, Anchorage, and Fireblocks would all be compounding custody balances. Where there is a listed parent, custody fee revenue in the institutional segment would show it. We have partial visibility into Coinbase's institutional line and effectively none into the private custodians. That absence is not proof of weakness. It is proof of unverifiability — which is the same defect the survey has.

What the Technology Layer Actually Is

The survey is implicitly a vote on a technology stack: MPC custody and tokenization middleware. MPC is not new. Threshold signature schemes — GG18, GG20, and the production ECDSA variants used in custody — have been deployed at scale since 2019. Fireblocks' differentiator is not cryptographic novelty. It is integration surface: an API network that lets an institution move an asset from a custodied wallet to an exchange, a dealer, or a chain without rebuilding the plumbing in-house.

That is a real and valuable product. It is not a paradigm shift. When a survey measures an established category, rising numbers mean adoption, not innovation. The cryptographic risk in MPC custody is understood and largely priced. The residual risks are operational and regulatory, and those are precisely the risks the 88% figure ignores.

There is also a philosophical split the survey flattens. MPC distributes key material across shards held by multiple parties and never reconstitutes the full private key; multi-signature custody, the Anchorage and classic BitGo approach, requires multiple independent keys to authorize a transaction. MPC optimizes for operational friction. Multi-sig optimizes for auditability and jurisdictional separation. These are different trust models, and a blended adoption percentage cannot distinguish which one an institution has actually deployed. The headline hides the architecture.

The Seven-Link Chain

Here is the transmission path from budget intent to on-chain impact. Count the links that can break.

1. Budget intent → approved business case. Requires internal risk sign-off. Breaks on policy ambiguity.
2. Business case → licensing. Requires a custody charter, trust registration, or equivalent. Breaks on jurisdiction.
3. Licensing → integration. Requires vendor selection and build capacity. Breaks on internal IT throughput.
4. Integration → production. Requires audit and operational controls. Breaks on the same gate that holds 84% of the pool.
5. Production → AUC. Requires clients to actually custody assets. Breaks on demand.
6. AUC → on-chain settlement. Breaks because institutional custody is largely omnibus and permissioned, and permissioned rails do not push volume into public DeFi.
7. Settlement → token demand. Only for the settlement assets with institutional depth: BTC, ETH, and stablecoins.

Seven links. Historically each carries a failure rate above zero. Apply a conservative 70% survival rate per link and the terminal probability of budget intent producing token demand lands under 8%. I am not claiming the true number is 8%. I am claiming the chain is long and that the survey reports on link one.

This is why the coverage of this report inverted the two figures. The 16% is a real reading on a real link. If it climbs to 30%, institutional infrastructure has crossed from pilot culture into operating culture. Until then, the 88% is a forecast, and forecasts are the cheapest instrument in finance.

What "Production" Might Mean

The definition of production is never stated, and the omission is load-bearing. Two readings are both defensible.

Narrow reading: production means live processing of customer assets at material volume across the full business line. Under this definition, 16% is a strong number, and the real adoption story is healthy.

Broad reading: production means a single business line has gone live, including small-value test transactions with real settlement. Under this definition, 16% is inflated, and the honest adoption figure is lower.

The document does not tell you which reading applies. In methodology terms, that is not a minor gap. It is the difference between a credible operating metric and a decoration. Any number that cannot be falsified is a narrative device, not a measurement.

The Centralization Nobody Audits

There is a structural point that vendor surveys never raise and that I raise constantly. Institutional custody is a permissioned, centralized trust model. The MPC provider manages shard allocation. The custodian holds administrative authority. Withdrawal policies, allowlists, and recovery procedures all sit with a single operator.

I have written for two years that Layer2 sequencers are effectively single centralized nodes and that "decentralized sequencing" has been a slide deck rather than a shipped system. Custody is the same architecture with better branding. The difference between a sequencer holding ordering power and a custodian holding key-shard authority is a matter of vertical, not of trust. Both are single points of operational failure. Both are disclosed in the security model and audited by nobody.

That does not make the model wrong. Institutional clients often require centralized accountability. But it does mean that "adoption" here is adoption of a permissioned rail, and a permissioned rail does not necessarily route capital onto public chains. The RWA bull case depends on that routing. The survey never tests it.

The LUNA Parallel

I have seen this exact narrative shape before. In May 2022, I published a LUNA forensics note roughly 48 hours ahead of the collapse. What I observed was not a surprised market. It was a crowded exit. Anchor Protocol deposits were draining through a specific cluster of wallets, and the 19.5% advertised yield was the bait. The mechanism was algorithmic. The failure was behavioral. Too many participants had internalized an unsustainable rate as their baseline.

The 88% is the same shape. It is not a yield, but it functions as a narrative yield — a number that pays a return in attention. Anyone who internalizes 88% as the base case for institutional adoption will misplan capex, hiring, and product roadmaps. Anyone who internalizes 16% will be approximately right.

There is a cultural reason the 88% travels and the 16% does not. Headline figures get posted. Funnel figures require arithmetic. The industry has a demonstrated preference for the former, and vendor IR teams price that preference into their release strategy. I have now watched this cycle three times: the survey drops, an aggregate account quotes the top line, price wiggles, no operating data follows, and the retraction is silent.

Contrarian: The 16% Is Not the Bad News

Everyone reading this report will treat the 16% as the disappointing number. I think that is backwards.

The 16% is the only figure that has been tested against reality. It reflects institutions that navigated licensing, integration, audit, and operational control — a gauntlet that has destroyed enterprise software rollouts in far friendlier regulatory climates. A one-in-six completion rate against that gauntlet, in an asset class with unresolved accounting treatment in several major jurisdictions, is not a failure. It is a floor.

The 88%, by contrast, has been tested against nothing. It is a survey answer. And survey answers in vendor research carry a specific, well-documented distortion: self-selection bias. The institutions most inclined to respond to a digital asset infrastructure survey are, by construction, the ones most inclined to say yes. The sample is not random. The sample is the answer, pre-committed.

Here is the counter-intuitive claim. The industry's real institutional adoption signal is not in custody budget surveys at all. It is in stablecoin settlement volume and tokenized treasury issuance. Those are the rails where permissioned capital actually touches a public chain, and they are measurable without asking anyone's permission. If institutional money is entering the system, it shows up there first — in transfer volume, in issuance, in redemption patterns — long before it shows up in a procurement questionnaire.

So the honest read is this: the 16% is a verifiable lower bound, the 88% is an unverifiable upper bound, and the gap between them is not a data problem. It is a storytelling problem, and the story is being told by the seller.

Takeaway: The Number to Watch Next Cycle

Ignore the budget figure. Track the production ratio. If the next survey reports production climbing from 16% toward 30%, institutional infrastructure has genuinely moved from planning to operating, and the transmission chain is shortening. If the production ratio stays flat while the intent figure grows, you are watching narrative inflation, not adoption.

One question decides everything: when the vendor publishes the next curve, will it publish the proof-of-concept row too — or will it keep selling you the cover?