On the Zcash test network, a shielded transaction now reaches three confirmations in seventy-five seconds. On mainnet, the identical transaction still takes three minutes and forty-five seconds. That gap is not a scheduling artifact or a stalled pull request. It is the visible edge of NU7, the next consensus upgrade, which cuts the block interval from 75 seconds to 25 and is currently waiting on final integration into the node software.
Contrary to consensus, the block-time cut is the least consequential of the three changes bundled into this upgrade. It is also the only one that will be widely covered. NU7 simultaneously routes 60 percent of every block's collected fees out of circulation — with a scheduled reversion to miners around February 2031 — and terminates the original Sprout privacy pool outright. Funds still sitting in Sprout at the activation height do not get burned, redistributed, or rescued. They become unspendable. Not frozen pending governance. Unspendable.
Strip the marketing away and you have a privacy chain performing three distinct operations at once: a monetary re-parameterization, a fiscal covenant with a nine-year maturity, and an irreversible state deletion imposed on a minority of holders. Each belongs to a different audience. Each carries a different failure mode. And in a market where the marginal buyer is an allocator running a risk memo rather than a user running a wallet, the two changes nobody is discussing are the two that actually move capital.
This is a stress test of that claim, not a preview of the feature list.
Zcash has carried a 75-second block interval since genesis in October 2016. The number was never sacred. It was a negotiated settlement between two constraints that dominated 2015-era engineering: block propagation across a slower internet, and the proving cost of the original Sprout zk-SNARK circuits, which were heavy enough that a wallet could take tens of seconds to construct a shielded spend on commodity hardware. Seventy-five seconds bought margin. It also fixed the chain's latency profile for a decade, and in that decade the rest of the industry moved from twelve-second blocks to two-second blocks to sub-second slots.
Zcash's privacy architecture evolved faster than its clock. Sprout, the genesis pool, relied on a trusted setup ceremony whose toxic waste problem was, for years, the single most cited objection to the entire project. Sapling arrived in 2018 with smaller proofs and a much better user story. Orchard arrived with NU5 in 2022, built on Halo 2 recursive proofs and — critically — no trusted setup at all. Three pools, three generations of cryptography, all running in parallel inside one consensus system. Entry into Sprout was disabled at the Canopy upgrade in late 2020. What remained was a deprecated pool with a shrinking balance and an unauditable cryptographic lineage.
Emission followed its own schedule. The block subsidy started at 12.5 ZEC and halved in November 2020 to 6.25, then halved again in November 2024 to 3.125. At 1,152 blocks per day, that puts current daily issuance in the neighborhood of 3,600 ZEC. A development fund that once absorbed 20 percent of block rewards expired in November 2024, replaced by community-directed grants. The chain now funds its own maintenance out of a shrinking subsidy and a fee market that has never been asked to carry real weight.
A Zcash network upgrade is not a soft fork with a friendly name. It is a consensus branch identifier change activated at a coordinated block height, and it requires node operators, miners, exchanges, and wallet vendors to all cross the same line within the same window. Cross it incompletely and you do not get a degraded network. You get two networks. For a transparent chain, that is an inconvenience measured in replay protection and relabeled tickers. For a privacy chain, it is worse: shielded note commitments diverge across the split, anchor sets stop agreeing, and a note that is spendable on one side may be permanently orphaned on the other. Every Zcash upgrade is a coordination event with a cryptographic penalty for failure.
That is the frame. NU7 is not shipping into a quiet network. It is shipping into a network whose entire history is a sequence of high-stakes coordinated state transitions, and it is asking for the most aggressive one yet.
Start with the number everyone is repeating and almost nobody is modeling correctly. Seventy-five seconds to twenty-five seconds is a three-fold compression of the interval between blocks. At 75 seconds, Zcash produces 1,152 blocks per day. At 25 seconds, it produces 3,456. The confirmation math is the part that gets quoted: three confirmations drop from 225 seconds to 75. Merchant settlement, exchange crediting, and bridge attestation all get a three-fold latency improvement for free.
The math that does not get quoted is the issuance math, and it is the reason this upgrade is far more delicate than a latency fix. A block interval is not a speed setting. It is a monetary parameter with a latency wrapper. Triple the block production rate while holding the 3.125 ZEC subsidy constant and daily issuance triples from roughly 3,600 ZEC to roughly 10,800 ZEC. That is a 200 percent inflation shock delivered to a chain trading at a fraction of its cycle highs, in the middle of a bear market, on a scheduled date. No monetary system survives that announcement intact.
Which means the subsidy per block has to fall to roughly 1.0417 ZEC to preserve the existing daily issuance. The upgrade cannot be a pure speed change. It is, mechanically and unavoidably, a re-parameterization of the emission curve — the single most sensitive variable a monetary chain controls.
And it does not stop at the per-block number. Zcash halvings are indexed to block heights, not to wall-clock dates. The first halving landed at block 1,046,400 in November 2020. If the chain produces blocks three times faster while the halving heights stay fixed, every subsequent halving arrives three times sooner in calendar terms. The interval between the 2024 halving and the next one compresses from roughly four years to roughly sixteen months. Either the halving heights get re-derived alongside the subsidy, or the emission schedule silently accelerates and the supply curve that every valuation model was built on stops describing reality.
The upgrade therefore requires the chain to re-specify the block-indexed coordinates of its entire emission curve in a single activation. That is the technical core of NU7, and it is buried three layers beneath a headline about confirmation times.
The latency improvement itself deserves more scrutiny than it is getting, because latency and security are coupled in ways that are easy to miss. Three confirmations at 75 seconds gives an integrator 225 seconds of wall-clock exposure. At 25 seconds, that same three-confirmation policy gives 75 seconds. Every exchange, every bridge, every payment processor that calibrated a confirmation depth against a 75-second block has implicitly calibrated a time budget. Compress the interval and you compress the time budget without anyone editing a config file. A twelve-confirmation policy that used to buy fifteen minutes of reorg resistance now buys five.
That is not a theoretical concern for me. The cumulative losses across cross-chain bridges — north of $2.5 billion by any reasonable tally — are usually attributed to key management and contract logic. A meaningful subset are latency-calibration failures: systems that credited value at a depth that was safe in wall-clock terms under the assumption that block production could not accelerate. NU7 accelerates block production by a factor of three and leaves every downstream integrator to re-derive their own safety margins.
There is a second-order interaction that makes the fee change and the block-time change compound rather than coexist. Fees are collected per block. Triple the block rate and you triple the number of fee-collection events per day, which means the 60 percent removal mechanism fires three times as often. Whatever fraction of fee flow the sink captures today, it captures three times more often after activation. The two changes are not independent line items. They multiply.
The distributional effects land on miners. Validating three times as many blocks triples header and proof verification work, triples the block-propagation load on the network — 2MB blocks arriving every 25 seconds rather than every 75 works out to roughly 80 KB/s of sustained average throughput, trivial in a data center, meaningfully expensive in a bandwidth-constrained region. Variance in block discovery rises. Small solo miners and low-resource node operators absorb that variance first. The block-time cut redistributes risk from users to miners, and the industry will report it as a user benefit.
The fee provision is the part of NU7 that will get a paragraph in most coverage and deserves an essay. The mechanism is simple: 60 percent of the fees collected in each block are removed from circulation, held outside the spendable supply, and returned to miners around February 2031. The framing writes itself — a deflationary mechanism, a supply reduction, a burn.
Run the numbers and the framing collapses. The conventional Zcash transaction fee is 0.0001 ZEC. Even at a generous 10,000 transactions per day, that is one ZEC of daily fee revenue against roughly 3,600 ZEC of daily subsidy — a ratio on the order of one to several thousand. Sixty percent of a rounding error is a smaller rounding error. The fee sink is not fiscal policy today; it is a constitutional precedent.
What the precedent says is more interesting than the amount. Until now, fees in Zcash belonged to miners by default, the same way they do in Bitcoin. NU7 asserts that the protocol has a prior claim on a majority of fee flow — that fees are not automatically miner property but protocol resources subject to collective decision. That is a jurisdictional claim dressed as an accounting entry, and it is the kind of claim that matters enormously later and almost not at all now.
Later arrives when the subsidy stops being the dominant revenue line. After the next halving, block rewards fall to 1.5625 ZEC per block. Post-2032, they fall again. Zcash is on a schedule that terminates in a fee-funded security budget, and every chain on that schedule faces the same question: at what point do fees become the thing that pays for security, and who decides how they are split. NU7 builds the plumbing for that decision years before it is needed. Fee revenue is a signal, not a subsidy.
The reversion date is the strangest and most revealing detail. February 2031 does not coincide with a halving boundary — those fall around 2028 and 2032. It does not coincide with an activation anniversary or a scheduled review cycle that anyone has published. It is a policy date, chosen rather than derived. Which means someone decided that a nine-year escrow was the right shape for this covenant, and that decision tells you what the designers actually believe about time horizons in this ecosystem.
A permanent burn would have been simpler and would have polled better. It would also have severed miners from any residual claim on fee growth, converting them from stakeholders into contractors. A deferred return keeps miners as residual claimants on the fee base while still removing supply for the intervening period. It is a lockbox with a maturity, closer to a bond covenant than to EIP-1559's straight incineration. The Ethereum comparison is instructive precisely because it is wrong: Ethereum burns permanently and tells a deflation story; Zcash escrows temporarily and tells a discipline story.
And a nine-year commitment is only credible if the governance process that made it survives nine years. A covenant with a maturity date is only as strong as the governance that honors it. That is the real test NU7 sets, and it is a test no one can run until 2031.
Then there is Sprout, which is the change with actual casualties.
Sprout is the original pool, the 2016 cryptography, the trusted setup whose ceremony transcript has been dissected for a decade. Entry was disabled at Canopy in 2020, so the pool has been closed to new deposits for years. NU7 does not deprecate it further or warn about it. NU7 removes it from consensus. Any note still in the Sprout pool at the activation height stops being spendable, permanently, with no recovery path and no governance override.
That is a unilateral state transition imposed on a set of holders who, by definition, are the least engaged participants on the network. Someone who shielded ZEC in 2017 and put the key on a paper wallet in a drawer is not reading the ZIP repository. Someone whose exchange custodied a legacy address and never migrated is not watching activation heights. The affected population is small — the Sprout balance has been shrinking for years — but the loss, where it lands, is total.
Finality is a promise until a state transition breaks it. Every chain eventually performs garbage collection on its state. Ethereum has been arguing about state expiry for years because its state bloat is a live consensus risk. Bitcoin's UTXO set grows monotonically and nobody has a clean answer. Solana charges rent. The honest question is never whether a chain prunes. It is whether pruning is announced, scheduled, and bounded — or improvised under duress. Zcash's answer is: announced in advance, scheduled at a block height, and irreversible once executed.
That answer is defensible, and the compliance case for it is stronger than the immutability purists will admit. A deprecated shielded pool running the oldest trusted-setup assumptions on the network, with no realistic audit path and no ability to attract new deposits, is a liability that compounds every year it survives. Every risk memo written about privacy assets in a tightening regulatory environment eventually reaches the same line item. Removing it is a moat-building action, not a technicality.
There is also a liquidity dimension. Whatever Sprout funds do move before activation arrive in Sapling or Orchard, which means the migration produces a one-time bump in shielded supply and a burst of on-chain activity in the weeks before the activation height. Watch that number. It is the cleanest available measure of how many dormant holders the ecosystem actually failed to reach, and it will be published on-chain whether anyone wants it published or not.
I have spent enough of my career writing stress tests to know that the useful ones are unpopular. Here is the version I would circulate internally before any allocation decision touching ZEC.
The emission re-parameterization is the highest-severity failure mode and the least discussed. If the per-block subsidy is mis-set at activation, the consequence is not a UI bug. It is either a tripling of daily issuance — an inflation shock delivered to a bear market — or a silent extension of the emission schedule that invalidates every supply model in circulation. Consensus changes of this class have historically produced the highest-severity incidents in the industry, and the failure surfaces only after the activation height is passed and the first blocks are indexed.
Propagation is the second. At 75 seconds, a two-to-three second p95 block propagation time consumes under four percent of the interval. At 25 seconds, the same propagation consumes eight to twelve percent, and a ten-second tail event consumes forty percent of the interval rather than thirteen. Orphan and stale rates do not scale linearly with interval compression; they scale with the tail. Small miners bear that cost first, which means the decentralization profile of the mining set degrades quietly, in a direction nobody measures quarterly.
The reorg-detection window is the third, and it is the one that reaches outside the chain. Every integrator — exchange, bridge, payment processor, custody provider — has an implicit depth-to-time mapping. Compress the interval by three and every one of those mappings becomes three times tighter in wall-clock terms without a single line of code changing. A bridge crediting at six confirmations had 450 seconds of detection window; it now has 150. The window is where fraud proofs, watchtowers, and manual review live. Shortening it is a security regression that presents as a performance improvement.
Governance coordination is the fourth. If node adoption is incomplete at the activation height, the network splits, and a privacy-chain split is materially worse than a transparent one. Pool accounting diverges, anchor sets stop agreeing, and holders discover which side of the fork their notes landed on by trying to spend them. The upgrade has been running on testnet and is waiting on final integration into node software — that is exactly the window in which the coordination risk lives, and it is invisible from the outside.
The Sprout deadline is the fifth, and it is the only one with a known, quantifiable, non-recoverable loss attached. Dormant holders, legacy custodians, forgotten paper wallets. Small numbers, permanent outcomes, and a reputational cost that is wildly asymmetric to the dollar amount. One documented loss becomes a case study, and case studies outlive bull markets.
The liquidity stress is the sixth, and it is the most ordinary. NU7 is a dated, announced, well-covered event. Dated events in thin order books are sell-the-activation setups, and ZEC's order books in this market are thin. The upgrade does not create demand. It creates a schedule against which existing demand can be expressed, and schedules cut both ways.
Now the part that will annoy people who have already written their NU7 threads.
Zcash does not have a throughput problem. Two-megabyte blocks at 75-second intervals produce roughly 26.7 KB/s of sustained capacity. Utilization is a small fraction of that. The chain has never had a sustained congestion event that required a fee market to sort out. Tripling block production triples the capacity of a pipe that is not full. NU7 solves a constraint that was never binding.
The binding constraints are elsewhere. Venue access is the first: privacy assets have been delisted or restricted across multiple major jurisdictions, and the marginal institutional buyer's problem is not that confirmations take four minutes, it is that their compliance function will not approve the trade at all. Market depth is the second: a thin book means any allocation of consequence moves the price against itself, which caps position sizing regardless of how fast the chain settles. Regulatory option value is the third, and it is the largest — ZEC trades as a claim on the probability that a permissive regime for opt-in privacy emerges, not as a claim on throughput.
So why cut the interval at all? Because latency is a proxy metric that institutions can underwrite. A 25-second block is not a scalability claim; it is a procurement checkbox. A compliance committee reviewing a digital-asset vendor can tick settlement latency, and a chain that fails that tick generates an exception memo that nobody wants to write. The upgrade's real product is a line in someone else's due-diligence template.
Which brings me back to the reframing. The block-time change is retail-facing and mostly symbolic. The fee sink and the Sprout removal are institution-facing and structurally load-bearing, because together they demonstrate that this chain can execute scheduled, irreversible, protocol-level decisions and honor the dates. The scarce asset in a bear market is not throughput; it is the demonstrated capacity to make a nine-year commitment and keep it.
That is the only claim in this upgrade that a risk auditor should underwrite.
Regulation does not remove risk; it prices it. My team spent a substantial part of 2025 quantifying exactly how much it prices it. Assessing compliance costs for three Northern European centralized venues under the EU's Markets in Crypto-Assets regime, we found that regulatory clarity reduced the counterparty risk premium those venues assigned to listed assets by roughly 40 percent. Clarity is not safety. It is a discount rate.
Applied to NU7, the regulatory ledger has three entries. The Sprout removal is a clear credit: eliminating the least auditable pool on the network reduces the residual exposure a venue carries on legacy shielded flows, and every venue risk function that has ever written an exception for Zcash benefits from that pool ceasing to exist. The fee sink is a soft credit: allocators who model protocol revenue read a scheduled 60 percent removal as fiscal discipline, and fiscal discipline is legible to the same committees that find privacy illegible.
The largest credit is structural and predates the upgrade. Zcash's privacy is opt-in. Transparent addresses exist, shielded addresses exist, and the chain is auditable at the address level when the user elects it. That is the difference between an asset a venue can construct a defensible posture around and an asset it cannot list at all. Mandatory-privacy designs face a categorically harder compliance argument in every jurisdiction that has implemented enhanced due diligence for anonymity-enhancing assets. Opt-in privacy is not a weaker product. It is the only version of the product that clears the bar.
That credit is time-boxed. Enhanced-due-diligence thresholds and implementing standards for anonymity-enhancing assets are tightening across the EU's AML framework through the middle of this decade. NU7 improves Zcash's compliance posture against a regime that is simultaneously getting stricter. The moat is real and it is narrowing at the same time. The upgrade buys time, not immunity.
Project forward, because the interesting question is not what NU7 does on activation but what it reveals about where the bottleneck moves next.
Across the industry, consensus latency is converging on zero. Sub-second slots exist and are being commoditized. Privacy latency is a different animal: the cost is not agreement, it is proof generation, and proof generation happens on the client. Halo 2 made recursive proofs practical and removed the trusted setup, but a shielded spend still requires a proving operation that takes seconds on commodity hardware and considerably longer on a phone.
Here is the non-obvious consequence of the block-time cut. At 75 seconds, an eight-second proof generation is eleven percent of the interval — invisible to the user. At 25 seconds, the same eight-second proof is a third of the interval. NU7 moves the performance bottleneck from consensus to client-side compute, and it does so on a date, without any corresponding improvement in proving hardware. Every wallet vendor on the network is about to discover that their UX constraint changed underneath them.
That is where the accrual vector sits. In 2026 I modeled decentralized compute networks and found that as AI demand scaled, the binding constraint shifted from capital to GPU availability, with value accruing to nodes providing low-latency inference rather than storage. Zero-knowledge proving is the same shape of workload: massively parallel, GPU-friendly, latency-sensitive, and increasingly separable from the chain that consumes it. I estimated a market in the low billions for AI-optimized blockchain infrastructure by 2028. Accelerated proving is a subset of that estimate with better unit economics, because the demand is protocol-mandated rather than speculative — every shielded transaction on a 25-second chain needs a proof, and it needs it inside the interval.
By 2031, when the fee escrow matures, two paths exist. On the first, fees remain immaterial, the reversion is a symbolic accounting entry, and the covenant's value was entirely in its demonstration effect. On the second, shielded settlement has become load-bearing for something — a payments corridor, a custody rail, an institutional venue that needed auditable privacy — and 60 percent of a material fee base becomes a genuine fiscal question decided by a governance process that will not resemble the one that set the date. Nobody knows which path obtains. The protocol will have to decide anyway, on a schedule it committed to nine years in advance.
The question NU7 poses is not whether 25 seconds is fast enough. It is whether a chain that can schedule the removal of 60 percent of its fees and the deletion of its oldest privacy pool — and honor both dates without a governance crisis — is more trustworthy than a chain that promises never to change anything. Speed is the easy part. The activation height has not been set yet, and that is the only number that matters.
