Investing in Bitcoin Ordinals

On April 20, 2024 the Bitcoin network’s users paid its miners 1,206.5 bitcoin in transaction fees. At that day’s price of $64,908 that is $78.3 million, in one day, for the right to put data in a block. Measured in dollars, no day in Bitcoin’s history has paid its miners more. Measured in bitcoin, one day has: December 22, 2017, at the height of that mania, paid 1,495.7 BTC, worth $21.4 million at the price of the day. What holds on either measure is the share. Fees made up 74.8% of everything miners earned on April 20, the highest daily share in the record, against an all-history average of 1.48%. The subsidy had just halved. The fees more than covered the difference.

What the money bought was mostly Runes — a fungible-token standard launched deliberately at the halving block, so that the first etchings would land in the same block as the reward cut. Ordinals had done something similar twice before: in May 2023, when a JSON-based token standard called BRC-20 pushed fees to 13.6% of miner revenue for a month, and in November and December 2023, when inscriptions took 17.28% of miner revenue across two months and Bitcoin topped the monthly cross-chain NFT sales tables that Ethereum had owned since 2017 — on a tracker’s count with a method problem that section 6 sets out.

Two years later, in the 143 days of 2026 through May 23, fees ran 0.59% of miner revenue. The average Bitcoin transaction cost $0.40. Blocks were still full — roughly 498,500 transactions a day, twice the 2022 rate — but almost nobody was bidding for the space. That collapse, from a third of miner revenue to well under one percent, is the whole story of this market compressed into one series, and it is the number to hold in your head while you read about the collections.

A bitcoin divides into 100,000,000 satoshis and no further, and ordinal theory gives each of those satoshis a serial number based on the order in which it was mined. The first satoshi of the genesis block is 0; the last satoshi of that block is 4,999,999,999. Roughly 2.1 quadrillion of them will ever exist. That is the entire foundational idea, and it requires no sidechain, no separate token and no change to Bitcoin’s consensus rules. The project’s own handbook is blunt about this: “It works right now.”

The second half of the idea is the transfer rule, and it is where the practical risk in this asset comes from. Satoshis move from a transaction’s inputs to its outputs first in, first out. Think of the inputs as a queue of numbered satoshis and the outputs as a row of empty slots; you walk the queue and drop each satoshi into the first slot with room. A transaction with inputs of 2, 1 and 3 satoshis and outputs of 4 and 2 sends the first four input satoshis to the first output and the last two to the second. Nothing is stored anywhere; the assignment is a deterministic function of transaction structure that anyone can recompute from the chain.

And fees fall out of the same rule. A transaction that pays a fee sends more satoshis in than come out, so the satoshis at the end of the queue have no slot. They go to the miner, appended to the coinbase transaction in the order their transactions appeared in the block. This is not a bug the protocol tolerates; it is the specification. The handbook says an inscribed satoshi can be “transferred, bought, sold, lost to fees, and recovered.” Section 9 is about not doing that.

Rarity, and why anyone cares about a plain sat

Because satoshis are now individually identifiable, they can be graded, and ordinal theorists grade them by Bitcoin’s own calendar. A common sat is any sat that is not the first of its block. An uncommon sat is the first sat of a block — one per block, roughly 144 a day. A rare sat is the first sat of a difficulty adjustment period, one every 2,016 blocks or about two weeks. An epic sat is the first of a halving epoch: four have ever existed. A legendary sat is the first of a cycle, where a halving and a difficulty adjustment coincide, which happens about every 24 years; the first conjunction is expected in 2032. A mythic sat is the first satoshi of the genesis block, and there is exactly one, and it has never moved.

This is numismatics rebuilt from first principles, and it should be read the way you read coin collecting: the scarcity is real and verifiable, the demand is a social fact that may or may not persist. Uncommon sats trade at small premiums to their face value in bitcoin, and rare sats as curios in a market of a few hundred people.

The epic sats are the exception that proves the rule. There are four, and the one mined in the 2024 halving block was auctioned on its own five days later by ViaBTC, the pool that mined it, for 33.3 BTC — $2,134,452 on April 25, 2024, on the exchange CoinEx. One satoshi, two million dollars, and no second sale to check it against. That is a market of one buyer, once. Do not confuse the arithmetic certainty of the supply with any certainty about the price.

An inscription is a content type and a byte string — the same data model as an HTTP response — written into a Bitcoin transaction and bound to a satoshi. The content lives entirely on-chain. There is no IPFS hash, no Arweave transaction, no server that has to keep paying its bills, which is the single strongest technical claim this asset class has against Ethereum NFTs, where much of the art is a pointer to a file somewhere else. Our Investing in Generative Art guide puts a number on how often that pointer stops resolving, and our Investing in Tokenized Real-World Assets guide draws the other distinction that matters here: an inscription is the thing itself, whereas a tokenised asset is a claim on something off the chain and is worth exactly what redeeming it is worth.

The mechanism is a piece of engineering that deserves to be understood before you spend money on its output. Inscription content is stored in taproot script-path spend scripts, which is possible because the taproot soft fork, activated in November 2021, put very few restrictions on what a script may contain and gave witness data a discount: a byte of witness data counts as a quarter of a byte against the block’s weight limit. The content is wrapped in an “envelope” — OP_FALSE OP_IF, then the pushes, then OP_ENDIF — which is an unexecuted conditional, a no-op that changes nothing about what the script does.

Two transactions are needed rather than one, because a taproot script spend can only be made from an existing taproot output. A commit transaction creates an output committing to the script; a reveal transaction spends it and publishes the content. Individual data pushes cannot exceed 520 bytes, so large content is chopped into many pushes.

That design fixes the cost of inscribing to the fee market, in a formula you can compute yourself. One million bytes of content becomes roughly 250,000 vbytes after the witness discount, which at the minimum relay fee of 1 satoshi per vbyte is 250,000 satoshis — 0.0025 BTC. The ord handbook quotes that as $50 per million bytes at $20,000 per bitcoin. At the $78,576 bitcoin price on our tape on September 8, 2026 the same megabyte costs $196. In a busy fee market at 50 sat/vB it costs about $9,800. The price of publishing is a function of two things you do not control, and the second one moved by two orders of magnitude inside 2023.

The properties that follow

Four consequences matter to a buyer. Inscriptions are immutable: there is no contract owner, no admin key, no upgrade path, so the thing you buy cannot be changed by its creator, which is not true of many Ethereum NFTs and is very hard for a non-technical buyer to check there. They are on-chain, so the content cannot be lost while the chain exists. They support no on-chain royalties at all, which the project frames as honesty rather than a limitation — Ethereum spent 2022 and 2023 discovering that on-chain royalties were unenforceable anyway. And they are numbered: inscriptions get sequential numbers in the order their reveal transactions appear in blocks, which makes “low inscription number” a collectible property in itself, the way a low serial number is on a banknote.

There is one wrinkle a buyer will meet in listings. Inscriptions made in ways early versions of ord did not recognise were assigned negative numbers and called “cursed”. At the jubilee — block 824,544, mined on January 6, 2024 — cursed inscriptions were “vindicated” and given positive numbers going forward. A listing that advertises a cursed inscription is describing a dated technical artifact, not a defect and not a virtue.

Inscriptions are unique by construction, and within three months of launch someone had built fungible tokens on top of them anyway. BRC-20, introduced in March 2023 by a pseudonymous developer known as domo, two months after Casey Rodarmor launched ordinal theory, works by inscribing small JSON documents that say “deploy this ticker”, “mint this many”, “transfer this many”. Bitcoin itself has no idea any of this is happening. The tokens exist only in the sense that indexers agree on how to read a sequence of inscriptions, which is a materially weaker guarantee than the one inscriptions themselves carry, and the reason a BRC-20 balance is not comparable to a bitcoin balance.

Runes is the correction, written by the same author as ordinal theory and specified properly. A Runes protocol message is a runestone, stored in a transaction output whose script begins with OP_RETURN followed by OP_13 and then data pushes that decode into a sequence of 128-bit integers. One runestone per transaction. Runes are identified as BLOCK:TX — the block they were etched in and the index of the etching transaction within it — so the twentieth transaction of the five-hundredth block etches rune 500:20. Because it uses OP_RETURN rather than inscribed JSON, a rune balance is a property of transaction outputs that any node can compute, and no separate indexer consensus is required to say who owns what.

Everything about a rune is fixed at etching and visible before you buy. Its name is one to twenty-six letters, A to Z, with optional spacers that do not affect uniqueness. Its divisibility, symbol, premine — the units the etcher allocates to themselves — and its mint terms are immutable once set, even to the etcher. A mint may be capped, bounded by start and end heights, or both, and it closes when any term fails. Malformed runestones are called cenotaphs: runes fed into a transaction with a cenotaph are burned, runes etched in one are unmintable, and mints in one count against the cap while the tokens are destroyed. That is deliberate, an upgrade mechanism that lets future versions add meaning without misleading old software.

Read that specification as a due-diligence checklist, because it is one. A rune with a 90% premine is a token whose issuer holds 90% of it, and the chain will tell you so before you send money. Nothing in the protocol makes the etcher’s promises enforceable, and nothing makes the token a claim on anything. Runes went live at the fourth halving block, 840,000, mined at 8:09 p.m. US Eastern on April 19, 2024 — which is just after midnight UTC on April 20, and the reason the fee record in the cold open falls on the 20th while the halving is usually dated the 19th. That record is what the launch cost the people who wanted to be first.

Every inscription, every BRC-20 mint and every rune etching is a Bitcoin transaction competing for the same 4 million weight units per block as a payment, and that competition is measurable in a way almost nothing else in this market is. We downloaded Coin Metrics’ community network dataset for Bitcoin on September 10, 2026 — 6,351 daily rows ending May 24, 2026 — and computed fee totals, miner issuance and transaction counts directly. Everything in this section comes from that file, not from a vendor’s summary.

The pre-Ordinals baseline is 2022: 5,375 BTC of fees across the year, $142M at daily prices, 1.59% of miner revenue, and 254,947 transactions a day. Then inscriptions arrived. Across the 23 months from February 2023 to December 2024 users paid 38,220 BTC in fees, about $1.71B, at 6.78% of miner revenue. Had the 2022 run rate simply continued, those months would have produced roughly $272M. The difference — call it $1.44B of incremental fee revenue on a 2022 counterfactual — is the clearest measure of what this asset class was worth to the people who secure Bitcoin.

Three episodes carry almost all of it. In May 2023, the month BRC-20 minting went vertical, fees averaged 146.5 BTC a day and took 13.61% of miner revenue; the network processed 525,756 transactions a day against 282,548 in January. In November and December 2023, when inscription volumes peaked, two months of fees came to 11,801 BTC, $481M, 17.28% of revenue, with December alone averaging 255.6 BTC a day and an average transaction cost of $19.28. And in the twelve days from April 19 to April 30, 2024, spanning the halving and the Runes launch, fees ran 3,077 BTC, $200M, 34.48% of everything miners earned.

Then the bid disappeared. Fees averaged 4.7 BTC a day across 2025 and 2.6 a day through May 2026; the average transaction cost fell from $5.20 in 2023 to $1.13 in 2025 and $0.40 in 2026. Transaction counts did not fall with them — 498,506 a day in 2026 against 254,947 in 2022 — which tells you the blocks are still full and the space is nearly free. Demand for Bitcoin block space at any meaningful price was a 2023–2024 phenomenon, and by every reading we can compute it has not returned.

Bitcoin pays for its own security with the block subsidy plus fees, and the subsidy halves every four years on a published schedule until it rounds to nothing around 2140. Every honest analysis of Bitcoin’s design treats the replacement of that subsidy by fee revenue as the central open question, and the Ordinals project made a specific claim about it. Inscription content is stored on-chain, the handbook argues, so demand for block space from inscriptions is effectively unlimited, which “creates a buyer of last resort for all Bitcoin block space” and raises a security budget that is “vital for safeguarding Bitcoin’s transition to a fee-dependent security model.”

That is a falsifiable claim, and three and a half years of data have now tested it. Over the wave itself, the claim held: fees at 6.5% of miner revenue for two consecutive years, against 1.6% in 2022, is a real and substantial contribution to the security budget, and $1.71B is real money. What did not hold is the durability. The buyer of last resort turned out to bid only when speculation in the tokens was hot, and when it cooled the bid went to 0.59% of miner revenue — lower than the pre-Ordinals baseline, not higher. A source of demand that vanishes precisely when asset prices fall is the opposite of what a security budget needs, because that is when miner revenue is already under pressure.

The arithmetic ahead is unforgiving. The subsidy is 3.125 BTC a block, roughly 450 BTC a day — issuance ran 446 a day across 2026 to May 23 — and the 2028 halving takes it to about 225. At 2026 fee levels of 2.6 BTC a day, fees would need to multiply roughly 170 times to replace the subsidy that exists today, and about 85 times to replace what remains after 2028. Neither multiple has ever been approached. The most fee-intensive stretch in section 4, the twelve days from April 19 to April 30, 2024, ran fees at 34.48% of miner revenue, which is fees at roughly half of issuance — a multiple of about 0.5, against the 170 they would need today.

Our Investing in Bitcoin guide works through the mining economics and the hashprice pressure in detail; the point here is narrower. If you are considering Ordinals because you believe inscriptions strengthen Bitcoin, the evidence to date is that they strengthened it for 23 months and then stopped.

Sales volume in this market has always been reported by trackers rather than by venues, and the trackers themselves have proved less durable than the chain. Bitcoin’s share of NFT sales went from nothing in January 2023 to topping the monthly cross-chain tables in December 2023, when CryptoSlam counted about $881 million of Bitcoin NFT sales against roughly $365 million on Ethereum. Read that comparison with its method in mind: CryptoSlam’s Bitcoin totals include BRC-20 token trades, while its Ethereum totals exclude ERC-20 trades, so the gap flatters Bitcoin by an amount nobody has published. What needs no such caveat is the block-space signature of that period, and it is the fee series in section 4: two spikes and a collapse.

The trackers then thinned out. DappRadar, one of the two standard sources for NFT volume, announced its shutdown on November 17, 2025 (CoinDesk), so there is no fourth-quarter 2025 report from it. CryptoSlam’s full-year 2025 figure for all chains was $5.63B, down 37% on 2024’s $8.8B, with an average sale price of $96 against $124 (Cointelegraph, January 2026). For Bitcoin specifically, monthly Ordinals volume ran roughly $53M in January 2026, $33.6M in February and $46.8M in March, as reported by Blockspace and carried by Yahoo Finance — secondary sourcing, dated, and the best available.

Hold that chart against the fee chart and you have the market’s condition stated as plainly as data can state it. Bitcoin still carries roughly twice the daily transaction count it carried before inscriptions existed, and users pay $0.40 a transaction to do it. The activity did not leave. The willingness to pay did — which is what a collapse in speculative demand looks like from the inside of a blockchain, and it is a cleaner signal than any volume tracker, because nobody can wash-trade a fee.

Four collections carry most of the recognisable value in Bitcoin-native collectibles — our characterisation, since nobody publishes a concentration measure for this market — and all four date from the fourteen months after launch. NodeMonkes is 10,000 pixel-art inscriptions, inscribed in February 2023 at inscription numbers 83,522 to 111,319 and so among the earliest ten-thousand-piece collections on the chain, though they sat unreleased until a Dutch auction in December 2023. Bitcoin Puppets is 10,001 hand-drawn images, free-minted on January 3–4, 2024, and the other that behaved like a blue chip. Both sat at the top of the Bitcoin-native market through 2024, and NodeMonkes recorded its best-known single sale — about $1 million, reported by CoinDesk on March 4, 2024.

The other two came from one team. Taproot Wizards, the project of Udi Wertheimer, Eric Wall and the developer known as 0xFAR, began with a single inscription in February 2023 that filled 3.96 MB of a four-megabyte block, a deliberate demonstration that the block-size debate had been settled by taproot rather than by argument. Quantum Cats, 3,333 pieces sold in late January 2024, was the same team’s second collection. Ordinal Maxi Biz, Runestone and a long tail of others exist beneath them.

What we cannot give you is a current floor, and the reason is itself the most important fact in this section. Our tape carries four NFT floors, all of them on Ethereum, because those four are the collections our vendor prices densely enough, day after day, for us to treat the number as a price. There is no Bitcoin collection on it.

Quoted floors for the larger Bitcoin collections do exist — vendors publish pages for NodeMonkes and Bitcoin Puppets — but they rest on a handful of trades, and a floor is only as good as the depth beneath it. In September 2026, pricing a NodeMonke means asking a marketplace with a handful of daily trades what its lowest ask is, and that number is a quote, not a price. Any floor you are shown for a Bitcoin collection should be treated as an advertisement until you have seen three of them agree.

Our tape also measures the thing that matters more than any floor. Across seven daily readings from September 1 to September 8, 2026, twenty-four-hour trading volume in that basket of four blue-chip Ethereum collections averaged $685,896 a day and read $566,125 on the last day. Those four collections carry a floor market capitalisation near $1.0B, so the whole blue-chip Ethereum complex turns over about 0.06% of itself per day. That is our own basket and our own method, not a market-wide figure, and it describes the deepest corner of the NFT market; our Investing in CryptoPunks guide takes the deepest collection in it apart line by line. The Bitcoin corner is thinner than any of those four, and that turnover figure is the yardstick to hold up against any pitch that describes Ordinals as liquid.

Trading an inscription is mechanically different from trading an Ethereum NFT, and the difference is a genuine advantage. There is no approval to grant, no contract with permission over your collection, and no blind signature. A sale is arranged with a partially signed Bitcoin transaction: the seller signs an input spending the inscription and an output paying themselves the price, the buyer completes it with their own inputs, and either the whole transaction confirms or nothing happens. You can read exactly what a PSBT does before you sign it. The Ethereum NFT market’s characteristic disaster — a phishing signature that drains a wallet through a marketplace approval granted months earlier — has no equivalent here.

The venue layer is where the weakness sits. Magic Eden, which ran the dominant Bitcoin Ordinals marketplace, shut it on March 9, 2026, along with its Ethereum-compatible markets (Blockspace, carried by Yahoo Finance). That is the single most consequential structural event in this market since Runes, and it should be read the way you would read the closure of the main auction house for a collecting category: the assets are unaffected, the ability to sell them at a known price is not. It was not the only exit. Ord.io, the explorer most of this market used to look inscriptions up, shut down on June 1, 2026 along with its companion app, its founder saying plainly that the money had run out.

What remains is OKX’s marketplace, Unisat, Ordinals Wallet, Gamma and a handful of smaller venues. Their fees differ by an order of magnitude, fall on the buyer at some venues and the seller at others, and change without notice; one of them was running a temporary zero-fee promotion during 2026. Obtain the schedule from the venue on the day you trade rather than from any guide, this one included.

Wallets: the list that got shorter

The wallet requirement is specific, because it is the wallet and not the marketplace that decides whether you keep the thing you bought: you need software that understands ordinal theory, meaning software that knows which unspent output holds which inscription and will not spend it by accident. Xverse, Unisat and the reference ord wallet do. A general-purpose Bitcoin wallet does not, and section 9 explains what that costs.

That list shortens, and one of the shortenings is dangerous rather than merely inconvenient. Leather, long one of the standard choices in this market, ended its support for Ordinals and BRC-20 on May 18, 2026 and now treats every UTXO as ordinary bitcoin. The inscriptions did not move when the support did: they are still sitting in those wallets, in software that has stopped seeing them, which is precisely the condition section 9 warns about.

If you are holding one there, the sequence is to stop making ordinary bitcoin payments from that wallet, and then move the inscription out through an ordinals-aware wallet or the ord tool, in that order. Any list of supported wallets, including the one three paragraphs above, is a statement about the day it was written — ours is September 10, 2026 — and support can be withdrawn by a software release without anything happening on the chain at all.

Hardware signing works throughout: the standard devices sign PSBTs. But the coin selection has to happen in software that knows what it is holding, and the device cannot supply that knowledge itself.

The failure mode unique to this asset class follows directly from the transfer rule in section 1, and it destroys value rather than merely losing money. An inscription is not a token in a registry; it is a satoshi sitting at a known position inside a specific unspent transaction output. If you send bitcoin from a wallet that holds that UTXO and the wallet’s coin-selection algorithm picks it as an input — which it will, because ordinary wallets choose inputs by size and age, not by what is inscribed on them — the inscription moves. It goes to whoever receives the change, or to the recipient, or, if it falls past the last output slot, to the miner as part of the fee.

The ord project’s own guidance states the case without softening it: receiving an inscription in an unsupported wallet is safe, but “if that wallet is then used to send bitcoin, it may select the UTXO containing the ordinal as an input, and send the inscription or spend it to fees.”

There is no undo. There is no contract to pause, no marketplace support desk, no chargeback. A miner who receives your inscription in a coinbase output owns it, and whether you get it back depends entirely on whether that pool has a policy of returning them and whether you can prove the loss quickly enough to matter.

The defence is sat control: a wallet that tracks which UTXO holds which inscription and never selects those UTXOs when constructing a spend. This is the whole reason the ordinals-aware wallets exist. It is also why the correct structure is not one wallet but two, and why the second one should hold no ordinary bitcoin at all.

The structure that removes the risk

Use a dedicated inscription wallet, derived from its own seed, on hardware, that has never held spendable bitcoin and never will. Fund each purchase by sending exactly what is needed from a separate spending wallet, and when the inscription arrives, leave it alone. Never send bitcoin out of the inscription wallet except by an explicit, ordinals-aware transfer of a named inscription. If you must consolidate dust or move funds, do it in the other wallet.

The rest of the custody stack is the same one our Bitcoin guide sets out in detail — a hardware signing device, a seed on steel and not on paper or in a photograph, and, above a threshold you set honestly, a multisignature arrangement so that no single device or location is a single point of failure. Two things are specific to inscriptions. First, insurance effectively does not exist for an individual: specialist crypto insurers cover custodial platforms, not a person with a signing device. Second, the physical-coercion risk that has grown with the bitcoin price applies with full force to anyone publicly identified as holding a valuable collection, and a collection is far easier to identify from public chain data and social media than a bitcoin balance is.

Ordinals carry no storage cost, no insurance premium and no annual fee, which makes them cheaper to hold than any physical collectible. Four layers of cost remain — entry, network, venue and spread — and all four can be priced before you commit. Three of them are published. The fourth, and the largest, is not.

The entry cost is buying bitcoin. Kraken Pro’s entry tier is 0.80% taker and 0.40% maker on the cross-platform tiers Kraken introduced on July 9, 2026, which roughly doubled the entry rate it had charged before — a reminder that an exchange schedule is a fact about a date, not a constant. Coinbase Advanced’s entry tier is 1.20% taker and 0.60% maker. Coinbase’s simple buy button is roughly 1.5% plus a spread near 0.5%, and a card purchase is about 4%. Schedules read September 10, 2026. Post a limit order and pay a fraction of what the buy button charges.

The network cost is the fee to move the inscription, which section 4 has already priced: the Bitcoin network averaged $0.40 a transaction across 2026 to May 23, and an inscription transfer is several times the size of a simple payment, so budget a few dollars in ordinary conditions and remember that the same transfer would have cost $19 on the average December 2023 day. The venue cost is the marketplace’s cut, which at the surviving Ordinals venues runs from zero to about 2.5% of the sale price and falls on the seller at some and the buyer at others; the schedules change without notice and must be read fresh. The 2% used in section 12 is an assumption chosen to be unkind, not a quote.

The layer that dominates all three is the spread, and it is the one layer nobody publishes. In a market where a collection trades a handful of times a day, the gap between the best bid and the lowest ask is not 1%; on our own reading of the surviving venues it is routinely 10–30%, and in a fast decline the bid side simply empties. Treat that range as our estimate rather than a measurement: no vendor computes a bid-ask series for Ordinals, which is itself the point. The absence of a carrying cost is real, but it buys you nothing on the way out: the exit is the cost here, and it is paid in one lump on the day you need it most.

Set against that, the round trip on a blue-chip Ethereum NFT runs roughly 5–15% at an unchanged floor, as our NFT guide prices it, and a physical collectible sold at auction gives up 25–30% between buyer’s premium and seller’s commission. Ordinals sit at the cheap end on published fees and at the expensive end on the spread, and only the published fees appear in anyone’s marketing.

United States federal treatment, general, as of September 10, 2026. Digital assets are property, so every disposition is a taxable event, and the specific hazards for an Ordinals buyer are sharper than for an NFT buyer on Ethereum because of what you pay with.

Spending bitcoin is a sale

Buying an inscription for 0.13 BTC is a disposition of that bitcoin at its dollar value on the day, and if you acquired it lower you realise the gain immediately, before the inscription has done anything at all. Someone who bought bitcoin at $20,000 and spent it at $78,576 realised $58,576 of gain per coin on the purchase. Long-term rates apply if the bitcoin was held more than a year; short-term ordinary rates if not. Keep the lot record for the bitcoin you spend, because under Revenue Procedure 2024-28 basis has been tracked wallet by wallet since January 1, 2025, and the universal-pooling method many holders used before that is gone.

The collectibles rate on the way out

IRS Notice 2023-27, issued March 21, 2023, announced a look-through analysis: an NFT is a collectible under section 408(m) if the underlying right or asset is one. Pending final guidance, a digital artwork sold as art is at real risk of collectible treatment, which caps the long-term rate at 28% rather than 20%, plus the 3.8% net investment income tax where it applies. The Notice also confirms that a collectible NFT cannot sit in an IRA without a deemed distribution. Whether a pixel-art inscription is a “work of art” for this purpose is unresolved. Plan for 28% and be pleasantly surprised.

Reporting, and the non-custodial myth

Form 1099-DA requires custodial brokers to report gross proceeds on sales effected from January 1, 2025, and cost basis on sales effected from January 1, 2026 where the asset was acquired in that broker’s custody on or after that date; basis stays voluntary for anything transferred in from elsewhere, which is most of what an Ordinals buyer holds. A PSBT trade on a non-custodial venue generates no form; the obligation to report the gain is identical. The IRS has the chain.

One advantage remains, for now. Section 1091’s wash-sale rule applies to “stock or securities,” and neither bitcoin nor inscriptions were covered as of September 10, 2026, so a loss can be harvested and the position rebought the same day. Successive proposals have sought to close this — most recently a bill introduced in Congress in June 2026 — and none has passed. Verify it for the tax year you are in rather than relying on this sentence.

The arithmetic below prices a single position from purchase to sale with every fee and both tax events shown. The purchase price is an assumption, not a quote; the fee rates and tax rates are the ones set out in sections 10 and 11. Federal only, single filer above the net investment income tax thresholds, no state tax.

The purchase

You already hold bitcoin, acquired more than a year earlier at $30,000 a coin. You spend 0.1273 BTC on an inscription when bitcoin is $78,576 — the price on our tape on September 8, 2026 — so the inscription’s cost basis is $10,000. That purchase is also a sale of the bitcoin: 0.1273 × ($78,576 − $30,000) = $6,184 of long-term capital gain, taxed at 20% plus the 3.8% net investment income tax, or $1,472, due that April. You have not yet made a cent on the inscription.

The sale, three years later

The collection has appreciated 50% and you sell for $15,000. The marketplace takes 2%, $300. Network fees on the transfer and settlement come to about $5. Proceeds are $14,695, giving a gain of $4,695 over your $10,000 basis. Treated as a collectible under the Notice 2023-27 look-through, the long-term rate is 28% plus 3.8% NIIT — 31.8%, or $1,493.

Of the $5,000 the price gave you, $3,270 — 65% — went to the venue, the network and the IRS: $300 in marketplace fees, $5 in network fees, $1,493 in tax on the sale and $1,472 in tax triggered by the purchase itself. You keep $1,730 on $10,000 over three years, a compound annual return of 5.5%. A 50% gross gain became a mid-single-digit net one, and the largest single line in that table is the tax on a bitcoin sale you may not have realised you were making.

Now price the exit honestly. Suppose that instead of hitting the quoted floor you sell into a bid 30% below it, at $10,500, which section 10 says is an ordinary outcome in a market this thin. The marketplace takes $210, proceeds are $10,285, the taxable gain is $285 and the tax $91. You are left with $10,194 from the inscription — and you have already paid $1,472 in tax on the purchase, so your net position is $8,722 on $10,000, a loss of 12.8% on a collection that rose 5%. Simply holding the 0.1273 BTC would have left you with $10,000 and no tax event at all.

Bitcoin has argued about arbitrary data in blocks since long before Ordinals, and the argument decides what future software may and may not do to the asset you hold. The position against is that Bitcoin is a monetary settlement network and that inscriptions are spam: they raise the cost of running a node, they crowd out payments, and they were enabled by a soft fork whose authors did not intend this. The position for is the one in section 5 — that a bidder for block space is exactly what a fee-dependent security model needs — plus the simpler point that a transaction paying the going rate is not spam by any definition a miner would recognise.

The mechanics decide who can do what about it, and the two protocols sit in different places. Inscriptions live in taproot witness data, inside an unexecuted conditional, and are therefore indistinguishable from an ordinary script spend without deliberate pattern-matching; a node that wanted to reject them would have to inspect and filter script contents, and a miner who accepted them would collect the fees anyway. Runestones live in OP_RETURN outputs, which nodes have always had an explicit policy setting for. That asymmetry is why the relay-policy debate has never been able to touch inscriptions and has repeatedly touched the fungible-token layer — and why the people who want inscriptions gone eventually stopped arguing about policy and proposed a change to the consensus rules instead.

The escalation began with the reference implementation. Bitcoin Core 30, published in October 2025, raised the default -datacarriersize from 83 bytes to 100,000 and allowed multiple OP_RETURN outputs to relay. That is a policy change and not a consensus change — it governs what a node passes along, not what a block may contain — but it was read by opponents as the reference implementation taking a side. Node operators who disagreed moved to Bitcoin Knots, which lets them refuse to relay inscriptions, BRC-20 data and runes, and whose share of reachable nodes has grown from a rounding error to a substantial minority; the published counts come from node crawlers and are disputed by both camps, so we do not print one.

From there the fight moved to consensus. Mainline Knots has enforced BIP-110, the “reduced data” temporary soft fork, since a May 2026 release. For roughly a year BIP-110 would cap most outputs at 34 bytes, restore the 83-byte OP_RETURN limit, cap witness elements at 256 bytes and ban OP_IF and OP_NOTIF in tapscript. That last item is aimed squarely at the inscription envelope in section 2, and the proposal is understood to block most new inscriptions while leaving most rune transactions valid.

It activated on August 8, 2026, at block 961,632, with about 2.53% of hashpower behind it. A soft fork enforced by 2.53% of hashpower is a minority chain, and this one behaved like one: two blocks in roughly eight hours while the main chain advanced by dozens, and then it stalled. That is where it stood when we read the block-space sources on September 10, 2026, five weeks later. Treat those five weeks as the freshest and least settled material in this guide. The mechanism underneath them is not fresh at all: consensus rules are enforced by the hashpower that builds the chain, so a rule miners will not enforce does not become a rule, whatever any node operator has installed.

Through all of it the reference implementation kept shipping. The ord project’s changelog records 68 releases, from 0.0.1 on June 5, 2022 to 0.29.0 on August 5, 2026, and its November 12, 2025 release updated the install script to Bitcoin Core 30. Development continued straight through the volume collapse and the block-space fight alike, which is a meaningful signal about durability even where it says nothing about price.

The practical conclusion for a holder is narrow, firm, and independent of how the current round ends. Relay policy is not consensus: no filter can invalidate an inscription already in a block, and none can stop a miner from including a paying transaction. A consensus change could, in principle — but only with the hashpower to enforce it, and August 2026 is the dated illustration of what happens without it. What policy can still do is make new inscriptions and new etchings harder or costlier to broadcast, which affects the supply of future competition to whatever you own and, on the evidence of the last three years, matters far less than whether anyone wants to buy it. The risk you are carrying is a market risk, not a protocol risk.

The sequence below is the one that removes the avoidable losses. It assumes you have already decided the position is small enough that a total loss would be an annoyance rather than an event, which is the only sensible frame for an asset whose main marketplace closed inside three years of its peak.

1. Read the chain before the pitch. Every claim about an inscription or a rune is verifiable from a block explorer: the inscription number, the content, the sat it sits on, the etching block, the cap, the premine. Verify the specific item, not the collection.
2. Build the two-wallet structure first, before you buy anything. A spending wallet you use normally, and a dedicated inscription wallet on its own seed, on hardware, that never makes an ordinary bitcoin payment. Confirm on the day you set it up that the wallet you pick still supports Ordinals: the supported list shortens, and one long-standing choice left it on May 18, 2026. Section 9 is the reason for both halves of this step.
3. Buy the bitcoin with a limit order on the cheaper venue — 0.40% maker on Kraken Pro’s entry tier against roughly 1.5% plus spread on a buy button, on the schedules in force in September 2026 — and record the date, the dollar cost and the amount. That is a tax lot, and you will need it twice.
4. Send only what the purchase requires to the inscription wallet. Do not park a balance there.
5. Buy through a PSBT you have read. Confirm which input spends which inscription and which output pays whom before signing. This is the step that Ethereum’s approval model does not let you take, and it is worth taking.
6. Record two cost bases on the day: the gain realised on the bitcoin you spent, and the dollar basis of the inscription. Section 11 explains why the first one surprises people in April.
7. Then leave it alone, and set a calendar reminder each January to recompute the fee-share test in section 5 from Coin Metrics’ free dataset.

Six readings would change the view, and each has a threshold and a source you can check yourself.

Fees as a share of miner revenue, computed annually

The test in section 5: above 5% for two consecutive calendar years without a token mania would mean block-space demand had become structural. It ran 6.50% in 2023, 6.52% in 2024, 1.03% in 2025 and 0.59% in 2026 through May 23, on Coin Metrics’ community dataset, which is free and updates daily.

The number of venues

After Magic Eden’s Bitcoin marketplace closed on March 9, 2026, a further closure among the remaining marketplaces would take the market from thin to bilateral. Watch for a venue publishing a sunset notice; the assets survive, the exit does not.

Whether a real floor ever appears

Quoted floors exist for the larger Bitcoin collections; what they lack is the daily trade density that makes a quote a price, which is why our tape carries four Ethereum floors and no Bitcoin collection as of September 10, 2026. A Bitcoin collection trading densely enough that independent vendors agree on its floor day after day would be the first hard evidence that this market had become priceable, and it is a low bar that has not been cleared in three years.

Final IRS guidance on the Notice 2023-27 look-through

Notice 2023-27 has governed since March 21, 2023 without final regulations. Guidance that put digital art clearly inside or outside section 408(m) would move the top federal long-term rate between 20% and 28% and would settle whether these can be held in an IRA.

A data-limiting soft fork with hashpower behind it

BIP-110 activated on a minority chain on August 8, 2026 with about 2.53% of hashpower and stalled inside a day. The reading that would matter is a successor proposal that miners actually signal for in the blocks they mine, over a full difficulty period, which is the only route by which any of this reaches the chain. Below that threshold, filtering is a relay-policy argument, and section 13 explains why relay policy cannot reach an inscription you already hold.

The 2028 halving, at block 1,050,000

The subsidy falls to about 225 BTC a day. At the 2026 fee run rate of 2.6 BTC a day, fees would cover roughly 1% of the security budget. Whether anything — inscriptions or otherwise — bids for that block space is the question this asset class was once claimed to answer.