
Cold hardware and NFC crypto wallets that will not survive the switch to a post-quantum signature
The NIST calendar sets a deadline for the old signature. Some cold NFC cards physically cannot make the new one: the chip is small and the firmware is locked.
In October 2026 the U.S. standards institute NIST set a deadline for three families of digital signatures: ECDSA, EdDSA and RSA. After 2030 they count as deprecated. After 2035 they are not allowed in systems that must follow NIST standards. Bitcoin, ether and almost every hardware wallet sign with one of these today.
The network will not stop on the night of 1 January 2031. What stops is something else: a cold card already sold, with a small payment chip and locked firmware, cannot learn the new signature. The maker will not send an update for the old plastic. The chip cannot be swapped.
Who NIST is, and why its calendar is heard in crypto
NIST, the National Institute of Standards and Technology, is a U.S. federal institute. It writes the technical standards used by banks, clouds, browsers and auditors. Its FIPS documents and special publications answer which algorithm is still allowed.
The institute does not run Bitcoin and does not ship an Ethereum upgrade. Its weight is elsewhere. An exchange, a fund or a custodian checked against U.S. or banking rules cannot tell an insurer they will think about it later, once the standard has disallowed the algorithm. The pressure lands on cold storage, audit reports and key requirements, not on the protocol itself.
The paper is NIST IR 8547, October 2026. It uses two different words.
Deprecated after 2030 means the algorithm may still be used, but the owner of the system accepts the risk and must explain why they have not moved. Disallowed after 2035 means the algorithm is gone from systems that follow NIST standards.
The most sensitive U.S. government systems have a shorter clock. They must move key exchange to post-quantum schemes by 31 December 2030, and digital signatures by 31 December 2031. That is an internal U.S. government schedule, not the day Bitcoin stops.
NIST had planned to disallow short keys on 1 January 2031. The post-quantum transition stretched the schedule: first the status deprecated, then a full ban inside its own standards after 2035.
What ECDSA, EdDSA and RSA are
A digital signature proves that the holder of the key sent the transfer and that the amount was not changed on the way. Anyone can check it. Only the secret key can create it.
ECDSA is a signature on an elliptic curve. Ordinary Bitcoin addresses and almost all Ethereum wallets use it. The curve is called secp256k1. When a wallet asks you to confirm a transfer, it makes an ECDSA signature.
EdDSA is a close relative, also a curve, with a different formula. In daily use this is usually Ed25519: it signs SSH logins and validator votes on many networks. For a quantum computer the difference between ECDSA and EdDSA does not matter. Both live on elliptic curves.
RSA is a 1970s scheme. Its strength is that a huge number is hard to factor. Blockchains almost never sign transfers with it. Some old website certificates, email and internal bank systems still do.
Artificial intelligence does not break these. AI steals a key that already exists, or slips in a fake program. The NIST calendar does not protect against that.
A quantum computer of sufficient power does, using Shor's algorithm, published in 1994. The secret key is recovered from the public key, which the network already shows after the first spend. Making the same curve key longer does not help. The mathematics has to change.
The hash that links Bitcoin blocks is not hit in the same way. This is about the signature: the right to spend coins and the right to confirm a block.
Encryption is different again. Someone can record traffic today and decrypt it years later, when the machine exists. A signature cannot be forged backwards like that. What can be forged is a new transaction from an address whose public key is already in the chain. On an ordinary wallet the key becomes visible with the first outgoing transaction. An address that has never sent anything still shows only a hash of the key.
NIST has already approved replacements. For new systems those are ML-DSA, lattices, FIPS 204, and SLH-DSA, hashes, FIPS 205. They are larger. A signature that is tens of bytes today grows to several kilobytes. Fewer transactions fit in a block of the same size. A network cannot turn the new algorithm on with a single checkbox.
SHRINCS, described in the press as protection of Bitcoin against quantum attacks, is not this replacement. It is a Bitcoin draft: signatures from 548 to 4,619 bytes, part of the scheme remembers state, and reusing a one-time key lets an attacker forge a signature. In November 2026 the authors themselves wrote that it has not passed an independent audit and that their analysis must not be the basis for deployment.
What Satoshi described back in 2010
Satoshi Nakamoto did not miss the quantum threat. Shor's algorithm is fifteen years older than Bitcoin. In 2008–2010 there was no swappable post-quantum signature fit for money. NIST published the replacement standards in 2024. ECDSA was the ordinary strong signature of that time, not a beginner's mistake.
On 14 June 2010, on BitcoinTalk, in a thread about SHA-256 collisions, Satoshi wrote that the function is very strong, not a small step from MD5 to SHA-1, and that without a major breakthrough it can last several decades. Then two cases.
If the hash breaks suddenly, participants agree which chain was honest before the failure, lock it, and continue with a new function.
If the weakness grows gradually, the software is programmed to use a new hash after a certain block number. Everyone must upgrade by that height. The software re-hashes the old blocks with the new function and stores those fingerprints, so a different old block with the same previous hash cannot be slipped in.
There is no such block number in the protocol. A certain number means a number chosen when the threat is real. Halvings and today's chain height have nothing to do with replacing SHA-256. There has been no replacement, and none is being prepared.
On 10 July 2010 Satoshi described the signature separately. If it is compromised suddenly, perhaps by integer factorization, perhaps by quantum computers, agreeing on the last honest block is already useless: the right to spend the coins can be forged. If the break is gradual, the upgraded program, the first time it is run, re-signs the money with a stronger algorithm. It does that with a transaction that sends the coins back to their owner.
The network does not do this by itself. Bitcoin does not hold other people's secret keys and cannot re-sign their coins. First the network accepts software that understands the new signature. Then a person runs it on their own wallet. The program builds a transfer to a new address, and the old key authorizes it. Whoever does not open the wallet stays on the old signature. A cold vault, a lost key and early coins that nobody uses are not moved by the network.
Bitcoin outputs from 2009–2011, including Satoshi's own coins, are stricter than modern addresses: the public key is written into the block immediately, not hidden behind a hash until the first spend. That part of the old code is more exposed to a quantum machine than an address that has never sent anything.
Which cold cards will not make the new signature
The new ML-DSA-65 signature, the middle NIST parameter set, is a secret key of about 4 KB and a signature of about 3.3 KB. Almost any chip can store that. A normal implementation needs about 50 KB of working memory to compute it. Payment-chip coprocessors do AES, RSA and elliptic curves. They do not do lattice math.
The list is not the whole market. It includes only devices whose public documents show either a factory ban on changing the program, or working memory measured in single or low double-digit kilobytes. Unbranded cards on the same Samsung and NXP chips are built the same way, but without a chip datasheet the name does not go on the list.
Tangem, card and ring. Firmware does not change after the factory. Older batches use the Samsung S3D232A, with 11 KB of RAM. Tangem's 2026 materials name the Samsung S3B512C: 512 KB of flash and 18.5 KB of RAM, of which 16 KB is for general work. Larger chips in the same family differ in flash, not in RAM. The key fits. The signature is not computed on this chip. A card already sold cannot be taught a new firmware: the maker designed it that way.
ELLIPAL X Card. Firmware is locked at the factory. The maker writes that the card has no updates. The chip is certified CC EAL6+, and the model and memory size are not published. For a card already sold that is enough: there is nowhere to write the new signature.
Arculus, the metal card from CompoSecure. The chip is CC EAL6+, the card leaves the factory locked, and the code on it cannot be updated. Chip model and memory are not published. A sold card cannot learn the new signature.
Keycard, the ordinary sealed card. The chip is an NXP JCOP 4 P71. Flash is about 180 KB, and the signing program gets about 3 KB of RAM. The card's cryptography is AES, RSA and elliptic curves. Retail cards are sealed in production. An unsealed developer card is a different product: you can load your own code, and that does not add memory.
CoolWallet S. The chip firmware can be updated. The chip is an NXP P5CD081, with 7.5 KB of RAM, and hardware AES, RSA and elliptic curves. An update replaces the program. It does not grow the memory.
CoolWallet Pro. Wallet specification roundups name an NXP J3R110, the same Java Card line as Keycard. CoolWallet itself does not publish the memory size. If the chip is that one, there is nowhere to compute the new signature.
Ballet, Coinkite SATSCARD and Satodime. These are not a chip that signs. To spend the coins you open the card and show the key itself. There is nothing there that can learn a new signature.
Satochip. The applet code is open, and it can be loaded again onto an unsealed card. The chips are NXP J3H145 and J3R110. Open source does not add RAM. On these cards it is a few kilobytes.
SafePal S1, S1 Pro and X1 are not on this list. Their firmware is updated with a file from the website, over USB. Until April 2025 the devices used a CC EAL5+ chip; new batches use a CC EAL6+ chip, and firmware does not raise the old chip to the new level. SafePal has not published the chip model or the RAM size. The documents do not prove that an S1 or X1 cannot make the new signature. They also do not prove the opposite.
Ledger on ST33 chips, Trezor on a separate microcontroller with hundreds of kilobytes of memory, and the ELLIPAL Titan with updatable firmware are built differently. A path through a new program is still open. That is not a promise that the maker will ship in time. It is different hardware.
What follows
For the Bitcoin and Ethereum protocols, the NIST calendar is a planning deadline, not a power switch. Old signatures stay in the history of blocks. After 2035 a regulated intermediary will not want to hold new keys on ECDSA.
For the owner of a card with a locked chip the conclusion is hard. The card lives as long as the old signature does, and it can make that signature once, to send coins to an address that can make the new one. It cannot store the new key. The maker did not promise a replacement card in the mail.
For the industry the order is this. Before 2030 the new signature is added beside the old one. From 2030 to 2035 exchanges, custodians and validators move to it. After 2035 the old signature remains in the block archive, and for a new key, at anyone who lives by NIST standards, it is no longer acceptable.
Whoever opens the wallet and signs the move takes their coins across. Whoever lost the key, or never replaces a card that can make only the old signature, stays on the algorithm NIST will by then have removed from its standards.