StarkWare reported that AI‑assisted coding reduced the estimated GPU cost of preparing a quantum‑resistant Bitcoin transaction to $66, down from an earlier estimate of $320.
The original $320 figure was derived from a transaction mined in August that required roughly 3,100 hours of GPU computing across a fleet of about 100 graphics processors. That cost covered the computational work performed before the transaction was submitted to the Bitcoin blockchain, separate from the on‑chain fee.
StarkWare opened the code to a public competition, inviting participants to accelerate the search process. Many contestants employed AI coding tools, and the leaderboard shows a leading submission checking about 881 million candidates per second, compared with roughly 146 million for the baseline code on the same benchmark GPU.
The speedup translates to an estimated five‑fold reduction in compute cost. CoinDesk’s own calculations, applying the contest’s speed improvements to the $320 baseline, arrive at roughly $83 per transaction, while StarkWare cites a $66 estimate based on further performance gains observed in later runs.
The technique adds a hash‑based layer of protection that is expected to withstand attacks from future quantum computers, and it operates within Bitcoin’s existing consensus rules, requiring no protocol upgrade. However, the method must be used by sending transactions directly to miners, and it does not safeguard coins whose public keys have already been exposed.
The $66 cost remains an estimate derived from test computations; it has not yet been demonstrated in a transaction that was actually mined on the Bitcoin network.
