Digital rails are often described in terms of speed alone, as if the only thing that mattered were the number of seconds between two timestamps. The honest reading is narrower and more useful: digital rails change which paths are available, and they make some operations faster — but they do not make every payment instant.

The gap they address

The problem is real. The number of active correspondent banks worldwide fell by about 22% between 2011 and 2019, even as payment volumes rose. In Latin America, active relationships fell by about 30% since 2012. Fewer corridors, more traffic: that is the pressure digital rails are answering, and it explains why the G20 set cross-border payment targets in 2021 — an average retail cost no higher than 1%, no corridor above 3%, and 75% of payments within an hour, mostly by end-2027. The FSB's 2025 report shows limited progress: in 2025 around 35% of payments arrived within an hour.

Speed is conditional

Practical speed depends on the corridor, the currencies, the counterparties and the compliance steps involved. A digital rail can shorten the settlement leg while leaving onboarding and validation unchanged. Where those steps are slow, the operation remains slow regardless of the rail underneath.

Where the win actually sits

The larger gain is architectural. When approved operations enter a single order book, offsetting flows cancel internally and only the remainder is routed out. If bank A sends €10m and participants B and C receive €8.5m in total, only €1.5m needs to leave the hub at all. AI routing then selects the best remaining path — hub participants, card networks, local instant payments, SWIFT, stablecoins or agent protocols — by cost, settlement time, reliability and regulatory fit.

In short

  • Digital rails change available paths; they do not make everything instant.
  • Correspondent relationships have been shrinking while volumes grew.
  • G20 targets exist and progress so far is limited.
  • Netting can reduce how much value leaves the network at all.
  • Routing, not any single rail, decides the practical outcome.

See how the layers produce that outcome: /protocol/