Finality speed only matters after you've solved decentralization and censorship resistance. A 400ms-final chain that 5 companies can collectively halt or censor is a fast federation, not a settlement layer.
Solana, TON, Sui, Algorand cannot function as a global financial OS.
Sub-second finality is bought with a 33% halt threshold concentrated in professional, jurisdictionally identifiable validators. That's not a tradeoff planetary-scale neutral money can make.
Look at how nations vote in the UN. Getting 2/3 of independent sovereign actors to agree on anything is practically impossible. Same arithmetic blocks any BFT chain from being decentralized at planetary scale.
True scalability isn't TPS or finality latency. It's surviving 33% of your validators getting subpoenaed, sanctioned, or jailed.
Two architectures do that:
1. Pure Nakamoto (Bitcoin, Cardano today): degrades smoothly at 33%, recovers without governance.
2. Nakamoto base additive BFT finality liquid stake: fast when validators agree, falls back to Nakamoto when they don't. Cardano's Peras is designed for exactly this.
Speed without that floor is a feature for cooperative environments. Useful, but not a global financial OS.
*Algorand leads Layer-1 blockchains in finality time 🏁