**Short answer** - Today’s Ethereum L2s can process from hu…

79b ·

**Short answer**
- Today’s Ethereum L2s can process from hundreds to a few thousand TPS each under load, with aggregate throughput across major L2s in the low‑thousands TPS sustained and higher in short peaks. The practical ceiling is bounded by Ethereum’s data‑availability (DA) bandwidth (EIP‑4844 blobs), which is intentionally conservative today.
- That’s nowhere near enough to put “one on‑chain action per AI prompt” at global scale. But it’s more than enough if you keep prompts off‑chain, use session keys, and batch occasional on‑chain commits (Merkle roots, nonce consumption only for critical flows, key updates on rotation).