Cardano’s Hydra scaling: five throughput myths debunked

Cardano’s Layer 1 processes 0.29 to 0.34 transactions per second on average, with a measured maximum of 11.62 transactions per second over the last 100 blocks. This baseline remains the fundamental reality for the network. Two governance actions in 2025 and 2026 raised the Plutus maxTxExecutionUnits memory parameter from 14,000,000 to 17,500,000 to increase computational headroom for smart contracts. This change does not change the order of magnitude for the base layer. Hydra provides a way to process transactions off-chain through state channels. A Hydra Head uses a BFT-style consensus where every participant agrees on every transaction before commitment. This requires a static, pre-configured peer topology.

The 1 million TPS myth

A single 2020 IOG blog post suggested that running 1,000 Hydra Heads in parallel could achieve a theoretical aggregate of 1,000,000 transactions per second. This figure refers to 1,000 isolated groups running their own channels, not a single network throughput capacity. Matthias Benkort, the former Cardano Foundation Technical Director, called the 1 million TPS narrative misleading. During the Hydra Doom Tournament in late 2024 and early 2025, Charles Hoskinson said the network crossed the 1 million TPS mark globally. Analysts noted those transactions never settled on Cardano mainnet because it was a testnet demo simulation. The Hydra FAQ now states that summing TPS across independent heads isn’t accurate.

Hydra delivers higher throughput than the Layer 1 capacity by running in an off-chain mini-ledger. DeltaDeFi uses Hydra for a perpetuals DEX with restricted deposits, and Masumi uses it for agent-to-agent commerce. Each Head’s ceiling depends on the hardware running it.

Hydra vs Mithril and Leios

Mithril helps node operators and light clients by using a stake-based threshold multi-signature. Stake pool operators sign a snapshot, and an aggregator combines these into one compact certificate. A Mithril client can restore a full mainnet node in under 20 minutes, whereas the previous process took days. A mainnet upgrade on 23 March 2026 decentralized light-client verification and cut average light-client bandwidth by 40%. Mithril does not increase transaction throughput per second.

Leios plans to split block production into two streams to increase Layer 1 capacity. Endorser Blocks, produced every 5 seconds, bundle transaction data, while Ranking Blocks, produced every 20 seconds, anchor the ordering. IOG projects Leios could reach 1,500+ TPS, representing a 30 to 65x improvement over current baseline levels. The Musashi Dojo public testnet launched on 23 June 2026 across five stages.

Midgard, an optimistic rollup from Anastasia Labs, uses fraud proofs for validation. Anastasia Labs reported a functional node for Midgard as of September 2025.

Is Hydra a general purpose solution?

Hydra operates as an isomorphic state channel, meaning it uses the same transaction format and Plutus script rules as the main chain. A contract that works on Layer 1 works inside a Head without modification. However, a Head cannot finalize while it holds minted-but-unburned native tokens, and it lacks support for Byron-era addresses. Hydra also has bounded tolerance for offline peers. These limitations exist because Hydra manages a known group of counterparties rather than an open, permissionless set of validators.

Does the current participant limit in a Head prevent widespread adoption? Hydra provides low latency for specific use cases like gaming or micropayments. It is not a substitute for open, permissionless scaling. A thousand closed rooms running fast does not make the building’s one front door any wider.