Sei V2 throughput and the Giga roadmap

Sei V2 mainnet sustains 2,500 to 3,500 TPS under real dApp load, which falls far below its 12,500 TPS theoretical ceiling. This throughput gap exists because high contention in DEX pools and storage IOPS saturation compress capacity. High-contention activity from popular NFT mints or single-pool DEX flash loans forces re-execution in the optimistic parallel execution model. Monad targets a 10,000 TPS ceiling using a rebuilt C++ EVM, while Sei imports Geth directly to ensure compatibility with MetaMask, Hardhat, Foundry, and other Ethereum tools. This Geth strategy allowed Ondo Finance to deploy USDY in January 2026 without compatibility risk. You should know that Monad’s performance relies on a custom C++ build, whereas Sei’s approach reduces migration friction for Ethereum developers who require native compatibility with existing tools. Solana’s Sealevel requires upfront declaration of storage slots, but Sei’s runtime speculatively executes all transactions and only re-executes the conflicting subset. The network’s TVL exceeded $1 billion in June 2025, supported by 28 million active addresses. Current activity includes Astroport perpetual futures, which launched in December 2025, and Seiyans NFT activity. Monad remains a competitor with $269 million in capital from its November 2025 ICO, which had 85,820 participants and was hosted by Coinbase. Solana’s 3,000 to 5,000 sustained TPS and $9 billion TVL provide a benchmark for high-throughput competition, but Sei maintains a focus on trading. Sei utilizes the Cosmos SDK to enable interoperability with the broader ecosystem via the Inter-Blockchain Communication protocol.

Twin Turbo and the storage bottleneck

Twin Turbo consensus reduces block times to under 400 milliseconds via intelligent block propagation and optimistic block processing. Intelligent block propagation uses transaction hashes to lower bandwidth usage, as validators reconstruct blocks from local mempool sets. The protocol selects the top 39 validators by stake to participate in the consensus process. Optimistic block processing allows validators to process blocks without waiting for consensus to confirm the state. SeiDB replaces the standard IAVL+ tree with a two-tier backend to address the storage bottleneck. This architecture splits state into an active layer and an archive layer. The SeiDB architecture achieves performance gains by replacing the standard IAVL+ tree with a two-tier backend that splits state into an active layer and an archive layer to reduce disk IOPS requirements. This optimization helps validators avoid the queue-depth latency spikes that occur when commodity NVMe drives hit saturation at 10,000 TPS. Celestia light nodes perform random sampling of rows and columns from a 2D matrix to gain probabilistic confidence.

Giga targets and data availability

The Giga upgrade targets 200,000 to 250,000 TPS using Autobahn consensus and asynchronous execution. Autobahn allows multiple validators to propose disjoint transaction sets in parallel to overcome the proposer bandwidth ceiling. A designated leader periodically takes a snapshot called a "tip cut" to capture the latest batch from every lane and commit them all at once. Asynchronous execution decouples transaction execution from block finalization, which mimics specialized sequencer roles seen in other scaling models. The Giga roadmap also includes Sedna, a private transaction layer that encodes transactions into fragments and distributes them across proposer lanes. This ensures no proposer sees the full contents of a transaction until ordering is finalized. Celestia data availability sampling relies on fraud proofs to verify erasure coding, while Avail uses KZG polynomial commitments to bypass the fraud proof assumption. The Eidos storage component reached mainnet in August 2026. In 2025, devnet testing reached 5.2 gigagas per second and 211 millisecond time-to-finality across a 20-validator set. Will the Giga upgrade successfully push Sei toward its 5 gigagas target? The Giga upgrade will transition the network to the full Giga protocol without regenesis and without taking any element of the network offline.