PulseVM claims 4–5μs actions

A blockchain action is the tiny unit of work a smart contract does, like “move $10,” “update a balance,” or “match this order.” Marshall Hayner said PulseVM is doing those actions in 4 to 5 microseconds inside the virtual machine, which is about 0.004 to 0.005 milliseconds. (x.com) A virtual machine is the software engine that reads contract code and executes it step by step, like the processor inside a game console reading a cartridge. PulseVM’s public code says it is “built for Metal Blockchain based on the XPR Network protocol,” which puts this claim at the execution-engine layer, not at the full network round trip a user sees in a wallet. (github.com) That distinction matters because blockchains have at least two clocks. One clock is how fast the machine can execute one action, and the other clock is how fast the network can package actions into blocks, get validators to agree, and mark them final. (github.com) PulseVM’s own repository says it charges a baseline of 50 microseconds of central processing unit time per action, and it says blocks are checked every 500 milliseconds with consensus taking around 200 milliseconds. Those numbers mean the 4 to 5 microsecond figure is much smaller than the block-production and finality numbers described in the same codebase. (github.com) Metal Blockchain says its A-Chain, short for the XPR Network chain inside Metal’s four-chain design, is the lane optimized for payments and decentralized finance. The docs also say XPR Network uses WebAssembly, which is a compact format that lets contract code run closer to native machine speed than older blockchain designs. (docs.metalblockchain.org) Metal’s docs describe the broader system as a layer-zero network with subnets, and they say each subnet can process 4,500 transactions per second. That means PulseVM is being pitched as one component inside a larger architecture that is already designed around separate chains and specialized workloads. (docs.metalblockchain.org) The comparison to Ethereum-style chains is really a comparison of execution models. Ethereum Virtual Machine chains usually measure user-visible speed in blocks and confirmations, while Ethereum’s own roadmap page still says a block takes about 15 minutes to finalize on mainnet today. (ethereum.org) The comparison to Solana is trickier because Solana is already fast by normal blockchain standards. Solana’s developer docs describe transactions as bundles of instructions, and third-party Solana infrastructure guides commonly put block time around 400 milliseconds and full finality around 12 to 13 seconds, which is still many orders of magnitude above a 4 to 5 microsecond execution claim. (solana.com, sanctum.so) If the 4 to 5 microsecond number survives outside a benchmark and holds under production load, the obvious target is finance systems that care about massive bursts of small operations, like order books, internal settlement, and payment routing. Metallicus has been aiming at that world for a while: its 2025 report said 12 credit unions were already in a stablecoin pilot and called A-Chain “a virtual machine precisely designed for banking.” (cdn.metalpay.com) The catch is that a virtual machine benchmark is not the same thing as end-to-end throughput on a live chain. Until there are public production metrics for full transactions, validator performance, failed-state handling, and sustained load, the 4 to 5 microsecond figure should be read as an execution-engine claim pointing at what the architecture might do, not yet a final verdict on what users will feel. (github.com, docs.metalblockchain.org)