Lightchain testnet is relaunching. Mainnet is untouched. Same chain id, 8200. New genesis, clean state. Testnet balances and contracts reset, the faucet refills, the explorer reindexes from block zero. Nothing on mainnet restarts, changes or needs your attention. The relaunch is the first public step of an architecture we have been building for the last eight months, and the new testnet is built to run it from genesis. Here is what we can say today. Every blockchain you have used so far makes one node do three jobs at once: decide the order of transactions, execute them, and finalize them. That is why the whole network moves at the speed of one machine. The new design pulls those three jobs apart and gives each its own engine. Ordering becomes a replicated sequencer. Your transaction is acknowledged in tens of milliseconds with a fixed place in one global sequence. It never executes anything, which is exactly why it can absorb bursts without flinching. Execution becomes independent EVM shards. Each shard is a complete chain. Each builds one block per tick, and ticks are measured in hundreds of milliseconds, not seconds. Every shard block is a pure function of public inputs: block number, timestamp, randomness, gas limit, all derived by a fixed rule with no operator discretion anywhere. That is not a detail. It is what makes the whole system provable later without a redesign. Finality stays on the chain you know. Every L1 block anchors each shard's state roots and commitments. The L1 becomes the settlement layer underneath, and inherits nothing new to execute. Value moves through escrow, never minting. A deposit locks on the L1 and lands on a shard in the next tick. An exit is proven against roots already anchored on the L1 and works even if the sequencer is gone. History is written once, content-addressed, with every hash on chain, so a shard's disk holds its state rather than its past. The AI compute marketplace moves to its own shard with a protocol rebuilt for it: fewer transactions per job, job ids you know before you send, encrypted prompts and responses kept off the chain with their hashes on it, workers that subscribe to events instead of polling the chain four times a second. And a second class of worker is coming that does not need a GPU at all. What stays: Proof of Intelligence as the useful-work layer, end-to-end encrypted prompts on the GPU tier, the validator set, the token, EVM compatibility and the tooling you already use. What changes: how far it scales, and how fast it feels. Date, genesis details and the operator guide will be posted here and in Discord before the switch. Benchmarks and the full technical write-up come when they are measured, not before.