MPC
Multi-Party Computation. A cryptographic technique where multiple parties jointly compute a function over their inputs without any party revealing its input to the others. Useful for shared computation on private data.
Also known as: multi-party computation, secure MPC
MPC is one of the older privacy-preserving cryptographic techniques, with roots going back to the 1980s. The core idea is that multiple parties can collectively compute a function over their combined inputs without ever sharing their inputs with each other. Each party only sees the final output, not the intermediate values or the other participants’ inputs. This solves a class of problems where you want collaborative computation but the participants don’t trust each other enough to reveal raw data.
The classic application is threshold signing: instead of one private key controlling a wallet, the key is mathematically split among several parties using MPC, and any threshold (say, 3 out of 5) can collectively sign transactions without ever reconstructing the full key. This is more secure than a single private key because compromising any individual party doesn’t compromise the wallet. Custodians like Fireblocks and Copper use MPC for institutional wallet security. NEAR’s Chain Signatures use MPC threshold signatures to let NEAR accounts sign transactions on Bitcoin, Ethereum, and other chains.
In DeAI, MPC powers projects like Nillion and Partisia. Nillion’s network lets clients submit private inputs that get computed on by a network of independent MPC nodes, and the output is returned without any single node ever seeing the raw input. The use case is private inference, secure data sharing for AI training, and federated computation across organisational boundaries that don’t trust each other. The performance overhead is significant (10-1000x slower than non-private computation, depending on the protocol) which is why MPC tends to be used for low-throughput high-value computations rather than mass inference.
The honest comparison to TEEs and ZK is that MPC doesn’t require trusting any specific hardware vendor and doesn’t require generating expensive proofs. Its cost is in network communication: every step of the computation requires multiple parties to exchange messages, which adds latency. For applications where the privacy guarantee is more important than throughput (signing, custody, sensitive analytics), MPC is often the right choice. For high-throughput AI inference, TEEs are still more practical today.