September 2026

ㅤ

Authored by

Arcana

@0x_Arcana

Abstract

Proof-of-work has historically been the standard consensus mechanism in blockchains, turning computation into economic security – a model many would argue built around “useless” random hashing. At the same time, a separate compute economy has emerged around AI, where hardware is used to train models and serve inference. The concept of Proof-of-Useful-Work (PoUW) was born from the question on whether proof-of-work consensus could be implemented on top of real-world computation.

The idea is simple. Instead of random hashing, miners could run AI models, generate proofs, or perform matrix multiplication. The same computation would produce two outputs: a useful result for a consumer and a claim on the network’s block reward. One system, two purposes.

But what makes proof-of-work effective is also what makes it hard to replace. Hashing is standardized, always available, easy to verify, and tightly linked to the chain. Real-world computation is none of these by default. Jobs vary, demand fluctuates, and outputs can be costly to verify. A protocol can prove work was done, but not that it was needed.

With the recent launch of Pearl Network and the growing attention around Nockchain, there has been some renewed interest around Proof-of-Useful-Work. New designs show that useful computation, specifically matrix multiplication, can be adapted into proof-of-work with minimal overhead. This opens a path toward reusing AI computation for consensus.

The question now, is no longer whether useful computation can secure a network, but whether it can do so in a way that remains economically and structurally coherent. Can computation become consensus? Possibly. Whether it can do so while remaining useful, secure, and economically sustainable is the real question we explore in this report.

PoUW: Can Computation Become Consensus?

Claim your free report