Wiki › Economy, compute and governance

Decentralized AI Compute

Open-protocol approach to AI compute where participants provide hardware and earn tokens, aiming to make inference a public competitive layer rather than a closed cloud monopoly.

Decentralized AI compute is the attempt to build AI infrastructure as an open protocol rather than a company-owned cloud. In the Liberman brothers' framing, the analogy is Bitcoin: participants contribute hardware, the protocol rewards useful contribution, and the network becomes a public layer that no board, company, or state can unilaterally close.¹

The proposed mechanism is not classic venture investment. A participant can rent or contribute GPUs, mine protocol tokens, or buy tokens from miners. The token is meant to represent future access to compute: use it now when the network is small, or hold it in expectation that the network becomes larger and therefore buys more useful computation later.¹

The strongest argument for this model is price discipline. If AI Compute Infrastructure remains concentrated in a few corporate clouds, providers can bundle access, exclude open-source agent frameworks, and price AI by customer value rather than marginal compute cost. A credible decentralized alternative would give agents, developers, governments, and users another route to inference and make it harder for closed providers to raise prices or cut off ecosystems.¹

The hard part is that AI compute is not Bitcoin hashing. Verification quality, latency, privacy, model compatibility, scheduling, enterprise reliability, and distributed training/inference are much more complex than checking a proof-of-work hash. Decentralized AI compute therefore has to prove not only ideological appeal but also reliability, cost, latency, and security against centralized clouds.¹

This page is marked contested because the interview presents a strong pro-decentralization thesis, while several implementation risks remain unresolved: hardware can still concentrate, token incentives can become speculative, and centralized data centers may retain major advantages in reliability and energy economics.¹