Where AI and Crypto Actually Intersect

A category map of the AI × crypto stack: data, compute, training, inference, agents, and verifiable ML — and which categories have real product vs. which are mostly narrative.

30 min · expert · part of AI × Crypto: The Convergence

Two Hype Waves That Actually Overlap

Crypto and artificial intelligence are the two biggest narrative engines of the 2020s. They are also frequently described together in ways that are mostly cosmetic — slapping "AI" on a tokenized product to lift the chart, or slapping "blockchain" on an AI startup deck to widen the addressable market. A lot of "AI × crypto" projects are, by honest assessment, vapor. But the overlap is not entirely fake. There are several places where the two genuinely need each other, where the properties of public-blockchain coordination (open token economies, permissionless participation, programmable settlement, censorship resistance) actually solve real problems for AI systems — and where AI systems in turn solve real problems for crypto-native applications. This lesson maps the categories so the rest of the module makes sense. There are six broad slices: **data networks** (compensating people for the inputs that train models), **decentralized compute** (peer-to-peer GPU rental), **training markets** (paying networks of strangers to improve models), **inference networks** (paying networks of strangers to serve model outputs onchain), **agent infrastructure** (autonomous software actors that hold their own crypto wallets), and **verifiable machine learning** (cryptographic proofs that a particular model produced a particular output). Each slice has both legitimate technical work and embarrassing grift. Part of the goal of this module is to give you the vocabulary to tell them apart.

Also in this lesson

Data Networks: Paying for the Inputs
Decentralized Compute: GPU Rental Markets
Training Markets and Inference Networks
Agents and Verifiable ML
The Grift Filter

Key terms

Foundation model: A large-scale machine learning model (e.g., GPT-4o, Claude Opus 4.7, Gemini 2.5, Llama 4) trained on broad data, intended to be adapted to many downstream tasks rather than serving one narrow purpose.
Decentralized compute: A token-coordinated marketplace that aggregates GPU or CPU capacity from independent providers and sells access to it, competing with hyperscaler clouds. Examples: Akash, io.net, Render, Aethir.
Bittensor (TAO): A protocol coordinating 64+ specialized subnets, each running an ML task; miners produce outputs, validators score them, and the Yuma consensus mechanism distributes TAO token emissions accordingly.
Agent wallet: A crypto wallet held and operated by an AI agent — typically an LLM with tools — that pays for services, holds tokens, and transacts without constant human approval. Key custody is often delegated to a custodial service like Privy or Coinbase CDP.
Verifiable ML: A class of techniques that cryptographically prove a specific machine-learning output was produced by a specific model with specific inputs. Implemented via zkML (SNARK/STARK proofs) or opML (optimistic challenge windows).
zkML: Zero-knowledge machine learning. Uses zero-knowledge proofs (typically Halo2 SNARKs or StarkNet STARKs) to verify the correctness of an ML inference cryptographically. Has 1000x-10000x performance overhead today.
opML: Optimistic machine learning. Assumes an ML output is honest unless challenged during a window, similar to optimistic rollups. Faster than zkML but introduces a trust/challenge model.
Data DAO: A user-owned cooperative that pools its members' data (e.g., browser history, ChatGPT exports, fitness data) and tokenizes access to it. Vana is the highest-profile example.
Yuma consensus: Bittensor's mechanism for distributing TAO rewards. Validators score miners' outputs; the Yuma algorithm weights miners by the consensus of all validators, rewarding both honest miners and honest validators.
Eliza framework: An open-source agent framework backed by the ai16z DAO, used to build AI agents that hold wallets, post to social media, and trade onchain. Released 2024.

Continue this lesson — 5 more sections in the CryptoBipto app.

Open lesson

Educational only — not financial advice.