KoinlyticsAI and crypto share a set of properties: both are permissionless, both benefit from distributed hardware, both have valuable outputs that need trust or verification. The overlap is real. So is the amount of slop with an "AI" prefix and no substance.
Real overlap: three things
- Decentralized compute markets. Akash, io.net, Render. Rent GPUs peer-to-peer, priced in tokens. Useful for hobbyists and startups undercutting AWS spot pricing. Real usage, real payments.
- Verifiable inference. ZKML projects (Modulus, Giza, EZKL) prove that a specific model was run correctly on specific inputs. Useful for on-chain applications that consume off-chain AI outputs (oracle-style).
- Agent commerce. Autonomous agents that hold wallets, transact, negotiate on-chain. Coinbase's AgentKit, Virtuals Protocol, Story Protocol. Early stage but shipping.
Marketing plays
- Tokens tied to "AI-powered trading bots" that are actually rebranded moving-average strategies.
- "AI" NFT projects with no model, just gpt4 prompts on generation.
- "DePIN for AI" projects with 12 real nodes and a $200M FDV.
Where the real capital flows
By TVL and revenue, decentralized compute is the largest real-usage sub-sector: Akash renders 30k+ concurrent workloads, io.net processes hundreds of thousands of GPU-hours monthly. Bittensor's subnets host actual model training and inference tasks and pay out real TAO tokens.
ZKML: the future many are betting on
Proving that a model gave a specific output enables on-chain markets to trust AI outputs: prediction markets settling on GPT-x consensus, oracles powered by AI reads of off-chain data, dApps that gate access based on user actions verified by an AI classifier. Currently expensive but proving costs are collapsing.
Agent economies
The most interesting long-term thesis: agents that autonomously earn, spend, and coordinate on-chain. USDC as their unit of account, MCP as their tool interface, on-chain identity as their reputation. Not there yet. But no other rail is even close.