My Honest Take on ZKML: Impressive Tech, Unclear Revenue

Frequently Asked Questions

01.01. What does ZKML actually do for a business founder?

ZKML lets a business prove that an AI model produced a specific output without revealing the model's weights, the input data, or any proprietary logic. For founders, this means you can offer verifiable AI decisions : loan approvals, risk scores, health screening results : to third parties who need cryptographic assurance rather than just trust in your brand.

02.02. Is ZKML production-ready for a business with a 2026 launch timeline?

For small-to-medium models in DeFi risk scoring and compliance reporting, yes : frameworks like EZKL are shipping production integrations. For complex transformers or large language models, the proof generation overhead is still between 10,000x and 100,000x standard inference time. A 2026 launch is feasible for narrow, well-scoped models; a full-scale LLM integration is 12 to 24 months away from practical deployment.

03.03. Who pays for ZK proof generation, and how does that fit a business model?

Proof generation is typically priced per inference request and runs on GPU clusters. Today, the cost ranges from fractions of a cent for tiny models to several dollars per proof for medium-complexity networks. The sustainable business model requires charging a verifiability premium to end users : regulators, institutional counterparties, or compliance-obligated clients : who are willing to pay more for provable outputs versus standard black-box AI.