Inference Labs

Inference Labs is a Web3 AI infrastructure project focused on verifiable machine learning (zkML). Its core business revolves around two proprietary frameworks: DSperse, a distributed inference system that slices AI models into smaller sub-models for parallel processing, and JSTprove, an in-house zero-knowledge proof engine optimized for AI operations using sumcheck and GKR protocols. The company's primary value proposition is enabling cryptographic verification of AI inference outputs without revealing the underlying model weights, addressing the dual challenges of trust and intellectual property protection in decentralized AI networks.

The market pain points Inference Labs tackles include the opacity of AI decision-making (users cannot verify whether outputs come from the claimed model), the computational infeasibility of proving large model inferences in zero-knowledge, and the tension between model privacy and verifiability. By converting AI computations into verifiable circuits and distributing the proving workload across nodes, the project reduces the barrier to trustless AI services.

In the past six months (2026-02-28 to 2026-08-31), Inference Labs has deployed all base models on its subnet, achieving proof coverage exceeding 30% per model. The DSperse and JSTprove frameworks have moved from research to production, with the team reporting successful parallel proof generation across distributed nodes. However, the project still faces significant technical limitations: despite model slicing, individual layer circuitization remains memory-intensive, and the proving overhead for large vision models (e.g., YoloV8) still exceeds consumer hardware capabilities. The team acknowledges that while current optimizations have improved efficiency, the system is not yet suitable for real-time applications requiring sub-second verification latency.