Inference Labs has announced a significant advancement in zkML (zero-knowledge machine learning) by focusing on system architecture rather than cryptography. Utilizing DSperse distributed slicing, large models are broken into smaller, parallel-verifiable segments, complemented by the JSTprove lightweight proof system. This innovation has resulted in a 65% increase in proof speed and reduced memory usage to less than 1GB. As of August 2025, over 281 million zkML proofs have been processed, marking a shift from theoretical to production-level applications. This development positions zkML as a scalable, verifiable AI infrastructure ready for real-world deployment.