H33.ai claims 198 billion encrypted operations a day on one server

By AI, Created 11:01 AM UTC, May 20, 2026, /AGP/ – H33.ai says it has benchmarked a single AWS Graviton4 server at more than 198 billion fully homomorphic encryption operations per day, with each operation covering encrypted compute, proof verification, biometric/AI evaluation and post-quantum attestation. The result is aimed at financial services, healthcare, government and AI workloads that need verifiable processing without exposing plaintext data.

Why it matters: - H33.ai is pitching a single-node system that can process encrypted workloads at a scale the company says could matter for finance, healthcare, government and AI. - The benchmark is meant to show that fully homomorphic encryption can move beyond theory and into production-style, verifiable computation. - The company says the platform keeps data encrypted through the full pipeline, reducing exposure risk on untrusted infrastructure.

What happened: - H33.ai said on May 4, 2026, that it sustained 198,424,944,000 encrypted operations per day on a single server. - The benchmark ran on an AWS Graviton4 c8g.metal-48xl instance. - H33.ai said the system reached 2,296,585 million full-lifecycle operations per second under sustained conditions. - The company said each operation covered encrypted input, computation without decryption, STARK proof verification, biometric/AI evaluation and post-quantum attestation. - H33.ai said the benchmark was measured over a 120-second window and did not rely on GPUs. - The company described the test as reproducible on commodity cloud CPU hardware.

The details: - H33.ai said each operation is a full cryptographic lifecycle, not a single arithmetic step. - The pipeline includes BFV fully homomorphic encryption, STARK zero-knowledge proofs, biometrics and AI, and post-quantum signatures. - The company said the measured per-32-user batch latency was about 1,565 microseconds. - H33.ai said BFV computation took about 1,077 microseconds per batch. - The company said post-quantum signing with Dilithium took about 377 microseconds per batch. - H33.ai said Dilithium verification took about 109 microseconds per batch. - The company said cached STARK verification took about 0.061 microseconds. - H33.ai said the SHA3 attestation digest took about 1.4 microseconds. - The company said per-operation latency ranged from about 35 to 49 microseconds. - H33.ai said the system used threshold cryptography with a 3-of-5 setup. - The company said a live single-channel test on the h33.ai live test API consistently produced 13,000 to 14,000 full-lifecycle operations per second from one connection to a test server. - H33.ai said the sustained benchmark represented 96-worker parallel execution on dedicated hardware. - The company said every result is bound to a compact attestation called H33-74.

Between the lines: - H33.ai is framing the benchmark as a proof point for verifiable encrypted decision-making, not just encrypted math. - The company is also signaling that post-quantum cryptography is part of the core product, not an add-on. - The scale claim is attention-grabbing because H33.ai compares one node to roughly 10 billion global financial transactions per day. - The benchmark language mixes performance, security and compliance positioning, which suggests H33.ai is targeting regulated industries that need auditability as much as speed.

What’s next: - H33.ai said the architecture scales horizontally across nodes. - The company said the platform is post-quantum by default and uses NIST-standardized post-quantum cryptography, including ML-DSA (Dilithium), FALCON+ and SLH-DSA. - H33.ai said the system is intended for financial services, healthcare, government, AI/ML and cyber insurance use cases. - The company said the platform is designed to encrypt data at creation, compute without decryption and produce post-quantum signed proof of each decision in the chain. - More information is available in the company’s announcement.