50+ dimensions. Every product. Every benchmark. Every price point. One chart.
| Dimension | Zama | H33 |
|---|---|---|
| Company | ||
| Founded | 2020, Paris | 2025, Tampa |
| Funding | $150M+, $1B valuation | Self-funded |
| Team | ~200, ~50% PhDs | Lean |
| Focus | Blockchain FHE | Post-quantum security infrastructure |
| Open Source License | BSD-3-Clear + patent license | Apache 2.0 (no patent fee) |
| FHE Engines | ||
| FHE Schemes | 1 scheme (TFHE only) | 4 schemes: BFV-128 (exact integer), BFV-256 (NIST L5), CKKS (float/ML), BFV-32 (ultra-batch) |
| Auto Engine Selection | No — one engine, one option | FHE-IQ auto-routes to optimal engine in <500 nanoseconds |
| Hybrid Circuit Splitting | Not possible (single engine) | Auto-splits float + integer ops across CKKS + BFV |
| SIMD Batching | Limited (no batch auth) | 4,096 slots — 32 users per ciphertext, constant-time |
| Per-Operation Latency | 124,000µs (124ms for a 64-bit add) | 35.25µs per auth — 3,222x faster |
| Throughput | 189K bootstraps/sec (requires 8x NVIDIA H100 GPUs) | 2,209,429 auth/sec on CPU — 11.5x higher on 7,500x less expensive hardware |
| GPU Required | Yes — H100 ($30K+/yr) for competitive performance | No GPU. Graviton4 ARM CPU only. |
| Hardware Cost | ~$15,000/mo per coprocessor operator | ~$2/hr spot ($1,460/mo) — 10x less expensive |
| Batch Variance | Not published | ±0.71% over 120 seconds sustained (production measured) |
| Python FHE Compiler | ||
| Compiler | Concrete (@fhe.compiler) | H33-Compile (@h33.compile) |
| Backend Targets | 1 (TFHE) | 4 (BFV-128, BFV-256, CKKS, BFV-32) |
| Auto-Routing | No | Yes (FHE-IQ) |
| Hybrid Circuits | No | Yes (auto-split across engines) |
| Type System | Runtime tracing | Compile-time type annotations |
| License | BSD-3 + patent license required | Apache 2.0 |
| Built In | C++/MLIR/Python | Rust + PyO3 |
| Zero-Knowledge Proofs | ||
| ZK Type | ZKPoK (encryption correctness only) | Lookup STARK + AIR STARK + STARK-IQ auto-router |
| Proves Computation | No | Yes (AIR STARK) |
| Proves Policy Compliance | No | Yes |
| Proof Generation | 6.9 seconds | 0.059µs cached (117,000,000x faster) |
| Post-Quantum Secure | No (acknowledged) | Yes (SHA3-256 hash-based) |
| Trusted Setup | No | No |
| Auto-Routing (Lookup vs AIR) | N/A | Yes (STARK-IQ) |
| Post-Quantum Signatures | ||
| Dilithium (ML-DSA) | None | ML-DSA-65/87 (FIPS 204) |
| Kyber (ML-KEM) | None | ML-KEM-768/1024 (FIPS 203) |
| FALCON | None | FALCON-512 |
| SPHINCS+ | None | Yes |
| Nested Hybrid (3-Key) | None | Ed25519 + Dilithium + FALCON |
| Attestation Latency | N/A | 189µs per batch |
| End-to-End PQ | FHE only (lattice) | FHE + ZK + Signatures + KEM |
| Biometrics | ||
| Production Biometrics | No (research demo only) | Yes (production, batched) |
| Users per Ciphertext | 1 (demo) | 32 (SIMD batched) |
| Verification Latency | Not benchmarked | 967µs for 32 users |
| Server Sees Biometric | Yes (demo uses FastAPI) | Never |
| Template Storage | Not optimized | 256KB/user (128x reduction) |
| Blockchain & DeFi | ||
| Primary Chain | Ethereum (fhEVM) | Solana |
| Smart Contracts | fhEVM library | 10 production contracts |
| TPS | ~20 (CPU) | Solana-native |
| Biometric Wallet | No | Yes (FHE-encrypted SCIF wallet) |
| Soulbound Identity | No | Yes (NFT + biometric binding) |
| PQ Transaction Signing | No (relies on ECDSA/Ed25519) | Yes (Dilithium attestation layer) |
| Token Economics | $ZAMA (5% inflation) | H33 token (deflationary burn) |
| ML on Encrypted Data | ||
| ML Framework | Concrete ML (sklearn, PyTorch, XGBoost) | CKKS + FHE-IQ + AI-Blind + H33-Compile |
| Encrypted LLM Fine-Tuning | Yes (LoRA-based) | Via CKKS approximate arithmetic |
| Auto Engine Selection for ML | No | Yes (FHE-IQ routes to CKKS for float) |
| AI Security Agents | None | 3 native agents (0.52–1.14µs) |
| Side-Channel Protection | “Not yet implemented” | AI agent (1.14µs) |
| Security & Compliance | ||
| SOC 2 | None | In progress (100% via Drata) |
| HIPAA | None | Compliant |
| ISO 27001 | None | Audit ordered |
| HATS Certification | N/A | Author & issuing authority |
| HICS Code Standard | N/A | Open source (Apache 2.0) |
| GDPR | No compliance program | Compliant by construction (FHE, zero data exposure) |
| CCPA / CPRA | None | Compliant (PII never in plaintext) |
| NIST AI RMF | None | HATS maps to NIST AI RMF 1.0 |
| EU AI Act | None | HATS maps to EU AI Act (2024/1689) |
| NIST FIPS 203/204 | No PQ signatures | ML-KEM (FIPS 203) + ML-DSA (FIPS 204) |
| PCI DSS | None | FHE protects cardholder data in-use |
| SOX Section 404 | None | HATS audit trail supports SOX evidence |
| CNSA 2.0 | Partial (FHE lattice only) | Full PQ stack (BFV-256 at NIST L5) |
| Patent Claims | All technology patented (license required) | 144 patent claims pending |
| Security Audit | Trail of Bits (blockchain only) | Full production audit |
| Side-Channel Mitigation | “Not yet implemented” | AI-based detection (1.14µs) |
| Products | ||
| Total Products | 5 | 39 |
| Product Categories | 3 (FHE, compiler, blockchain) | 14 |
| Product List | TFHE-rs, Concrete, Concrete ML, fhEVM, Zama Hub | FHE (4 engines) · ZK (4 products) · PQC (5 sig schemes) · Biometrics (3) · Blockchain (4) · AI/ML (4) · Compliance (3) · Platform (6) · DevTools (3) · Compiler (1) · MPC (1) · Video (1) · Search (1) · Storage (1) |
| Pricing | ||
| Free Tier | Research use only | 1,000 credits + BotShield 2,500/mo |
| Commercial License | Patent negotiation required | Self-service API key |
| Per-Operation Cost | $0.005–$1.00 (blockchain) | $0.001 per auth (at scale) |
| Infrastructure | Self-hosted (~$15K/mo) | Managed API (no infra) |
| GPU Costs | ~$30K+/year per H100 node | $0 (CPU only) |
| Patent Fee | Required for any commercial deployment | None (Apache 2.0) |
| Developer Experience | ||
| API Style | Library (self-host) | REST API + SDKs + RPC (via Helius) |
| Languages | Rust, Python, C | Any (REST, SDK, RPC) |
| SDKs | Library only (no managed SDK) | Native SDKs + Solana RPC via Helius |
| Setup Time | Days (compile, configure, self-host) | Minutes (API key) |
| FHE Parameter Expertise | Required | Not required (FHE-IQ handles it) |
| Documentation | GitHub READMEs | Full docs site at h33.ai/docs |