SODALABS

Powered by garbled-circuit MPC.

Sensitive data is split across independent nodes; no single node ever sees meaningful information, and no trusted third party is required as computations run directly on encrypted data via secret sharing.

SODA LABS · RESEARCHThe gcVM:a Virtual Machine forConfidential ExecutionSoda LabsAbstract1
The cryptographic construction behind Bubble

The gcVM: a Virtual Machine for Confidential Execution

The full construction, published in the open: how arbitrary programs run over ciphertext with no trusted third party. Read it, verify it, cite it.

Technical paper · 17 pages · PDF
How it works

Four layers, zero plaintext.

  1. 01

    GC-MPC engine

    Computation runs directly on encrypted data via garbled circuits and secret sharing. Sensitive data is split across independent nodes; no single node ever sees meaningful information, and no trusted third party is required.

  2. 02

    Quantum-resistant AES-256

    Inputs are encrypted client-side with AES-256: standards-based, battle-tested, and resistant to quantum attacks.

  3. 03

    The gcVM

    A virtual machine for confidential execution: arbitrary programs run over ciphertext, so any financial logic (transfers, auctions, rebalancing) can be made confidential.

    Read the gcVM paper →
  4. 04

    Selective disclosure

    Data is never decrypted during computation; the only disclosure path is the on-chain access list (ACL), through which authorized entities can request scoped decryption. Confidentiality for the market, transparency for oversight.

CONFIDENTIAL EXECUTION
Encrypted inputAES-256
▼ never decrypted ▼
GC-MPC computationgcVM
▼ still encrypted ▼
Encrypted settlementon-chain
▼ selective disclosure ▼
Auditor viewACL grant
AES-256
Quantum-resistant, standards-based encryption
GC-MPC
No trusted third party, no single point of failure
10+
L1s & L2s supported, plus cross-chain
0
Plaintext in computation. Decryption only by access-list approval
Benchmarks

Built for production. Proven at scale.

The Soda Labs stack is independently audited by Hacken, protected by three granted US patents, and runs in production as the privacy layer of COTI mainnet with 125,000,000+ confidential transactions processed.

Measured end-to-end
500cTPS sustained

Peaking at ~750, over 15,000 private ERC-20 transfers on Arbitrum, July 2026. Methodology and raw results on request.

No other confidential-transaction system publishes an end-to-end figure measured on a live chain.

Compute only
3,400cTPS per CPU core

Analytical ceiling on a commodity CPU core, before parallelization. Parallelized across cores, that projects past 30,000 cTPS.

For scale: the fastest published FHE figure is 1,040 tx/s on a node of eight NVIDIA H100 GPUs, which its authors describe as compute throughput rather than settlement on a live chain.

Measured end-to-end on Arbitrum over 15,000 confidential transfers (July 2026): 500 cTPS sustained, 750 cTPS peak, at $0.14 per million transfers.

A practical comparison of garbled-circuit MPC used by Soda Bubble network versus FHE used in Zama network.

CapabilitySoda Bubble (GC-MPC)FHE (Zama and others)
Encryption
Encrypted amounts (arguments)YesYes
Encrypted addresses (anonymity support)YesNo
Encryption typeBattle-tested AESTFHE
Encryption adoptionAES: the worldwide standard (internet, banking, government)Early: no large-scale production deployments yet
Performance
LatencyNear zero – HTTPS equivalentHigh latency – requires expensive client-side ZK proofs
Computation speedFast: near-native circuit evaluationSlow: encrypted operations take orders of magnitude longer
Throughput500 cTPS sustained, 750 peak (measured end-to-end)20 tx/s in production (vendor-published, CPU)
Cost per transferEssentially $0 ($0.14 per million transfers, measured)A ~$10,000/month GPU operator before the first transfer
Performant on CPUYesNo, relies on GPU or ASIC
Compliance & compatibility
Full EVM/SVM compatibilityYesNo (doesn't support 256-bit and hash operations)

Detailed capability comparison: see the Privacy Hub taxonomy for a neutral summary.

The landscape

How Bubble compares.

GC-MPC vs FHE, ZK, and TEE: an honest, side-by-side look at latency, trust assumptions, and best-fit workloads.

See the taxonomy

Put GC-MPC to work.

Walk through the architecture with our engineers and see it running on your use case.