Permissionless miner and validator program live on Sui testnet; hardware-verified scoring, on-chain rewards, and imminent mainnet launch.
VRAM AI today announced a $10 million Token Subscription Facility from GEM Digital Limited, alongside the public launch of VRAM Network — a decentralized LLM training coordination network built on the Sui blockchain. End-to-end testnet validation is complete. VRAM Network is now preparing for imminent mainnet launch on Sui.
VRAM Network makes frontier training trustless. Every gradient score is enclave-attested. Every reward is settled on-chain.”
— Sid, CEO at VRAM AI
The announcement positions VRAM AI at the intersection of two active narratives in crypto infrastructure: decentralized AI compute and the Sui ecosystem. VRAM Network brings cryptographic guarantees to distributed LLM training that existing decentralized networks do not provide: hardware-attested loss evaluation via AWS Nitro Enclaves, credential privacy via Sui Seal’s Identity-Based Encryption, and on-chain reward settlement via Sui Move smart contracts.
Also Read: CIO Influence Interview with Kyle Wickert, Field CTO at AlgoSec
“Frontier model training doesn’t require a hyperscaler, and it doesn’t require trust either,” said Sid, CEO at VRAM AI. “VRAM Network makes it trustless. Every gradient score is enclave-attested. Every reward is settled on-chain. No operator, no whitelist, no trusted coordinator.”
Verified Protocol Performance
VRAM Network has completed end-to-end testing across multiple miners and validators, running synchronized training windows on a small LLM. The full coordination pipeline runs in pure Rust: gradient compression, Seal IBE credential decryption, Nitro Enclave loss evaluation, OpenSkill peer ranking, and on-chain reward distribution. The top-performing miner captured 5.5% of reward weight based on verified gradient scoring and enclave-attested loss evaluation.
The system supports pluggable training adapters. Miners running existing distributed-training infrastructure connect without rewriting their training stack.
How It Works
VRAM Network operates in synchronized 10-minute windows. Each window, miners train on a data subset derived from a per-miner-per-window SHA256 seed, compress gradients using DCT and top-k sparsification, and upload to storage with read credentials encrypted on-chain via Sui Seal IBE. Only staked validators decrypt them. No trusted third party holds keys.
Validators download miner gradients, send them to an AWS Nitro Enclave running the Nautilus evaluator, and receive Ed25519-signed loss improvement scores. These scores cannot be forged or manipulated outside the hardware boundary. Signed scores go to Sui’s score_ledger.move contract, which verifies the enclave signature cheaply. Attestation was verified once at enclave registration, not every window. Rewards flow from reward_distributor.move proportional to each miner’s OpenSkill Bayesian peer rating, a probabilistic ranking model that resists collusion.
GEM Digital Facility
The $10 million Token Subscription Facility provides VRAM AI with structured capital, from a long-only investor, to scale the miner network, complete mainnet deployment, and expand the validator set. GEM Digital has previously backed decentralized infrastructure projects, and operates across more than 30 exchanges globally.
Catch more CIO Insights: The CIO as a Value Creator: Moving Beyond Cost Centers to Revenue Drivers
[To share your insights with us, please write to psen@itechseries.com ]
