VRAM AI raises $10M from GEM Digital to launch decentralised LLM training network on Sui
What's the deal? Hong Kong-based VRAM AI has secured a $10 million token subscription facility from GEM DigitalDealroom has a profile for this one. Try Dealroom → Limited to launch VRAM Network — a decentralised coordination layer for training large language models, built on the SuiDealroom has a profile for this one. Try Dealroom → blockchain. The facility, from a long-only investor, will fund miner network expansion, mainnet deployment, and validator growth.
VRAM Network lets distributed miners train LLMs without a central coordinator. Gradient scores are verified inside AWS Nitro Enclaves, credentials are encrypted on-chain via Sui Seal, and rewards are settled through Sui Move smart contracts. The project has completed end-to-end testnet validation and says mainnet launch is imminent.
"Frontier model training doesn't require a hyperscaler, and it doesn't require trust either," said Sid, chief executive officer at VRAM AI.
Why now? The project sits at the intersection of two hot crypto infrastructure narratives: decentralised AI compute and the growing Sui ecosystem. Demand for GPU capacity to train frontier models has exploded, yet access remains concentrated among a handful of hyperscalers. At the same time, Sui has been attracting builders with its Move-based smart contract environment and privacy tooling like Sui Seal.
VRAM AI's timing also reflects broader market appetite for AI-adjacent token projects, with GEM Digital — which operates across more than 30 exchanges globally — providing structured capital to help the network scale before mainnet goes live.
What could go wrong? Token subscription facilities are not the same as traditional equity rounds. GEM Digital's commitment is structured around future token purchases, meaning actual capital inflows depend on market conditions and token liquidity. If the token underperforms or never reaches sufficient trading volume, the full $10 million may never materialise.
Decentralised training networks also face steep technical hurdles. Coordinating gradient updates across permissionless miners introduces latency and quality-control challenges that centralised setups avoid by design. Whether VRAM Network's enclave-attested scoring and Bayesian peer ranking can reliably match centralised training quality at scale remains unproven beyond testnet.
The competitive field is crowded, too. Projects like Gensyn, Together AI, and several others are racing to crack decentralised AI compute, each with different architectural bets.
The signal: GEM Digital, classified on Dealroom as an investment fund, has built a playbook around token subscription facilities for early-stage crypto projects — structured commitments that tie actual capital deployment to future token market conditions rather than upfront wire transfers. That model suits speculative infrastructure bets like decentralised AI training, but it also means the headline figure overstates guaranteed funding. For VRAM AI, the real proof point will be whether its mainnet launch generates enough token demand to convert a paper facility into working capital.
Read more: worldonlinenewsreports.com
