MCG Live · July 1, 2026 · 45 min
c0mpute is a decentralized AI inference network where users can run private, uncensored models on idle GPUs across the world. Founder Leighton, a 21-year-old electrical engineer, built sharding technology that splits large models across multiple consumer GPUs—solving the latency problem that previously made decentralized inference impractical—allowing enterprises to access frontier-quality AI models cheaply and privately for the first time.
Founder Leighton started c0mpute in January after his $200 Claude subscription kept going down for hours, realizing millions of idle GPUs worldwide could provide cheaper, private AI inference.
Sharding splits large language model layers across multiple consumer GPUs instead of running on one device, but single tokens moving through the network created latency problems.
Speculative decoding solves the latency issue: a small draft model predicts multiple tokens at once in a batch, then other workers verify correctness instead of processing one token sequentially.
c0mpute has been live for about 20 days and grown from 10 workers (including the founder) to 80 workers online, with peaks reaching over 100 workers.
Partnership with Use Pod lets stakers of the $ZERO token sell their daily inference allowance to users who want to buy compute on a marketplace.
Staking $ZERO token provides daily inference credits weighted by stake amount, and workers who stake get 80% revenue share versus 70% for non-stakers, with remainder funding the treasury.
$ZERO token ran to 12 million market cap generating substantial fees, putting the treasury in a massive surplus; Leighton spends approximately $200 daily renting GPUs on Vast.ai for testing the sharding engine.
Leighton chose Solana because it's where he made the most money and feels at home, and the $ZERO token is on Solana while the compute network runs separately from the blockchain.
The entire codebase is open source, allowing people to verify claims and fork the project; Leighton expects copycats to fail if they merely extract token value without building a real product.
This summary was written from the recording's transcript by a language model. For the exact words, ask the MCG Live archive: every answer there is quoted word for word and linked to the second it was said. Market Bubble Search is independent and not affiliated with the show. Nothing here is advice.