0G co-founder and CEO Michael Heinrich is building a decentralized infrastructure stack positioned as a trust layer for artificial intelligence, where AI work can be executed, stored and verified outside centralized providers. The project combines verifiable compute, encrypted storage, private inference and an EVM-compatible Layer 1 blockchain built for AI workloads. The interview lands as the four largest hyperscalers committed more than $500 billion to AI capital expenditure in 2025 alone, with spending on pace to double this year. For business, the issue is direct: GPU hours now gate models, agents and products, yet access remains concentrated. Heinrich frames the response as making compute ownable and auditable rather than only rentable.
How 0G turns compute into a usable asset
Heinrich, a former software engineer at SAP Labs and Microsoft who later worked at Bain, Bridgewater Associates and J. P. Morgan and founded Y Combinator company Garten, started 0G in 2023 with co-founders Thomas Yao, Fan Long and Ming Wu. The stated motive was opacity in AI infrastructure: customers could not inspect what ran, where data went or what compute actually cost, while capital requirements compounded every quarter. 0G answers with a modular stack that developers can adopt in parts, from storage or compute alone to fully onchain applications. Components include private compute, open-weight 0gm models, safety mechanisms such as sparse inference alignment and TEE verification, encrypted decentralized storage and the 0G blockchain. Data, models and agent actions are designed to be audited and verified.
At the center is Compute Finance, termed ComFi, which treats compute like other productive assets that pay dividends or coupons. Instead of paying hourly cloud rent, a user locks 0G tokens as collateral, receives a0G, mints iAI and commits that iAI to earn usable compute across supported 0G AI services. Under initial parameters, committed iAI can yield compute credits worth more than $1 per day, while the underlying collateral stays locked and can be recovered when iAI is burned. Heinrich stresses that users receive access to compute, not cash or a financial return. The flow runs on the 0G blockchain, where usage can be verified rather than accepted on trust. Ascend and iAI are presented as the first implementations of ComFi as an open category.
The timing reflects a shift in compute economics from purchased input to binding constraint on intelligence itself. Every model improvement and agent deployment ultimately consumes GPU hours, so access determines which companies, universities and countries can build. Yet compute, unlike capital and energy, has no transparent market for holding, hedging or allocating claims, only procurement contracts with a handful of vendors. Heinrich compares the path to oil, electricity and bandwidth, which moved from production to pricing to financing, and argues compute is entering that third stage. 0G Compute addresses this with a decentralized marketplace where GPU providers set prices and compete for workloads. The company cites up to 95 percent lower costs for comparable open-model inference and storage at more than 90 percent below AWS S3, partly from competition and use of idle capacity.
What decentralized verifiable compute means for business
For companies running models or agents, the practical change would be more predictable access to capacity outside long-term hyperscaler relationships. Infrastructure providers gain another channel to sell capacity beyond traditional cloud contracts, while developers can budget around a resource they hold rather than hourly rented quotas. AI companies could secure compute for private inference, application building and agent operations with audit trails for data and actions. Smaller teams working with open models stand to benefit first, since lower inference and storage costs reduce the barrier to testing and deployment. Larger enterprises could use modular components selectively, for example encrypted storage or verifiable inference, without moving entire workloads onchain.
The model also introduces conditions that buyers should examine before planning around it. Compute earned through ComFi is usable access, not income, and collateral remains locked while the position is active, which affects liquidity and treasury decisions. Cost comparisons cited by 0G refer to open-model inference and its own storage layer, not necessarily closed frontier models on dedicated infrastructure. Elasticity depends on independent GPU providers joining the marketplace and offering suitable performance, availability and support. Prospective users need to verify supported services, credit redemption terms, verification methods including TEE processes, and responsibilities for data privacy and compliance. The interview does not by itself prove sustained pricing or capacity at scale.
A concrete marker to follow is whether ComFi moves beyond Ascend and iAI to additional implementations on the 0G network with disclosed usage and pricing. Growth in staked positions, redeemed compute credits and independent providers on 0G Compute would show whether the marketplace delivers elasticity during demand peaks. Hyperscaler capital spending trends provide the other side of the test for cost pressure. If both lines develop, verifiable decentralized infrastructure could become a parallel procurement option for business AI.
