0G Labs

A distributed AI training network has published a verification framework that proves honest training across all nodes for their 107 billion parameter model - completed nine months ago, 48% larger than recent 72B benchmarks. **Key developments:** - Model trained with 107B parameters across distributed nodes - New verification framework addresses trust in decentralized training - Solves the core challenge: proving honest participation without centralized oversight **Why it matters:** Distributed AI training faces a fundamental problem - how to verify that remote nodes actually performed legitimate training work rather than submitting fraudulent results. This framework provides cryptographic proof of honest training. The two-step approach separates model training from verification, allowing networks to scale AI development while maintaining integrity across independent participants. [Read the technical details]()

**Payment infrastructure is rapidly adapting to AI agents:** - Visa launched a CLI tool for AI agent payments - Stripe integrated machine payments on Solana - Over 18,000 agents are already transacting on-chain - CFTC Chair stated "AI needs blockchain" **The shift is fundamental:** Traditional payment systems were built for human users, but the emerging agent economy requires different infrastructure. As agents begin executing autonomous transactions at scale, the need for blockchain-based identity, verified compute, and decentralized storage becomes critical. This isn't speculative—major financial institutions are building the rails now.

Jake Salerno will present on AI verification at EthCC Cannes on April 1, addressing a critical infrastructure gap in decentralized AI systems. **Key Focus:** - Hardware-level proof of honest AI inference - Moving beyond trust-based or log-based verification - Building verification as a core component of AI systems The talk explores how to cryptographically verify what an AI model actually computed - described as the missing layer for the agent economy. This addresses the fundamental challenge of ensuring AI outputs are authentic and untampered. A detailed analysis of verification approaches is available at [0g.ai/blog](https://0g.ai/blog/why-verification-matters-decentralized-ai-training).

$620 million flowed into a single AI subnet over 12 months, signaling a fundamental market shift. **Major institutional moves:** - Grayscale filed its first AI crypto investment vehicle - Kraken launched 0G margin trading The market is moving beyond generic Layer 1 blockchains toward **purpose-built AI infrastructure** that integrates chain, compute, storage, and data availability in unified stacks. This follows broader trends where Bitcoin miners, facing profitability pressures around $500 per BTC, are pivoting their GPU infrastructure toward AI compute as a yielding asset. Wall Street is actively financing this transition. The era of general-purpose chains appears to be giving way to specialized infrastructure designed specifically for AI workloads.

**0G Labs has deployed two critical technologies for decentralized AI training:** - **DiLoCoX**: Delivers 357x speedup in communication efficiency - **TEE (Trusted Execution Environment)**: Ensures computation integrity through hardware verification These dual capabilities address the core bottlenecks in distributed AI systems - communication overhead and trust in computation results. The combination enables production-grade decentralized AI training at scale. This development positions 0G Labs as a key infrastructure provider in the emerging decentralized AI landscape, where both speed and verifiability are essential for practical deployment.

**Major shift in AI compliance landscape** Training costs for AI models are skyrocketing from $100M+ toward $1B, while AI agents increasingly execute real financial transactions autonomously. **Key regulatory change:** - EU AI Act now requires proof of how models are trained - Verification for decentralized training transitions from optional to mandatory - Compliance becomes critical as AI agents handle actual financial decisions **Why it matters:** As AI agents make exponentially more transactions than humans, transparent training verification becomes essential infrastructure—not just for regulatory compliance, but for trust in autonomous systems managing real money.

**0G Labs has successfully trained a 107-billion parameter AI model** using their DiLoCoX framework, marking the largest decentralized AI system ever documented. **Key achievements:** - The model is **48% larger** than any system built by Bittensor - Runs on **standard 1 Gbps internet** connections - Operates **without traditional data centers** - Training completed in July 2025 The breakthrough demonstrates that large-scale AI training can occur across distributed networks using consumer-grade internet infrastructure. The DiLoCoX Framework enables cost-efficient training by coordinating compute resources across multiple nodes without requiring centralized data center facilities. 0G Labs plans to open-source their approach, potentially lowering barriers to entry for AI development and reducing reliance on expensive cloud infrastructure. [Read the full technical report](https://mpost.io/0g-labs-reports-107b-decentralized-ai-breakthrough-highlighting-cost-efficient-training-and-open-source-plans/)

**Efficiency breakthrough reshapes AI landscape** A 9-billion parameter model now delivers performance equivalent to 120-billion parameter systems, while 72B models match GPT-5 capabilities. This compression makes open-source AI viable for decentralized GPU networks. **Key implications:** - Smaller models enable distributed computing without data centers - 0G Compute Network demonstrates viability with 107B parameter system - Standard 1 Gbps internet sufficient for operation - 48% larger than previous decentralized systems **The fundamental shift** The critical question evolves from *who builds AI models* to *who controls their operation*. As models shrink while maintaining performance, centralized infrastructure becomes optional rather than mandatory. This efficiency gain removes traditional barriers to decentralized AI deployment, potentially redistributing control from major tech companies to distributed networks.

**Flashback Labs has migrated from centralized cloud storage to 0G Labs' decentralized infrastructure.** Most AI memory applications store voice recordings and personal data on AWS or Google Cloud, requiring users to trust these corporations with sensitive information. Flashback took a different approach. The platform now operates entirely on decentralized storage, eliminating reliance on corporate cloud providers. Current metrics show **802 wallets** and over **3,300 memories** stored without any centralized infrastructure. This shift means no single company controls access to users' family stories and personal recordings. The move represents a practical application of decentralized AI infrastructure for consumer-facing products.

**FlashbackLabs reaches major milestone** with over 10,000 transactions on its privacy-focused AI platform. The platform enables: - Real user interactions with AI - User-owned memory and data storage - Privacy-first architecture powered by 0G DeAI Stack This follows 0G Labs' recent launch with Americanfort of an AI-native private transaction system, allowing AI agents to transact using names instead of wallet addresses while maintaining cryptographic security.