9B AI Models Now Match 120B Performance as Decentralized Computing Shifts Power Dynamics

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.