Vitalik Buterin has highlighted the growing intersection between Ethereum infrastructure and artificial intelligence, emphasizing the need for optimized models for relevant use cases. He announced the release of Deepseek V4, a 2-bit quantized model that operates within 90GB of memory, achieving speeds of 35 tokens/second on Apple hardware but only 7 tokens/second on AMD. Buterin stressed the importance of supporting multiple hardware vendors to differentiate decentralized AI from centralized AI models. Additionally, Buterin discussed Mistral's Leanstral model, which focuses on Lean code writing and can run within 70GB, offering performance comparable to a 1TB parameter large model. He underscored the role of AI-assisted formal verification in enhancing code security, particularly for core components like STARK, consensus algorithms, and EVM. Buterin urged the Ethereum community to fine-tune AI models for Ethereum-specific applications and ensure efficient support across diverse hardware platforms.