Vitalik Buterin links DeepSeek V4 to Ethereum privacy via CROPS access layer

Vitalik Buterin says DeepSeek V4 could support Ethereum’s privacy goals by enabling locally run AI models within a CROPS-based access layer. He argues that local setups can reduce exposure of metadata, IP addresses, and wallet balances to centralized RPC providers, and points to ZK-based paid remote LLM calls and private RPC reads as overlapping needs. Buterin also highlights hardware notes such as a 2-bit quantized DeepSeek V4 using 90GB of memory and recommends at least 96GB-128GB of Unified Memory or VRAM for running it locally.