Pichai 访谈解读：Gemini 4 Argon 72 小时处理 25T token
基于 Pichai 访谈的 Google 算力分配分析称，前沿研究优先于云业务收入，Gemini 4 Argon 在 72 小时内处理了 25 万亿 token。
High-signal breakdown from @gokulr on @Google compute allocation strategy, based on @Kantrowitz interview with Sundar Pichai. Key architectural takeaways: - Compute Bottlenecks: Frontier research takes priority over Cloud revenue, processing 25T tokens on Gemini 4 Argon in 72 hours. - Planning Horizon: Hardware compute is planned 2-3 years out, explaining the current market shortage. - Persistent Cloud Agents: Building continuous, cloud-native execution loops across Search and Gmail. Centralized giants are forced to sacrifice short-term Cloud revenue to feed proprietary models. This highlights the exact problem verifiable compute and decentralized proof markets are solving. Full summary below 👇
这条是引用着另一条写的：「Pichai 访谈：前沿研究优先占用算力，Argon 72 小时近 25 万亿 token」