Meta locked $21B for inference. my sessions still break.

CoreWeave just expanded its deal with Meta to $21 billion for AI cloud capacity through 2032. The same week, they signed a multi-year agreement with Anthropic for Claude workloads. CoreWeave now serves nine of the ten largest language model developers. If you are building anything that touches AI inference, the infrastructure underneath you is consolidating fast. Last session was 375 merged events across Letta, Claude Code, and Codex. Most of the day was prototype parity corrections, pixel-perfect alignment, tablet and mobile regressions. I corrected title alignment five times. I authorized a desktop Rust fix, dispatched 20 agents for deep research, and resolved a backyard blocker chain. But the thing that actually mattered was none of that. The real carry-forward was session rehydration. The product can show you a session list, but it cannot show you the actual conversation history. When a session ends or the app restarts, the context is gone. That is the difference between a prototype and a product. Rehydration means: when you reopen a session, you see the full conversation as it happened. Not a summary. Not a placeholder. The real thread, across runtimes. This is the hardest infrastructure problem in the app right now because sessions can come from Codex, Claude Code, ACP, or a local agent, and each has a different shape. This stream is about designing and building the rehydration layer. If you write frontend code and struggle with state that doesn't survive navigation, this is the same problem at a different scale. AI radio is on. If you are coding and trying to make things actually persist instead of just look right, this is the room. Sources: - https://www.coreweave.com/news/coreweave-and-meta-announce-21-billion-expanded-ai-infrastructure-agreement - https://cointelegraph.com/news/coreweave-agreement-anthropic-ai - https://techxplore.com/news/2026-04-anthropic-mythos-ai-uncovers-thousands.html