xAI continues aggressive scaling of its Colossus supercluster in Memphis. Colossus 2 (1 GW scale) is now operational with significant GB200/GB300 deployments alongside earlier H100/H200 units, targeting expansions toward 555k+ GPUs and multi-GW capacity. Roadmap points to 1M GPU equivalents. Primary source: xAI site and updates
Meta deepens NVIDIA partnership with multi-year, multi-generational deal for millions of Blackwell and upcoming Rubin GPUs, plus NVIDIA CPUs, Spectrum-X networking, and Vera Rubin NVL72 rack-scale systems. Meta also advancing AMD deployments (multi-GW scale starting H2 2026) and its own MTIA custom silicon. Zuckerberg emphasized personal superintelligence clusters.
Microsoft/OpenAI ecosystem continues massive NVIDIA GPU intake (hundreds of thousands of GB200s reported in planning) alongside custom Maia efforts. Broader hyperscaler capex remains in the hundreds of billions annually
Google Cloud expanding co-engineered infrastructure with NVIDIA (including early Vera Rubin access in H2 2026) while pushing custom TPUs (v7/v8 generations) via Broadcom partnership locked through 2031. TPU focus on training (8t) and inference (8i).
xAI has rolled out the Grok Build Plugin Marketplace, bringing a built-in plugin system to its terminal-native coding agent. The marketplace allows users to browse, install, and manage plugins directly inside Grok Build without leaving the terminal. Plugins bundle skills, slash commands, agents, hooks, MCP servers, and LSPs into single installable packages. Everything is pinned…