👀In Case You Missed It
Meta poaches Thinking Machines Lab co-founder Andrew Tulloch to join its Superintelligence Lab — part of a wider AI reorg and infrastructure spree.
xAI hires ex-Nvidia researchers to build AI-generated 3D “world models”, with Musk teasing a playable game by 2026.
OpenAI claims GPT-5 “instant” and “thinking” modes reduced political bias by ~30% across 500 hot-button prompts.
Microsoft warns Azure capacity constraints will continue into 2026, especially in Northern Virginia and Texas, limiting new signups and rerouting workloads.
Read more here
🧠 Cloud & Infrastructure Watch
OpenAI and Broadcom to build 10GW custom AI accelerators first racks 2H 2026 full by end 2029
Read MoreMicrosoft launches MAI Image 1 in house image model top 10 on LMArena coming to Bing Image Creator and Copilot
Read MoreSalesforce rolls out new Slack AI rebuilt Slackbot Channel Expert Agentforce and ChatGPT integrations
Read MoreWalmart adds shopping in ChatGPT link account browse and checkout in chat
Read MoreOracle unveils Zettascale10, designed to link hundreds of thousands of Nvidia GPUs across multiple regions as “the largest in the cloud.”
Read MoreCloudflare’s Connectivity Cloud is going native on Oracle Cloud Infrastructure to secure and accelerate multicloud AI workloads.
Read MoreNvidia signals shift to 1MW IT racks and 800V DC power, with partners preparing for next-gen thermal and electrical demands
Read More
💻 Hardware & Semiconductor Moves
Nvidia’s DGX Spark desktop AI system now on sale about $3,999 and ~1 PFLOP class aimed at local model work
Read MoreBroadcom launches Thor Ultra 800G NIC for AI data centres as rivalry with Nvidia heats up
Read MoreAMD lands new OCI deal Oracle to deploy 50k Instinct MI450 GPUs starting 2026 building on prior AMD wins
Read MoreOpenAI and Oracle to deploy roughly 450,000 Nvidia GB200 GPUs at the Stargate site in Abilene, Texas.
Read MoreAccelsius launches the MR250 in-row CDU, enabling up to 250kW direct-to-chip cooling per rack.
Read MoreNvidia’s 1MW rack and 800V roadmap is becoming the baseline for next-generation AI hardware design.
