Vertiv's 109% run and $15B backlog make it the default AI infrastructure pick, but a 30% pullback is telling platform buyers something about pricing power and lead times.
Researchers decoded 315,320 hidden reasoning blocks and pulled 182 credentials out of them. The three frontier vendors patched. The architectural debt didn't go away.
Google's Gemini Omni 1.1 Flash pushes generative video to 4K, 40-second scenes, and 360p previews at one third the cost. What it means for teams shipping AI video.
Jeff Dean is out, Gemini 3.5 Pro is cancelled, and GCP is compounding past 100% YoY. The build-vs-rent math for AI platform teams just shifted hard.
Thomson Reuters just shipped its own LLM, trained on under 10% of its archive, and claims parity with frontier models. The economics deserve a hard look.
Moonshot's Kimi K3 ships open-source on 27 July with 2.8 trillion parameters, beating Fable in blind tests and knocking 27% off Zhipu's stock.
DeepSeek dropped an MIT-licensed agent harness where models, tools, sandboxes, and the loop itself are all swappable plugins. Here's what it actually changes.