The AI infrastructure race and the model race are both accelerating in parallel: NAVER is scaling sovereign AI compute toward gigawatt levels on NVIDIA's DSX platform while DeepSeek V4 Pro and Xiaomi's MiMo Ultraspeed claim to leapfrog GPT-5.5 Pro and ChatGPT on precision and speed respectively. OpenAI is reportedly prepping its biggest ChatGPT redesign since launch as agentic interfaces replace chat, and Microsoft, LangChain, and Karpathy are all pushing on what it actually takes to run autonomous agents reliably - sandboxes, dedicated compute, and new workflows. Meanwhile the security story is sobering: Microsoft pulled ~70 GitHub-hosted open-source tools tied to Azure, Claude Code, Gemini CLI and VS Code after malware was found stealing developer credentials, a reminder that the tooling supply chain around AI development is now a prime attack surface. Underneath it all, several writers are converging on the same point - the model is no longer the bottleneck, trust and governance are.
- NAVER scaling sovereign AI infra toward gigawatt levels with NVIDIA's DSX platform
- DeepSeek V4 Pro claims precision edge over GPT-5.5 Pro
- OpenAI preps biggest ChatGPT overhaul since launch
- Microsoft pulled ~70 hacked open-source tools stealing AI dev credentials
- Agents are getting their own dedicated sandboxed compute
Model & Infrastructure Race
Agentic Engineering Gets Real
Design Meets AITrust, Governance, and Security Catch-Up