today's ai news brings practical tools and big moves. google research tackles cloud memory costs with a new caching method. tokenspeed-kernel simplifies llm inference across different hardware. hugging face now lets you launch a vllm server with one command. netris raises funds to speed up ai data center setup. general intuition uses video games to train robots. on-device neural architecture search adapts tiny models in real time. adobe acquires topaz labs for video and image ai. we also look at open source omni models, an ai architect roadmap, amazon's india investment, europe's chip war pushback, and google's talent drain.

  1. linear elastic caching cuts cloud memory costs - this matters because it treats memory as a utility, dynamically adjusting to workloads to lower total cost of ownership in cloud databases.
  2. tokenspeed-kernel: clean apis for multi-silicon llm inference - it decouples llm runtimes from hardware kernels, making it easier to run models on different chips without vendor lock-in.
  3. spin up a vllm server on hugging face jobs in one command - this simplifies deploying private, openai-compatible llm endpoints, paying per second with no server management.
  4. netris raises $15m to speed up ai data center setup - network automation helps neoclouds go live in days instead of months, critical for scaling ai infrastructure quickly.
  5. video games train ai for real-world robots - general intuition's approach uses gameplay to teach spatial reasoning, then transfers skills to physical robots, backed by $320m in funding.
  6. on-device neural architecture search for sensor data - this allows tiny neural networks to adapt to new users and sensor data directly on embedded devices, improving personalization without cloud reliance.
  7. adobe buys topaz labs for ai video and image tools - the acquisition brings topaz's upscaling and enhancement models into creative cloud and firefly, boosting adobe's ai capabilities for creators.

also today, a look at five open source omni ai models for multimodal tasks, a 2026 ai architect roadmap, amazon's additional $13b for india ai infrastructure, europe's pushback on us chip export rules, and top google ai researchers leaving for anthropic and openai. check the site for full stories.