Model economics and AI infrastructure dominate the day: DeepSeek pushed its production V4 Flash model onto Ollama's cloud with a built-in speculative decoding module, GPT-5.6's Luna tier got 80% cheaper even as aggregate token spend keeps rising, and the EU committed $11.4B to seven new AI gigafactories. Leadership news broke too, with Scale AI tapping former Google Cloud COO Francis deSouza as CEO. On the systems side, engineers are writing about heap-size wins, platform engineering's staying power, and picking models for speed over raw intelligence, while Reddit's CEO publicly questioned whether Google's AI Overviews are actually a win for content creators. For a senior engineer, the throughline is that model costs are dropping per-token but total AI spend, and the infrastructure build-out behind it, keeps accelerating.

• DeepSeek's V4 Flash lands on Ollama's cloud with speculative decoding for agentic tasks
• Scale AI names ex-Google Cloud COO Francis deSouza as its new CEO
• EU commits $11.4B to build seven AI compute gigafactories
• GPT-5.6 Luna gets 80% cheaper even as total AI token spend keeps climbing
• Reddit's CEO says Google's AI Overviews still aren't a win-win for publishers