Groq, Hotel Delivery Robots, and Mark Launches a Company

It’s been a travel-heavy hiatus—Mark’s been living in Spain and Shashank’s been bouncing across Asia (including a month in China)—but they’re back to unpack a packed week of AI news. They start with the headline hardware story: the Groq (GROQ) deal/partnership dynamics and why ultra-fast inference is becoming the next battleground, plus how this could reshape access to cutting-edge serving across the ecosystem. From there, they pivot to NVIDIA’s CES announcements and what “Vera Rubin” implies for data center upgrades, cost-per-token curves, and the messy real-world math of rolling hardware generations. Shashank then brings the future to life with on-the-ground stories from China: a Huawei “everything store” that feels like an Apple Store meets a luxury dealership, folding devices that look straight out of sci-fi, and a parade of robots—from coffee bots to delivery robots that can ride elevators and deliver to your hotel room. They also touch on companion-style consumer robots and why “cute” might be a serious product strategy. Finally, Mark announces the launch of Novacut, a long-form AI video editor built to turn hours of travel footage into a coherent vlog draft—plus export workflows for Premiere, DaVinci Resolve, and Final Cut. They close by talking about the 2026 shift from single model calls to “agentic” systems, including a fun (and slightly alarming) lesson from LLM outcome bias using poker hand reviews. Topics include: Groq inference, NVIDIA + CES, Vera Rubin GPUs, GPU depreciation math, China robotics, Huawei ecosystem, hotel delivery bots, companion robots, Novacut launch, Cursor vs agent workflows, and why agents still struggle with sparse feedback loops.

Link mentioned: Novacut — https://novacut.ai