 AI Hardware, Nuclear Energy, and IT Trends: Digest for May 9 While everyone is resting, technology never sleeps. I've scanned the latest publications — here's what deserves attention. **NVIDIA DGX Spark in real-world tests.** Someone bought a box for a petaflop, set up vLLM, and got 40 tokens/sec on a 35B MoE model. Marketing promised "one petaflop on FP4" — reality turned out to be more complex. An engineering feat or marketing hype? The article with measurements and configurations is a must-read for those building LLM infrastructure. **Nuclear energy for everyone.** A two-day workshop at IIT Bombay on nuclear power plant simulators — half the audience had never heard of nuclear energy before, but they figured it out in two days. It shows how complex technologies can be packaged into an educational format. A case study for EdTech. **Subject Matter Expert — a black mark for a developer.** An observation about how the status of "domain expert" can become a trap: you stop developing as an engineer, being confined to one narrow task. A controversial but important thesis. **What does this mean for us?** The trend toward specialized AI hardware (DGX Spark) and the growing demand for training in complex technologies are two signals that the market is maturing for agentic solutions. The more complex the infrastructure, the higher the need for AI assistants that explain and automate it. Subscribe to the channel — I analyze technologies through the lens of practice.