Craig Churchill
CEO & CBO @NormalComputing
3 Apr 2026OpenAI $852B, Legora $100M ARR, SpaxeX IPO, April Fools Award
Normal Computing: Series B and the Compute Energy Wall
Faris and Craig join fresh off Normal Computing's Series B. Faris describes the mission as recursive self-improvement in hardware — AI designing the next generation of chips that run the next generation of AI — and warns of a roughly 49-gigawatt US power shortfall for compute around 2028 that demands unconventional architectures built in partnership with AI. Craig says demand pull is exceptional on both sides, citing Nvidia compressing chip spins from three years to 12 months and the memory shortage, and both credit ARIA, the UK agency chaired by Matt Clifford, as a key partner; the two are calling from Germany.
Thermodynamic Computing: Noise as a Resource and ASICs for Diffusion
Faris explains thermodynamic computing treats physical noise as a computational resource and does computation directly in memory, easing GPU bandwidth limits; the first ASICs target diffusion-model inference like video generation, noting OpenAI reportedly spent about $15m a day running Sora. He envisions thousands of workload-specific ASICs rather than single flagship chips, and attributes skyrocketing memory stocks to agentic workloads' larger KV caches colliding with slow fab ramps. Craig compares today's pace to his Sun Microsystems years and signs off with a shoutout to the show and the company's UK-based silicon team.