capexScore 65/100Add to thesis
NVIDIA per-gigawatt compute economics rising: $18B (Hopper) to $25B (Blackwell) to $40B+ (Vera Rubin)
“In the gigawatt, Hopper was $18 billion of economics for NVIDIA per gigawatt. Grace Blackwell is about 25. The Vera Rubin generation is 40 plus billion dollars per gigawatt of total NVIDIA compute economics... there's like seven different types of NVIDIA chips that has to go into an AI factory. An XPU would be a replacement of one of those.”
Why it matters
This quantifies NVIDIA's revenue intensity per gigawatt of AI factory buildout roughly doubling from Blackwell to Vera Rubin, driven by networking, switching, and CPU content beyond the GPU. It frames the bull case that custom XPU displacement of GPUs only removes one of seven chip categories per AI factory.
Investment implication
Implies networking and interconnect (NVLink switches, Spectrum-X) become a growing share of NVIDIA revenue per data center dollar, supporting networking suppliers (e.g., CoWoS packaging, optical interconnect vendors) and suggesting hyperscaler capex per gigawatt flows increasingly through NVIDIA's full-stack platform.