evidence · section 03 of 10
When does owning GPUs beat renting them, and what sets the threshold?
What this section answers
Specialist clouds charge 50–70% less than hyperscalers for the same H100, and the on-prem crossover splits by procurement scope: a lean operator breaks even against hyperscaler on-demand near 21% utilization, an enterprise node-loaded buyer near 57%.
Boundary
The crossover bands move across roughly 35–65% with the amortization window and staffing assumptions named in the model.
Source coordinate
Report heading GPU ownership — three verified claims and a crossover model, under What the evidence establishes, in The real cost of AI: August 2026.
The GPU notebook has 7 references (a compiled 8-source evidence file grade C + direct captures including A6 grade A, A7 grade A) and 3 verified claims, plus the TCO crossover model (analysis/tco-crossover-2026-08.md):
C1 (verified, 3 independent sources): Specialist GPU clouds (“neoclouds” like RunPod, Lambda, Together AI) charge 50-70% less than hyperscalers for the same H100 GPU-hour: median $3.99/GPU-hr vs $7.89/GPU-hr (+98%). The ~2x hyperscaler premium is consistent across H100, A100, H200, and B200.
C2 (verified, 4 independent sources): On-premises GPU clusters break even with hyperscaler on-demand at roughly 50-83% sustained utilization, but rarely win against specialist GPU clouds at any utilization once fully loaded.
The measured price gap between specialist clouds and hyperscalers, and the two procurement regimes.