NVIDIA's reported ~3.76M 2023 data-center-GPU shipment figure traces only to a paywalled TechInsights study relayed by non-independent trade journalism

ai-energyjevons-paradoxgpunvidiatechinsightsunverified-quantquantitative-verification

Every version of the ~3.76 million NVIDIA 2023 data-center-GPU-shipment figure found in this trace attributes to one unreached primary: "a study conducted by semiconductor analyst firm TechInsights." HPCwire (Agam Shah, 2024-06-10) reports the figure directly; DataCenterDynamics (Charlotte Trueman, 2024-06-12) reports the same number two days later — but DCD's own text credits HPCwire as its route to the finding ("As reported by HPCwire"), so the two articles are not independent corroboration in the single-source-concentration sense. Both ultimately rest on the same unreached TechInsights report. No NVIDIA first-party disclosure of GPU unit-shipment counts (as opposed to Data Center segment revenue) was located during this search; unit shipments are not a metric NVIDIA is known to break out in its own investor materials.

This is the quantitative claim underlying claim-luccioni-2025-nvidia-gpu-figure-misdated-2023-as-2024 and the illustrative figure in claim-jevons-paradox-invoked-in-ai-energy-debate-luccioni-2025. The sourcing floor for quantitative claims (00-meta/specs/sources.md) requires Tier 1–2; at every hop this search could reach, the number is a market-research estimate relayed by journalism, not an original data disclosure — Tier 3 at best. TechInsights' own product page, the closest thing to a primary reached, states a different figure again — see claim-techinsights-own-page-contradicts-journalisms-376m-gpu-figure.