---
title: "Verify the market figures Luccioni, Strubell & Crawford (2025) cite for AI's rebound effect — e.g. ≈3.7M GPUs shipped in 2024 — against primary sources"
type: "question"
status: "answered"
writer_model: "claude-opus-4-8"
date_raised: "2026-07-12T00:00:00.000Z"
tags: ["verification","ai-energy","jevons-paradox","gpu","nvidia","rebound-effect","unverified-quant"]
answered_log: ["2026-08-17 — [[claim-luccioni-2025-nvidia-gpu-figure-misdated-2023-as-2024]], [[claim-nvidia-2023-gpu-shipment-figure-traces-only-to-paywalled-techinsights-study]], [[claim-techinsights-own-page-contradicts-journalisms-376m-gpu-figure]] — traced the citation chain as far as it goes without paid access: the paper's own footnoted source dates the figure to 2023, not 2024, contradicting the paper's own sentence; and the underlying number itself is unreachable above Tier 3 — a paywalled TechInsights study relayed only by non-independent trade journalism, whose own product page states a different unit count and market share for the same year. Verdict: the figure does not check out as the paper states it, and this line of inquiry is settled at 'not independently verifiable further,' not merely 'not yet verified.'"]
---


[[claim-jevons-paradox-invoked-in-ai-energy-debate-luccioni-2025]] is Tier-1 for
its core invocation (verbatim quote preserved), but the paper's supporting market
figures — notably that NVIDIA shipped on the order of ≈3.7M GPUs in 2024 despite
per-unit efficiency gains, and the general claim that AI's rebound effect can
exceed 100% — are the paper's own downstream citations, not independently
sourced numbers.

**What to read / pull:**
- The relevant section of arXiv:2501.16548 and the specific citation it gives for
  the GPU-shipment figure (trace it to the market-research primary, e.g. the
  quoted analyst report).
- NVIDIA data-center shipment/revenue disclosures or a reputable market tracker
  for 2024 GPU volumes.
- The rebound-effect economics literature for whether >100% rebound is empirically
  established or a bounding argument.

**What it gates:** clearing the `[unverified-quant]` flag on the note's
illustrative figure. The core claim (that the paper *invokes* the Jevons paradox)
is already Tier-1 verified and does not depend on this.


## Progress log

- 2026-08-17 — [[claim-luccioni-2025-nvidia-gpu-figure-misdated-2023-as-2024]], [[claim-nvidia-2023-gpu-shipment-figure-traces-only-to-paywalled-techinsights-study]], [[claim-techinsights-own-page-contradicts-journalisms-376m-gpu-figure]] — traced the citation chain as far as it goes without paid access: the paper's own footnoted source dates the figure to 2023, not 2024, contradicting the paper's own sentence; and the underlying number itself is unreachable above Tier 3 — a paywalled TechInsights study relayed only by non-independent trade journalism, whose own product page states a different unit count and market share for the same year. Verdict: the figure does not check out as the paper states it, and this line of inquiry is settled at 'not independently verifiable further,' not merely 'not yet verified.'