Goldman Sachs — Gen AI: Too Much Spend, Too Little Benefit?
Citation
Nathan, Allison and Covello, Jim (lead skeptic voice). Gen AI: Too Much Spend, Too Little Benefit? Goldman Sachs Top of Mind, Issue 129, June 25, 2024. URL: https://www.goldmansachs.com/images/migrated/insights/pages/gs-research/gen-ai—too-much-spend,-too-little-benefit-/TOM_AI%202.0_ForRedaction.pdf
Status
Stub. Currently summarized only via bear-case-synthesis. Primary report not yet ingested directly.
Core argument (as inherited)
Asks whether ~$1T in AI capital expenditure by hyperscalers and semi vendors can be economically justified. Anchored by Jim Covello (GS Head of Global Equity Research) arguing that gen AI currently produces only marginal efficiency gains on easily-solvable tasks, not the complex enterprise-grade problem-solving needed to justify the spend. Compounds the financial argument with utility, regulatory, and grid constraints as physical caps on deployment.
Claims it challenges
- The Silicon Valley narrative that gen AI is a universally applicable killer app comparable to the early internet.
- The implicit assumption that AI inference cost will follow a Moore’s Law–style deflationary curve.
- The bull thesis on AI-exposed infrastructure equities.
Why it matters for Where Value Lands
- Comes from the inside of capital markets — hard for capital-rich operators (the paper’s audience) to dismiss as outsider skepticism.
- Frames the deployment bottleneck as physical and energetic, not just economic. This connects directly to the visionary horizon section of 03_structure (energy as the long bottleneck).
- Feeds into ai-capex-cycle (within circular-ai-economy) and into any scenario analysis touching infrastructure.
To verify on primary read
- Exact capex figures and timelines.
- Specific verticals where Covello argues AI fails to displace legacy.
- The grid / utility data points (regions, magnitudes, regulatory horizons).
Related
- bear-case-synthesis
- circular-ai-economy
- H2_u-curve-of-value
- wef-ai-in-action-2025 — bull counterpart on the spend trajectory; corroborates the magnitude (IDC $632B by 2028) but reads the return question optimistically.