Generative AI Economics: Two-Year Update Shows Value Chain Still Inverted
Summary
- • AI ecosystem grew ~5x to $435B annualized revenue but semiconductors still capture 70% of revenues
- • NVIDIA alone added $175B incrementally — roughly 3x the size of the entire app layer today
- • OpenAI and Anthropic jointly hold 75% of the $60B application layer
- • Analysis: semis are a one-player game; apps a two-player game; infra is the only competitive layer
Details
AI ecosystem grew ~5x from $90B to $435B annualized
The overall market expansion is dramatic, but the distribution of that growth is highly skewed — semiconductors absorbed the vast majority of incremental dollars.
Semi layer ~$300B, 70% of total AI revenues, 79% of gross profit
NVIDIA's data center business alone annualizes to ~$250B; Broadcom's AI semiconductor business adds ~$34B. NVIDIA holds roughly 80% of the semiconductor layer.
Infra layer ~$75B; each hyperscaler contributing $10-20B
Azure, AWS, GCP, Oracle each contribute $10-20B in AI-attributable revenue; CoreWeave adds ~$6B. Infra runs at ~55% gross margins, the middle tier in both revenue and profitability.
App layer ~$60B; OpenAI and Anthropic = $45B (75%)
The app layer grew 12x in two years but added only ~$55B in absolute revenue versus ~$225B for semis. App gross margins sit at ~33%, the lowest of the three layers.
NVIDIA added ~$175B incrementally — ~3x the entire app layer today
This comparison underscores why the value chain inversion persists: even the fastest-growing application companies cannot keep pace with the hardware layer in absolute dollar terms.
Hyperscaler capex $443B in 2025, projected $600B+ in 2026
Approximately 75% (~$450B) of 2026 projected capex is directed at AI infrastructure, meaning semiconductor and infra layer revenues are structurally supported for the near term.
Over a decade for apps to reach cloud-era software revenue share
In cloud, the application layer eventually captured the largest share of value. The analysis contends the AI value chain is moving in that direction, but the pace is far slower than optimists assume.
Semis = one-player game, apps = two-player game, infra = contested
NVIDIA's dominance in semis and OpenAI/Anthropic's combined 75% app-layer share leave infrastructure as the primary battleground among hyperscalers, GPU providers, and emerging entrants.
Financials = revenue/margin/capex data, Market Impact = competitive scale comparisons, Insight = attributed analytical argument from the source
What This Means
Despite explosive overall growth, the analysis argues the generative AI value chain remains structurally tilted toward hardware — NVIDIA in particular — and that the application layer, for all its growth rate momentum, is unlikely to close the gap with semis in absolute dollar terms for a very long time. For anyone building, investing in, or competing within the app layer, the implication is that high growth rates mask a persistent ceiling on value capture relative to the infrastructure and semiconductor layers beneath them.