Ramp Data Shows 600x Divide in US Corporate AI Spend; a16z Charts Rise of Neoclouds
Ramp credit-card data shows US corporate AI spending is both accelerating and highly polarized: in July, the top 1% of AI spenders invested a record $7,400–7,500 per employee per month, the top 10% $650–660, while the median company spent only $12 — a gap of over 600x. Separately, a16z published charts and analysis arguing that cloud computing is splitting into generalist clouds and Neoclouds, with former crypto miners leveraging existing power rights and infrastructure experience to pivot into AI compute.
Confirmed
- Ramp data (consistent across multiple posters): top 1% $7,400–7,500 per employee per month, top 10% $650–660, median $11.95–12
- @rohanpaulai calculates the gap at 625x; a top company's afternoon spend can equal a median company's annual AI budget
- a16z's cloud industry chart and article split the market into generalist clouds and Neoclouds
Why it matters
- @rohanpaulai argues that even a small catch-up by median companies would dramatically increase inference demand, making this spending distribution a leading indicator
- @DavidLinthicum notes AI training's core need is not GPUs alone but synchronizing thousands of accelerators as one machine — demanding continuity, high-speed interconnect, storage, scheduling, power and cooling — so the scarce resource is systems engineering, not chips
- a16z draws historical parallels (railroads laying fiber to become Sprint, gas pipeline firms becoming WorldCom), suggesting crypto miners with power rights and construction experience are naturally positioned to become Neoclouds
2026-08-17 ~ 2026-08-18 · 5 related posts
Primary sources
- [source] AI Spending Gap 625x: Top 1% Spend $7,500/Employee/Month vs Median $12 — rohanpaul_ai · 2026-08-17
- Top 1% US firms spend record $7,400 per employee on AI as investment concentrates — luisdans · 2026-08-17
- [source] a16z Charts: How Neoclouds Repurpose Crypto Infrastructure for AI — FinanceYF5 · 2026-08-18
- Neoclouds Rise as AI Demands Synchronized Compute, Not Just Chips — DavidLinthicum · 2026-08-18
1 near-duplicate retellings: FinanceYF5