Ara Kharazian 2 items
‹ All posts People /Ara Kharazian /News ARCHIVE
News · Wed, 23 Sept 2026 · 15:30 ET

Ramp: AI was 5.73% of business spend in August

Ara Kharazian adds AI share of non-payroll business spend: 5.73% in August, almost entirely tech + technical services on Ramp’s panel.

primary Ara Kharazian RampAnthropicOpenAI Source ↗
Outlet: Ramp / Ara KharazianPublished: Wed, 23 Sept 2026 · 14:51 ET

Opening

Ramp’s Ara Kharazian published a new intensity metric: AI was 5.73% of business spend excluding payroll in August — “almost entirely” tech and technical services. In an X post tied to the Ramp AI Index, Kharazian says Ramp’s card / invoice / ACH panel covers $10B+ in AI spend annually, and that the firm added AI share of business spend as an economy-wide intensity indicator. The August print is 5.73% of business spend excluding payroll, concentrated in tech + technical services. This is distinct from the Sep 9 “Cracks in the AI Thesis, part 2” letter (Anthropic 43.8% / OpenAI 39.8% adoption; top-1% median AI spend per employee −9.7% to $7,205), which did not headline this 5.73% share figure.

The numbers

AI share of business spend ex-payroll — 5.73% in August — Ara Kharazian / Ramp (panel of businesses on Ramp).

AI spend covered — $10B+ annually across corporate card / invoice / ACH — Ara Kharazian / Ramp.

Sector concentration — almost entirely tech + technical services — Ara Kharazian / Ramp.

Why it matters

A single intensity print does not settle whether hyperscaler capex holds; it does put a falsifiable business-spend denominator under the “software AI is eating opex” story that sits beside token-price cuts already in the archive. Concentration in tech / technical services means the figure can look large while remaining thin outside those verticals — Kharazian states that concentration explicitly. Company-data framing: Ramp is talking its own spend panel.

Positioning

AI capex durability — Opex AI intensity on a card panel is adjacent evidence, not a hyperscaler capex guide; theme still breaks on a top-four sequential capex cut or inference cost destroying new-build need — neither is in this print.

Inference margin inversion — Same desk window as cheaper Opus / Sol-Luna list cuts; a rising or high AI share of spend can coexist with falling $ per token if volume offsets price — this print alone does not split price vs volume.

Connects to

Ramp Astra vs Fable spend share — Prior Ramp principal thread on frontier mix (discussion).

Opus 5.5 / Sol-Luna cheaper list — Serving-price cuts that interact with measured business AI spend (news).

The Open/Close  ·  Research commentary, not investment advice. Positions may be held in securities mentioned.