Silicon Data · Ramp AI Index

AI Compute Tape

Token expenditure, GPU rental spot, the rental term structure, and business AI adoption and spend — captured daily. None of these sources publish history, so this archive is the only record of what came before.

Tokens & curve GPU spot Ramp

01 — Price

GPU rental price vs token expenditure index

Rental prices on the left axis in dollars per GPU-hour; the token expenditure index on the right in dollars per million tokens. Raw levels, no rebasing. Token indices share one hue in three shades; each GPU has its own hue, with hyperscaler lighter and dashed against neo-cloud. L and R in the legend mark which axis a series is read against.

From 22 Aug, lengthening daily. Silicon Data publishes a rolling 7-day window with no backfill, so the record starts the day the scrape started — Citadel licence the history, ours accrues. With two scales the visual gap between the lines is arbitrary: only the shape of each line, and where it turns, carries information.

Data table

Rental term structure and forward curve

All three GPUs out to 36 months. Each GPU is a hue; the term rate is solid, the forward rate dashed and lighter. Tenor 0 is spot, where term and forward meet by construction. Switch to Compare vintages to overlay today's curve against a week, a month and six months ago for one GPU.

The public site only plots 0–6 months. The 36-month point — the one used to talk about backwardation — is in the payload but never shown on screen.

Data table

02 — Adoption

AI adoption, headline

Share of firms with AI spend in the month. The solid line is the Ramp customer base; the dashed line is Ramp's estimate for US businesses overall — much lower, because the Ramp base skews to earlier adopters.

Data table

Adoption by sector

NAICS sector, share of firms with AI spend.

Data table

Adoption by business size

Small, medium and large firms.

Data table

Adoption by model provider

Share of firms with spend at each provider. A firm can appear under several.

Data table

03 — Spend per employee

Spend per employee, overall distribution

Monthly AI spend per employee across the distribution. Plotted on a log axis — the top 1% and the median differ by roughly 600×, and on a linear axis the median would be pinned flat to the floor.

Naming trap. The figure usually quoted as the “90th percentile” is Top 10% median, not P90 — 650 ÷ 11.95 = 54×, the widely cited ratio. P90 ÷ median is only 22×. Both are on this chart; they are not interchangeable.

Data table

Spend per employee by business size

Median monthly spend per employee.

Data table

Spend per employee by financing status

Median monthly spend per employee.

Data table

Spend per employee by sector

Median monthly spend per employee, by NAICS sector.

Data table