Ramp

Sources#

• 2026 State of Scaling: The Great Sorting
• A New Look at AI's Impact on Jobs: Firm-Level AI Spending and Workforce Adjustment
• Ramp's latest data on China vs. the American AI Labs
• September 2026 Ramp AI Index: Cracks in the AI thesis, part 2

What it is#

A US corporate-card and bill-pay platform. It appears in this wiki not as a product but as an instrument: Ramp's line-item transaction records are the only source in the corpus that observes which firms pay which AI vendors, when, and how much, and its research arm publishes cuts of that data under the Ramp AI Index brand. Three raw documents in the corpus run on it, all classified by the same vendor/line-item classifier (foundational LLMs, GPU cloud, model serving and inference, coding agents, API tokens, AI image/video, AI search).

Ara Kharazian is Ramp's lead economist and the author of both index letters below; he previously led economic research at Square. Ramp's own framing of the index — "our flagship research… to track how American businesses are using AI" — is the relevant conflict of interest in one sentence: the dataset is a marketing asset as well as a measurement.

What it has published into this corpus#

Part 1 of the Cracks in the AI thesis serial (the August 2026 edition) is not in the corpus, so every figure the September letter attributes to "last month" reaches the wiki second-hand.

Three properties of the instrument#

Sampling and conflict of interest. Ramp measures its own customers — a business-spend-active, VC-forward-skewed base, not a random sample of US businesses — and has a commercial interest in owning the authoritative AI-adoption dataset. The index's paid-adoption level (~56% of eligible businesses in August 2026) sits far above the Census BTOS firm-weighted 18% and above OECD's 20.2% of European enterprises, and the gap is mostly population, not method. The rule the wiki applies: directions are far more defensible than levels, and every load-bearing figure is attributed inline.

Aperture. The rail sees purchases that cross a card or bill-pay flow. It is a good instrument for per-seat and per-API purchasing and a poor one for enterprise agreements, negotiated invoices and cloud committed-spend drawdowns — which is why the two vendors with the largest enterprise contract businesses, Microsoft (2.01%) and Google (6.15%), are the two smallest lines in a series where Anthropic reads 43.8%. It is also blind by construction to self-hosted open weights, which is the limit that matters most at The Open-Weight Frontier Gap. Denominators shift between cuts and are easy to conflate: vendor shares are a share of eligible businesses, while model-serving and spend series are shares of AI-spending businesses.

Revisions. Recent months are incomplete on first print and revise upward as late transactions land. The September 2026 letter discloses one directly — July top-1% per-employee spend restated from ~$7.4K to ~$8.0K — and comparing the two ingested editions on their overlapping months shows the same effect in the other dollar-and-count series (June 2026 median per-employee spend $10.59 → $10.94; June model-serving adoption 5.77% → 6.00%) while the binary vendor-share series barely moves (June overall adoption 54.96% → 54.95%). The consequence is mechanical and easy to miss: any month-over-month change computed off the latest point of this index is biased downward, because the newest month is the least complete. The arithmetic is worked at Firm AI-Spend Intensity and Headcount Growth.

Why it matters to this wiki#

Ramp is the second member of the platform-administrative-records instrument family catalogued at Telemetry vs. Survey Measurement — whoever runs a rail can count what crosses it, for free, in near-real time, for exactly the population that transacts there. Indeed brokers vacancies and counts labor demand; Ramp brokers payments and counts who firms pay for AI. Both are published by the operator's own research arm, which folds the vendor incentive and the instrument into one party. Against surveys whose adoption estimates for the same period span 18%–78%, a payment trace is behavior rather than feeling; against telemetry, it sees purchase and not use.

Connections#

• Firm AI-Spend Intensity and Headcount Growth — the instrument's main home: the Revelio-linked headcount panel, the monthly index's adoption and per-employee-spend cuts, the aperture argument, and the cross-edition revision finding

• Telemetry vs. Survey Measurement — where the instrument family is defined and Ramp is weighed against surveys, telemetry and randomization

• The Open-Weight Frontier Gap — the demand-side cut and the limit that bites hardest there: a payment rail cannot see self-hosted open weights, and a routing platform sells closed models too

• Cost-per-Task Over Cost-per-Token — the blended token-price and model-tier series, read as a buyer-side verdict on model selection

• Standardize the Infrastructure, Not the Tools — the organizational reading of the same tier series: firms imposing company-wide model defaults, outcome observed and mechanism invisible to this rail

• Anthropic, OpenAI — the two vendors whose business-adoption race this index is most often quoted for; the crossover it dates to May 2026 is the corpus's first cross-instrument agreement on that reordering

• Cited as a benchmark by a third party (2026-09-22). ICONIQ's 2026 State of Scaling uses the Ramp AI Index as its source for AI-adoption breadth — 7.5% of U.S. businesses paying for AI in January 2023 rising to 50%+ by July 2026 — and reproduces Ramp's own understatement caveat verbatim (free tools and spend through employees' personal accounts are invisible to a corporate-card panel). A growth investor reaching for a card-spend index rather than a survey when it needs an adoption number is a small but real signal of where the instrument now sits; see Telemetry vs. Survey Measurement.

Sources#

• A New Look at AI's Impact on Jobs: Firm-Level AI Spending and Workforce Adjustment — Kharazian, Simon & Stevens (Ramp × Revelio Labs, June 2026), empirical: the research use of the classifier. Full treatment and evidence note at Firm AI-Spend Intensity and Headcount Growth
• Ramp's latest data on China vs. the American AI Labs — Ara Kharazian, Ramp AI Index monthly letter, 2026-07-08, empirical. Four Datawrapper charts with no static fallback; the ingest pass recovered each chart's dataset.csv
• September 2026 Ramp AI Index: Cracks in the AI thesis, part 2 — Ara Kharazian, September 2026 Ramp AI Index: Cracks in the AI thesis, part 2, 2026-09-09, empirical. Five recovered Datawrapper datasets; two provenance traps (a revised-vintage July spend cell in the CSV against an as-reported figure in the prose, and a CDN-cached price series whose tail trails the prose) are documented in the raw's note: field and in the source notes
• Ramp's methodology — denominator definitions, the eligibility rule and the vendor classifier — is documented only at ramp.com/data/ai-index, which is not in the corpus
• 2026 State of Scaling: The Great Sorting — ICONIQ Venture & Growth, 2026 State of Scaling: The Great Sorting (September 2026, empirical). Third-party citation only: the AI-adoption series (7.5% → 50%+) and the index's methodology box, reproduced on p.17