Anthropic: 55% at Menlo, 60.5% at Ramp. Two Numbers, Two Rulebooks

In short

• Menlo, an investment fund, estimates that Anthropic accounts for 55% of what enterprises spend on language-model APIs, versus 40% a year earlier. It is an estimate built from a survey of nearly 400 US decision-makers, run in September 2026.
• Ramp counts 60.5% for Anthropic in the token spend of its customers who use its tracking tool, week ending 23/10/2026, versus 51.3% in the week ending 25/09.
• Another Menlo line: open source accounts for 19% of the tokens consumed but 4% of model spend. A million tokens costs 0.13 $ on average there, versus 0.70 $ for closed models.

The smoking gun

Picture a Monday morning. You are preparing a committee. Two briefing notes on your desk. The first: "Anthropic, 55% of enterprise API spend." The second: "Anthropic, 60.5% of token spend." What do you do? Write 58% as the average? No.

Both notes talk about money, companies and Anthropic. They count neither the same dollars, nor the same companies, nor the same period. The rest of this article lives in that gap.

Menlo: a survey, a calculation model, a company in its portfolio

Menlo Ventures is an investment fund. Its annual report rests on a survey of nearly 400 US enterprise decision-makers who together handle more than 4 billion dollars of AI spending, run by the polling firm Morning Consult from 08/09/2026 to 26/09/2026.

Here is what the 55% measures. Market shares of language models, says the methodology, are estimated dollars spent based on the proportion of production API usage. Respondents reported what share of their AI workloads goes through each model, weighted by their AI spend. Anthropic's 55% is therefore a share of an estimated API spend; Menlo gave 40% last year. OpenAI goes from 27% to 31%, Google from 21% to 9%.

And one detail the report displays itself: Anthropic appears among Menlo Ventures' portfolio companies, marked with an asterisk in the text, and Menlo lists it among the companies it partners with. I put it on the table, nothing more. For Anthropic's financial context, see our rundown of what we know about Anthropic's IPO.

Ramp: usage data, not a survey

Ramp is a platform for corporate cards and bill payment. Its AI Index page measures AI adoption every month and, for tokens, also gives the latest complete week. What is measured: among the companies that use its product for tracking token spend, what share of their token spend goes to each AI provider.

These figures come from a tool that imports usage and billing data from the AI providers the customer has connected. Ramp states that this sample is not representative of everything its customers spend on AI, even if the results point the same way. The "85,000 companies" cited in the methodology describe the index as a whole, not that group, which the page does not size.

Result, week ending 23/10/2026 versus the week ending 25/09:

• Anthropic: 60.5% of token spend, versus 51.3%;
• OpenAI: 35.1%, versus 44.2%;
• in token volume: Anthropic 59.8% (56.1% a month earlier).

Total token spend in the sample rose 8.9% and volume 37.1% over the period. No total dollar amount here either.

Why we neither average nor add

Two meters, side by side:

Both point the same way: Anthropic's share rises (+15 points at Menlo, +9.2 points at Ramp). But it is not the same measure repeated twice, so it is not double the proof.

One last reading trap: the number 55 turns up three times in these pages, for three unrelated things.

• At Menlo, 55% is Anthropic's share of enterprise API spend.
• Still at Menlo, 55% is also the share of the extra money put into AI that comes from new budgets, rather than from existing software or labor budgets.
• At Ramp, 55.9% is the share of customer companies that paid for AI in September 2026.

The bill in dollars

Back to the rule: a percentage without an amount is worth nothing. Here is what the two sources give, biggest to smallest.

• Menlo estimates 176.8 billion dollars as what enterprises spend on using AI ("deployment") as of September 2026, versus 33 billion in 2025. Menlo writes "5.4x"; our division gives the same result.
• Within that total, Anthropic, OpenAI and Google share 122.7 billion, or 69% (50% a year earlier).
• Anthropic's "55%" is a share of API spend only. Menlo does not give, in this report, the dollar amount of that API spend. So do not multiply 55% by 176.8 billion: the two numbers do not have the same base.

Ramp, for its part, gives dollars (apart from the price per million tokens) only per employee, across all AI spending and not only tokens: in September 2026, 15.70 $ a month for the median company paying for AI, and 9,567.53 $ for the 1% of companies that spend the most (median of that group). If your company had 100 employees: 1,570 $ a month on one side, 956,753 $ on the other. I already ran into that benchmark in our rereading of Apollo's report on AI outside tech. Menlo also notes that a small group of heavy users carries a disproportionate share of AI spend.

Open source: 19% of usage, 4% of spend

The other line of the anchor. At Menlo, open source (models whose weights are public) accounts for 19% of tokens, versus 11% a year earlier. Its share of model spend falls from 6% to 4%. 19% is a share of tokens, the unit of text an AI bills. 4% is a share of the dollars spent on models.

Menlo sets the prices side by side: a million tokens costs 0.13 $ on average for open source, 0.70 $ for closed models (Menlo speaks of a blended price). If you consume 1 billion tokens: 130 $ on one side, 700 $ on the other. Ratio of the two prices: 5.4 to 1. Within open-source spend, Menlo puts Chinese models at 89%, with Kimi at 61%. To see what Kimi K3 demands in practice, see our reading of the repo that runs Kimi K3 in 8 GB of RAM.

What we don't know yet

• The amount of API spend. Menlo gives 55% without saying how many dollars. Without that, there is no way to put a dollar figure on Anthropic's weight.
• The size of Ramp's sample. The page does not say how many companies are in the token measure.
• The following weeks. As of 08/11/2026, the Ramp page still shows the week ending 23/10 as the latest complete week.

Verdict: Anthropic gains ground in both sources. How many dollars, across which companies, neither says.

Key terms

• Token: the unit of text an AI reads or writes, and that its provider bills.
• API spend: money paid to call an AI model directly from a program, rather than through a ready-made app.
• Open source (open-weights models): models whose weights are published; a company can run them itself or at a host.
• Deployment (at Menlo): spending to use AI (infrastructure and applications), as opposed to developing models (training compute and data).
• Share of token spend (at Ramp): the share an AI provider represents in the dollars that customers of Ramp's tool spend on tokens.

Source sheet

• Menlo: annual report 2026: The State of Enterprise AI, Menlo Ventures, published 03/11/2026, by Tim Tully, Joff Redfern, Deedy Das and Derek Xiao; copy of 04/11/2026 12:38 UTC, reread online on 08/11/2026 around 06:40 UTC, text unchanged.
• Ramp: Ramp AI Index page (Ramp Economics Lab), monthly data through September 2026 and weekly data through the week ending 23/10/2026; read on 06/11/2026 around 17:05 UTC, reread on 08/11/2026 around 06:40 UTC: same figures, no more recent week.
• Links: Menlo report · Ramp AI Index

Source passages

Pages read on 08/11/2026 (dated copies in the article folder). Amounts are in US dollars.

Show the 13 sources

Ref.	Quote (original language)	URL
S1↑1↑2↑3↑4↑5↑6+2	"Enterprise AI deployment spend scaled to $177B by September 2026, up from $33B last year." ; "Anthropic’s share of enterprise API spend increased from 40% to 55%, OpenAI’s from 27% to 31%, while Google’s fell from 21% to 9%. Open-source remains 4% of spend despite 19% of usage." ; "By September 2026, spending reached $176.8B, up 5.4x from last year." ; "*Menlo Ventures portfolio company"	https://menlovc.com/perspective/2026-the-state-of-enterprise-ai/
S2↑1↑2↑3↑4↑5↑6	"Anthropic widened last year’s lead, growing from 40% to 55%" ; "OpenAI also gained share, from 27% to 31%" ; "Google fell furthest, from 21% to 9%." ; "We estimate Anthropic, OpenAI, and Google command $122.7B of enterprise spend—69% of all deployment dollars, up from 50% last year"	idem
S3↑1↑2↑3	"It draws on a survey of nearly 400 U.S. enterprise decision-makers responsible for more than $4 billion in AI spending, fielded by Morning Consult from September 8–26, 2026." ; "LLM market shares represent estimated dollars spent based on proportion of production API usage. Survey respondents reported the share of their AI workloads using each model, weighted by AI spend." ; "Token economics ($/M tokens) come from a bottoms-up model of token growth [...] triangulated against third-party sources including OpenRouter, Ramp, and Vercel."	idem
S4↑1↑2↑3↑4↑5	"Open-source now handles about one in five tokens but accounts for only about one in 25 dollars of model spending" ; "From September 2025 to September 2026, its share of tokens rose from 11% to 19%, while its share of spending fell from 6% to 4%" ; "The average open-source token fell 3.3x in price to $0.13 per million, while the blended price of closed-model tokens fell 24% to $0.70." ; "Chinese models grew to 89% of open-source spend, led by Kimi (61%)"	idem
S5↑	"Anthropic’s* Opus 4.5" ; "*Menlo Ventures portfolio company" ; "we’re fortunate to partner with many of the companies at the forefront of this wave, including Anthropic at the frontier"	idem
S6↑1↑2	"55% of the incremental AI spend" (new budgets) ; "a small group of power users accounts for a disproportionate share of AI spend"	idem
S7↑1↑2↑3↑4↑5↑6	"In the week ending October 23, 2026, Anthropic captured the largest share of AI token spending among businesses on Ramp, at 60.5%." ; "Share of token spend: Anthropic 60.5% 51.3% +9.2 pts" ; "Share of token spend: OpenAI 35.1% 44.2% -9.0 pts" ; "Share of token volume: Anthropic 59.8% 56.1% +3.7 pts" ; "Total token spending rose 8.9%. Total token volume rose 37.1%."	https://ramp.com/data/ai-index
S8↑1↑2↑3↑4	"The product imports usage and billing data from connected AI providers and records costs using provider and model identifiers. It represents a sample of Ramp businesses that spend on AI tokens and use the TSM product. It is not representative of Ramp’s total AI spend volume, though results are directionally similar." ; "Our sample includes more than 85,000 American businesses"	idem
S9↑1↑2	"In September 2026, AI adoption reached 55.9% of U.S. businesses on Ramp" ; "Among firms paying for AI, the top 1% spent a median of $9,567.53 per employee per month on AI, compared with $15.70 for the median firm."	idem
R1↑	Kira's calculation, S1S2 : 55 − 40 = 15 points (Menlo); 60.5 − 51.3 = 9.2 points (Ramp) ; two different measures, not comparable point to point	— (in-house calculation)
R2↑1↑2	Kira's calculation, S2 : 122.7 / 176.8 = 69.4%, which is Menlo's 69% ; 176.8 / 33 = 5.4	— (in-house calculation)
R3↑1↑2↑3	Kira's calculation, S4 : 1 billion tokens = 1,000 million ; 1,000 × 0.13 $ = 130 $ ; 1,000 × 0.70 $ = 700 $ ; 0.70 / 0.13 = 5.4	— (in-house calculation)
R4↑	Kira's calculation, S9 : 100 × 15.70 $ = 1,570 $ ; 100 × 9,567.53 $ = 956,753 $ per month	— (in-house calculation)