Google's 9% Enterprise AI API Share, Checked

Google's share of enterprise AI API spending fell to 9%, the headline says, and every figure in it matches Menlo Ventures' report. The 21% a year earlier matches. Anthropic's climb from 40% to 55% matches. The trouble starts when you ask what the report measured.

Verdict: half true. Menlo's method note says "LLM market shares represent estimated dollars spent based on proportion of production API usage," so the 9% is an estimate, and no word in the headline tells you so.

The 9% comes from a survey of U.S. buyers

Menlo surveyed nearly 400 U.S. enterprise decision-makers responsible for more than $4 billion in AI spending. Morning Consult fielded the survey from September 8–26, 2026. Respondents reported the share of their AI workloads on each model, "weighted by AI spend."

Menlo then converts those workload shares into dollars. My reading is that a respondent who runs most workloads on Gemini Flash may spend little of the budget there, and the report does not show how the weighting handled it.

No Google revenue figure sits behind the 9%. A secondary analysis of the report says the shares are "not from vendor revenue."

Two qualifiers follow. The sample is American, and the headline never says so. The denominator is enterprise LLM API spend, narrower than "enterprise AI API spending."

Menlo gives no Google dollar figure and no absolute size for the API market. The $176.8B figure, up 5.4x, measures enterprise deployment spend, a different denominator. Coverage still sets the 9% beside the 69%, which is Anthropic, OpenAI and Google combined at $122.7B of deployment dollars.

I infer a lower share is compatible with higher dollars in a market growing that fast. The report does not settle it.

Menlo does offer a cause. Gemini 3 Pro "put the lab at the frontier in November 2025," Google "hasn't matched Anthropic's and OpenAI's flagship models since," and "its API spend now skews toward cheaper Flash models." Menlo places Gemini Flash in its small tier, which fits. It never says how much of the drop from 21% to 9% is lost customers and how much is customers buying the cheaper tier. No supporting data appears.

The same report says Anthropic and OpenAI now account for 86% of enterprise LLM API spend, up from 67% last year. The 86% points the same way as the headline.

The comparison base moved as well. Menlo reclassified prior years, shifting data labeling vendors like Mercor from deployment to development, so the $33B in coverage differs from the $37B last year's edition published for 2025.

Disclosure is not evidence of bias, but it is not a margin of error either

The strongest case for the headline deserves its full weight. Every figure matches the primary. The press release carries the line "*Menlo Ventures is an investor in Anthropic and OpenRouter," and the report spells out method and sample. The analysis I quoted grants "Disclosure is not evidence of bias."

Menlo's text says Google "fell furthest," and I infer a move this large is unlikely to be survey noise. Ramp's AI Index counts businesses rather than dollars and found 43.8% paying Anthropic and 40.6% paying OpenAI, a closer race. The analysis holds both datasets true together. Nothing in the research contradicts the 9%, and no audited lab revenue exists to replace it.

I accept the direction. The same analysis calls the direction of Menlo's findings credible and the magnitudes less certain, which is my claim. Its sentence continues: the lab rankings are "an investor's estimate, not an audited count." Disclosure answers whether Menlo hid something. It gives you no margin of error for the 9%.

The checks offered as confirmation share a source. Menlo's cause is Menlo's assertion, and the three lab shares plus the remainder for xAI and open-source models come from one survey. The Ramp passage gives no Google figure. So the 9% has no independent corroboration in the research.

I am not claiming Menlo tilted anything. The research shows no sign of it, and I do not infer it from the stake.

You might object that every share figure in a young market is an approximation. It is, and the objection stands. What separates this case is a list of omissions you check yourself: the U.S. sample, the denominator, the missing dollars, and the stake the press release disclosed and the coverage dropped.

Each retelling drops another qualifier

Two other write-ups repeat 55%, 31% and 9% with no qualification. One calls the 69% "the market." It also says "Only 22% of enterprises' software budgets were redirected towards AI," where Menlo says 22% of incremental AI spend came from software budgets. The error has nothing to do with Google, but it shows what happens to a survey estimate in secondary coverage.

The number worth asking for next is Google's API revenue in dollars. Menlo's report does not carry it, and neither does any coverage I checked.