From Sanskriti Khandelwal | Product & Market Analysis

Enterprise AI Spend: Menlo Says $177B, a16z and Ramp Never Give a Total

On this page

Menlo Ventures says enterprise AI deployment spend reached $176.8 billion by September 2026. a16z and Ramp, the two reports most often quoted beside it, publish no market total at all. a16z reports about $7 million of LLM spend per large firm. Ramp reports that 55.9% of businesses on its platform pay for AI. The enterprise generative AI spend debate is really three different units.

Key takeaways

  • Only Menlo publishes a total for enterprise AI spend. Its November 2026 report puts deployment at $176.8 billion, stated as of September 2026, from a model informed by a survey of nearly 400 US decision-makers.
  • Menlo's own history moved between editions. Its 2025 total went from $37 billion to $33 billion and its 2024 total from $11.5 billion to $8.5 billion, which lifted 2025 growth from 3.2x to 3.9x.
  • a16z and Ramp give you units, not a market. a16z's 100 Global 2000 executives reported about $7 million of average LLM spend, and Ramp counts firms with a paid AI transaction, at 55.9% in September 2026.
  • Vendor revenue leaves Menlo little slack. Anthropic's third quarter annualised plus OpenAI's reported run rate comes to about $118 billion, consumer revenue included, against Menlo's $122.7 billion for three labs' enterprise spend.

How much do enterprises spend on AI? The only published total is Menlo Ventures' estimate of $176.8 billion in annualised deployment spend as of September 2026. It is a modelled figure from an investor in Anthropic. a16z and Ramp do not size the market, so neither one confirms or contradicts that number directly.

$176.8BEnterprise AI deployment spend as of September 2026, a modelled total. Source: Menlo Ventures, November 2026.
~$7MAverage enterprise spend on LLMs among 100 Global 2000 respondents, per firm. Source: a16z, January 2026.
55.9%Share of US businesses on Ramp paying for AI in September 2026. A count of payers, not dollars. Source: Ramp AI Index.

Enterprise AI spend has one published total and two other units

The headline suggests a disagreement, and the honest finding is stranger. Menlo, a16z and Ramp cannot disagree on the total because only one of them states a total. The other two answer different questions, and writers turn their answers into market claims.

Each report picks a different unit. Menlo sizes annualised dollars across a market. a16z reports average dollars inside a sample of very large firms. Ramp reports the share of its customers that paid an AI vendor, and spend per employee at firms that paid. If you want to know who leads on share of those dollars, that is a separate question, covered in why the enterprise LLM market share trackers name different leaders.

Two more sources belong on the table because readers confuse them with these three. Gartner forecasts all AI spending worldwide. Company filings report what vendors actually recognised as revenue. Together the five form a ladder from broad forecast to audited-style reporting, and the table below sets them side by side.

How each enterprise AI spend source is built
SourceEvidence typeSample or baseUnit reportedMarket total?
Menlo Ventures, 3 Nov 2026Survey plus bottom-up model, public filings, third-party reportsNearly 400 US enterprise decision-makers, 8 to 26 Sep 2026Annualised dollars by layerYes: $176.8B deployment, $102.5B development
a16z, 30 Jan 2026Self-reported survey via an expert network100 VPs and C-level executives, Global 2000, not randomAverage dollars per firm, share of walletNo
Ramp AI Index, Sep 2026 dataObserved card, invoice and ACH paymentsMore than 85,000 US businesses on RampShare of firms paying, spend per employeeNo
Company filings and investor figuresRecognised or annualised revenueOne company at a timeCompany revenue, business and consumer mixedNo, and recognition policies differ
Gartner, Sep 2026Analyst forecastWorldwide, all AI spendingCalendar 2026 dollarsYes, at $2.7T, for a far wider market

Menlo's $177B enterprise AI spend figure and what it counts

Menlo's 2026 State of Enterprise AI is the most useful single document in this comparison, and it should be quoted carefully. The survey is one input. The dollar figure comes from a bottom-up model that Menlo says is informed by survey responses, public filings and third-party reports.

Menlo says the survey respondents are responsible for more than $4 billion in AI spending. That is about 2.3% of the deployment total, if the $4 billion is an annual figure, which the page does not say. So at least 97% of the headline number is inference from outside the sample. That is normal for market sizing, and it is not what most people mean when they write "a survey found".

Deployment and development are separate totals

The $176.8 billion is deployment, meaning the cost of using AI. It splits into $99.5 billion of infrastructure and $77.3 billion of applications. Applications include $34.0 billion of coding tools and $25.0 billion of workplace assistants, with Microsoft Copilot named as the example.

Development is a further $102.5 billion, made up of $87.2 billion of training compute and $15.3 billion of training data and services. Add the two and you get $279.3 billion, a sum Menlo does not headline. Mixing them up changes the number by 58%, so always name the layer when you quote it.

Two scope gaps deserve a note. The page never defines "enterprise", so you cannot tell how much startup, small business or individual developer spend is inside. It also never says whether the dollar total is US or global, although the survey is US only.

A September run rate is not a 2026 total

Menlo's methodology says revenue is shown as annualised run rate, and the headline is dated September 2026. That makes $176.8 billion a pace at a moment, not a sum of invoices across a year. Actual calendar 2026 spend in a fast-growing market will differ from a late-year pace, and it may land well below it because the early months ran lower.

Say "annualised as of September 2026" every time. A slide that reads "enterprises spent $177 billion in 2026" states something the report does not claim.

Menlo restated its own enterprise AI spend history

The figures most often cited from last year's report no longer match this year's report. The December 2025 edition said companies spent $37 billion on generative AI in 2025, up from $11.5 billion in 2024. The November 2026 edition gives $33 billion and $8.5 billion for those same years.

Menlo's enterprise AI spend, as first published and as restated (billions of dollars)
YearDecember 2025 editionNovember 2026 editionChange
2023$1.7B$1.7BNone
2024$11.5B$8.5BDown 26%
2025$37B$33BDown 11%
2025 growth multiple3.2x3.9xFaster after restatement

Dollar figures are Menlo's. Percentage changes are Zan Digital's arithmetic. The 2025 edition's 3.2x is its own; the 2026 edition's 3.9x is its own, on the restated base.

The same two years, one edition apart Menlo Ventures enterprise AI spend, in $ billions December 2025 edition November 2026 edition 2024 $11.5B $8.5B 2025 $37B $33B Lowering the older year more than the newer one raises the 2025 growth multiple from 3.2x to 3.9x. Source: Menlo Ventures, December 2025 and November 2026 editions.
Notice that the restatement cut 2024 harder than 2025. A smaller base makes the same year's growth look steeper.

Menlo gives one explanation, that prior years were reclassified to this year's taxonomy, which moved data labelling vendors such as Mercor from deployment to development. It does not publish a bridge. One piece of arithmetic is suggestive: the 2025 edition itemised $4.0 billion of model training infrastructure, and $37 billion minus $4.0 billion is exactly $33 billion. That is my inference, not Menlo's statement.

Why I would not quote the 5.4x growth rate

The 2026 report says deployment went from $33 billion to $177 billion, a 5.4x increase. The two ends are different objects. One is described as spend in 2025, and the other is dated September 2026 with revenue shown as annualised run rate.

There may also be a scope break. Menlo's 2025 edition excluded inference and model serving, naming AWS, Azure, Google Cloud and Fireworks. Its 2026 infrastructure line includes frontier model inference and inference platforms. The stated reclassification does not say whether the $33 billion was rebuilt to include them. Quote the level, label its date, and leave the multiple out until Menlo publishes the bridge.

a16z measures enterprise LLM spend per firm, not a market

a16z's third annual CIO survey, posted on 30 January 2026, polled 100 VPs and C-level executives in the Global 2000. All had at least $500 million of revenue and 88% had more than $1 billion. It reports that average LLM spend rose from about $4.5 million to about $7 million over 2 years, and that respondents expected about $11.6 million next.

Application spend came in at nearly $6 million against an expected $3.9 million. a16z deserves credit for its disclosure: participants were not randomly sampled, responses are self-reported, and the findings may over-represent firms further along in AI. The survey also leaves out startups by design, and a16z states that it is an investor in OpenAI.

a16z's own series does not chain cleanly

In March 2024, a16z said average 2023 spend across foundation model APIs, self-hosting and fine-tuning was $7 million, across dozens of companies it spoke to. In January 2026 it reports an average that rose to about $7 million. The same number appears for different years, samples and definitions, so do not plot a16z's editions as a single trend line. The wider problem of comparing such surveys is set out in the piece on the methodology gap in AI ROI surveys.

Scaling a16z's average gives a rough market check

Because a16z gives a per-firm average, you can scale it, crudely. Add about $7 million of LLM spend to nearly $6 million of application spend and multiply by 2,000 firms. The result is about $26 billion, or 79% of Menlo's restated $33 billion for 2025.

That is a surprisingly close match for two unrelated methods. Forward expectations are where the gap opens. Multiply the expected $11.6 million by 2,000 and you get about $23.2 billion of LLM spend, while Menlo puts three labs alone at $122.7 billion, 5.3 times as much. Part of that difference is firms outside the Global 2000, and part is that a January expectation was overtaken by the year.

The calculation is mine and it is rough. a16z warns that its sample tilts toward advanced adopters, which inflates the average, while restricting it to the Global 2000 leaves out every smaller buyer. Use it to sanity-check magnitude, never as an estimate to publish on its own.

Ramp card data counts payers and shows a wide spend spread

The Ramp AI Index is the only source here built on observed transactions. A firm counts as adopting AI if it has a positive transaction amount for an AI product or service in a month, identified from merchant names and line items across card, invoice and ACH payments. In September 2026 that was 55.9% of US businesses on Ramp, split into 65.7% of large firms, 62.5% of medium firms and 50.0% of small firms.

How Ramp turned that dataset into a distribution engine is covered in the analysis of the Ramp AI Index as a research marketing asset. For this comparison, what matters is what the rail can and cannot see.

What card data sees and misses

Ramp's page says its results likely underestimate adoption and spend, because free tools produce no transaction and employees may pay through personal accounts. The population is Ramp's own customer base, not a random draw of US firms. Its stated sample is more than 85,000 businesses, while the notes on the St. Louis Fed's FRED series still say more than 70,000.

Two gaps matter for anyone tempted to derive dollars. Ramp does not say whether a model bought through AWS, Azure or Google Cloud is counted against the model vendor. It also reports AI at 5.77% of business spend excluding payroll without defining the denominator. Ramp publishes no market total, and I would not build one from these fields.

The median firm and the top 1% are 609x apart

Ramp's most underused data is spend per employee. Among firms paying for AI, the median firm spent $15.70 per employee per month in September 2026. The top 10% group has a median of $813.43, and the top 1% group a median of $9,567.53.

AI spend per employee per month, firms paying for AI Ramp AI Index, September 2026. Each group's value is the median within that group. Median firm $15.70, or $188,400 a year at 1,000 staff Top 10% group $813.43, or $9.8M a year at 1,000 staff Top 1% group $9,567.53, or $114.8M a year at 1,000 staff The 1,000-staff annual figures are Zan Digital's illustration: value x 1,000 x 12. Source: Ramp AI Index, data through September 2026.
The median bar is nearly invisible at this scale, and that is the point. Any average of enterprise AI spend is mostly a statement about the top of this distribution.

At 1,000 employees, those medians convert to about $188,400, $9.8 million and $114.8 million a year. The top 10% figure sits right at a16z's Global 2000 average, at a company a fraction of the size. Averages in this market are dominated by a few heavy spenders, which is why a survey's total swings on who answered it.

Vendor revenue is the check on every enterprise AI spend survey

Revenue that companies report is the closest thing to a hard floor, and it exposes the pressure on Menlo's top line. Menlo says Anthropic, OpenAI and Google together take $122.7 billion, or 69% of deployment dollars, as of September 2026.

Now compare what two of those labs report. CNBC's reading of Anthropic's amended S-1 puts third quarter revenue at roughly $17 billion, about $68 billion annualised. CNBC also confirmed that OpenAI told investors it hit roughly $50 billion in annualised revenue at the end of September. Together that is about $118 billion.

Menlo's enterprise lab total against what two labs report, all customers Annualised, $ billions. Anthropic is Q3 revenue x 4. OpenAI figures are for end of September. Menlo estimate Anthropic OpenAI Three labs, enterprise Menlo: Anthropic, OpenAI, Google$122.7B Two labs, as reported $68B $50B$118B Two labs, OpenAI gross $68B $68B$136B The two lower bars include consumer subscriptions and exclude Google. The top bar is enterprise only and includes Google. Source: Menlo Ventures, November 2026; CNBC, 8 October and 2 November 2026. Sums are Zan Digital's arithmetic.
Read the dashed line. On reported revenue, two labs' entire business, consumers included, falls short of Menlo's enterprise-only figure for three labs.

Those two companies' whole revenue, consumer subscriptions included, comes in below Menlo's enterprise-only line for three labs. There are ways to close the gap, because Google is missing from the sum and a September pace can exceed a quarterly average. There is still not much room.

Accounting decides how much room. CNBC reports that OpenAI's previously circulated $68 billion figure included gross revenue from partners, to allow a more direct comparison with Anthropic. On that basis, the two labs total about $136 billion. The Menlo figure reconciles comfortably only on a gross basis, where a cloud reseller's sale counts in full. The mechanics of the S-1 itself are worked through in the reconciliation of Anthropic's S-1 loss.

I read this as a modest warning, not a refutation. Menlo's total could be right as a measure of what buyers pay, including resellers' margin. It is not the same as revenue the labs keep. Anyone quoting it as lab revenue overstates the labs.

Which enterprise AI spend number to cite for what

Pick the source that matches the question, and write the label into the sentence. My rule is simple: the number should travel with its unit, its date and its owner, or it should stay off the slide.

Which source answers which enterprise AI spend question
Your questionCiteWrite it asDo not use
How big is enterprise AI use?Menlo"$176.8B annualised as of September 2026, Menlo Ventures estimate""Enterprises spent $177B in 2026"
How much does a large firm spend on models?a16z"About $7M average LLM spend among 100 Global 2000 respondents, January 2026"a16z's figure multiplied into a published market size
How many firms pay for AI at all?Ramp"55.9% of US businesses on Ramp, September 2026""55.9% of US businesses"
How unequal is spend?RampMedian $15.70 per employee per month, paying firms onlySurvey averages
What does a lab actually earn?Filings and confirmed investor figuresPeriod, basis and whether partner revenue is grossMenlo's vendor dollar lines as revenue
What is all AI spend including data centres?Gartner"$2.7T worldwide in 2026, Gartner forecast, September 2026"Beside Menlo as if they measure one market

Gartner's September update lands at $2.7 trillion, up 49.5%, roughly 15 times Menlo's deployment figure. That is not a disagreement either. The Gartner number spans the wider AI economy, so a comparison to Menlo says only how far apart two definitions sit. My view is that Menlo is still the right citation for enterprise spend, provided the sentence names the firm and its interest.

Where this comparison of enterprise AI spend reports is weakest

Three of the checks in this post are my own arithmetic on other people's numbers. They deserve the same scepticism I have applied to Menlo.

Menlo may be closer to right than the check implies

Anthropic's run rate was growing fast, so a third quarter average understates its September pace. Anthropic told investors its run rate hit $65 billion at the end of July, and a higher figure at the end of September would close the gap. Google's enterprise revenue is also absent from my sum. The chart compares unlike things on purpose, and it establishes tightness rather than error.

Menlo also publishes more of its method than most market sizers. A restatement with a stated reason is better behaviour than silent revisions, and it should not be read as proof of unreliability on its own.

The evidence here has real gaps

The a16z survey is 10 months old, and it does not state the calendar year of its averages. I did not read Anthropic's S-1 directly; its third quarter and annualised treatment here rely on CNBC's report of the amended filing. OpenAI's $50 billion is an investor figure confirmed by press, not a filing. I could not confirm how Anthropic's filing treats gross revenue from cloud partners, and Gartner's press release was not accessible, so that figure comes from CIO.com's coverage.

Finally, every source here has an interest. Menlo backs Anthropic, a16z backs OpenAI, and Ramp sells the spend platform that creates its data. None of that makes the numbers false, and it should change how much weight a single report carries.

Frequently asked questions

How much do enterprises spend on generative AI?

Menlo Ventures estimates enterprise AI deployment spend at $176.8 billion annualised as of September 2026, plus $102.5 billion of development spend. That is the only published market total among the reports people usually compare. It is a modelled estimate from a firm that invests in Anthropic, so it should be cited with the date, the layer and the name of the estimator.

What is the Menlo Ventures enterprise AI report methodology?

Menlo's 2026 report combines a survey of nearly 400 US enterprise decision-makers, fielded by Morning Consult from 8 to 26 September 2026, with a bottom-up model informed by public filings and third-party reports. Revenue is shown as annualised run rate. Prior years were reclassified, which moved the 2025 total from $37 billion to $33 billion and the 2024 total from $11.5 billion to $8.5 billion.

Does a16z publish a total enterprise AI spend figure?

No. a16z's January 2026 CIO survey reports averages from 100 Global 2000 executives, about $7 million of LLM spend per firm and about $11.6 million expected next, plus wallet shares. It offers no market total. The sample was not random and may over-represent advanced adopters, so multiplying its averages into a market size gives only a rough check on magnitude.

What does the Ramp AI Index measure?

The Ramp AI Index measures the share of US businesses on Ramp with a paid AI transaction in a given month, reaching 55.9% in September 2026. It uses card, invoice and ACH payments from more than 85,000 businesses. It also reports spend per employee at paying firms. It excludes free tools and personal accounts, and it does not estimate total market spending.

Which enterprise AI spend number should I cite?

Use Menlo for the size of enterprise deployment spend, with the date and investor interest stated. Use a16z for what large firms report spending per firm. Use Ramp for the share of businesses paying and the spread of spend per employee. Use company filings for revenue. Use Gartner only when you mean all AI spending worldwide, including infrastructure.

Before you put an enterprise AI spend number on a slide

Run a 10-minute check on the next AI market figure you plan to use. Write down its unit, its date, its owner and whether it was observed, surveyed or modelled. If you cannot fill all four boxes from the source itself, replace the figure with one you can.

Then make your own number. Pull 12 months of AI line items from your card, invoice and cloud bills, divide by headcount, and place yourself against Ramp's $15.70 median. That single figure will tell your board more than any market total, and the gaps in FinOps tooling for AI spend explain where those bills tend to hide.

References

  1. Menlo Ventures, 2026: The State of Enterprise AI, 3 November 2026. Retrieved 9 November 2026. Used for deployment and development totals, the survey sample, restated figures and vendor dollars. Anthropic is a Menlo portfolio company.
  2. Menlo Ventures, 2025: The State of Generative AI in the Enterprise, 9 December 2025. Retrieved 9 November 2026. Used for originally published totals, scope exclusions and category lines.
  3. a16z, Leaders, gainers and unexpected winners in the Enterprise AI arms race, 30 January 2026. Retrieved 9 November 2026. Used for per-firm averages and sample disclosures. a16z is an investor in OpenAI.
  4. a16z, 16 Changes to the Way Enterprises Are Building and Buying Generative AI, 21 March 2024. Retrieved 9 November 2026. Used for the earlier $7 million average.
  5. Ramp, Ramp AI Index, data through September 2026. Retrieved 9 November 2026. Used for adoption, per-employee spend and definitions.
  6. CNBC, Anthropic reports $17 billion in Q3 revenue in revised S-1, 2 November 2026. Retrieved 9 November 2026. Secondary report of the S-1/A filed with the SEC; the citation should be upgraded to the filing.
  7. CNBC, Nvidia, Oracle, CoreWeave and other AI stocks sink on OpenAI revenue report, 8 October 2026. Retrieved 9 November 2026. Used for OpenAI's $50 billion and $68 billion figures and Anthropic's July run rate.
  8. CIO.com, AI spend will jump 49.5% in 2026, says Gartner, 23 September 2026. Retrieved 9 November 2026. Secondary report of Gartner's forecast.

Last verified 9 November 2026. The weakest part of this source base is that every report comes from a party with a commercial interest in the market it describes, and that the revenue check rests on press accounts of the filing rather than the filing itself.

SK
Sanskriti Khandelwal
Writes for Zan Digital about AI product economics, B2B software markets and what the numbers behind vendor claims actually say.

Related reading