Enterprise generative AI spend hit USD 37 billion in 2025, tripling in a single year, yet only about 5.5% of organizations can attribute 5% or more of their EBIT to AI. That gap between deployment and measurable return is the defining fact of enterprise AI in 2026. Adoption itself is close to saturated at 88% of organizations, model share has been upended with Anthropic taking 40% of enterprise usage against OpenAI’s 27%, and buying has decisively beaten building at 76% of use cases. The figures below come from Menlo Ventures’ enterprise survey, McKinsey’s State of AI global survey, and Stanford HAI’s 2026 AI Index.
TL;DR
- Enterprise generative AI spend reached USD 37 billion in 2025, up 3.2 times year over year (Menlo Ventures)
- That is up from USD 11.5 billion in 2024 and USD 1.7 billion in 2023 (Menlo Ventures)
- Enterprise AI now accounts for roughly 6% of the global SaaS market (Menlo Ventures)
- Applications took USD 19 billion and infrastructure USD 18 billion (Menlo Ventures)
- Anthropic leads enterprise LLM usage at 40%, ahead of OpenAI at 27% and Google at 21% (Menlo Ventures)
- In coding specifically, Anthropic holds 54% against OpenAI’s 21% (Menlo Ventures)
- Open-source models fell to 11% of enterprise share, from 19% (Menlo Ventures)
- 76% of use cases are now bought rather than built, reversing 2024’s split (Menlo Ventures)
- 88% of organizations use AI in some form (McKinsey, Stanford HAI)
- 79% use generative AI in at least one business function (McKinsey)
- Only 39% attribute any EBIT impact to AI, and most of those under 5% (McKinsey)
- Roughly 6% qualify as AI high performers (McKinsey)
- Only 16% of enterprise production systems are true agent deployments (Menlo Ventures)
- Global corporate AI investment reached USD 581.7 billion, up 130% (Stanford HAI, 2026)
1. The Spend: Tripling, Then Tripling Again
Three consecutive years of roughly 3x growth is not a normal software adoption curve. Spend went from USD 1.7 billion in 2023 to USD 11.5 billion in 2024 to USD 37 billion in 2025, and the category now represents about 6% of all global SaaS spending after existing commercially for barely three years. For comparison, Stanford HAI puts total global corporate AI investment, including infrastructure and research rather than just enterprise software purchasing, at USD 581.7 billion.
| Metric | Value | Source |
|---|---|---|
| Enterprise generative AI spend, 2025 | USD 37 billion | Menlo Ventures |
| Enterprise generative AI spend, 2024 | USD 11.5 billion | Menlo Ventures |
| Enterprise generative AI spend, 2023 | USD 1.7 billion | Menlo Ventures |
| Year-over-year growth | 3.2x | Menlo Ventures |
| Share of global SaaS market | approx. 6% | Menlo Ventures |
| Global corporate AI investment, 2025 | USD 581.7 billion, up 130% | Stanford HAI, 2026 |
| Generative AI investment specifically | USD 170.9 billion, up 404% | Stanford HAI, 2026 |
| Products above USD 1 billion ARR | at least 10 | Menlo Ventures |
| Products above USD 100 million ARR | 50 | Menlo Ventures |
| Survey sample | approx. 500 US enterprise decision-makers | Menlo Ventures |
| Survey field dates | November 7 to 25, 2025 | Menlo Ventures |
The two investment figures measure different things and should not be added or compared: Menlo counts what enterprises pay for AI software, Stanford counts what the world invests in AI overall. Source: Menlo Ventures, 2025 State of Generative AI in the Enterprise.
2. Where the Money Actually Goes
The near-even split between applications and infrastructure is the structural surprise. Applications took USD 19 billion and infrastructure USD 18 billion, with foundation model APIs alone accounting for USD 12.5 billion of the infrastructure side. Inside applications, copilots dominate horizontal spend at USD 7.2 billion, or 86% of that category, while the agent platforms attracting most of the industry’s attention took USD 750 million.
| Metric | Value | Source |
|---|---|---|
| Applications layer total | USD 19 billion | Menlo Ventures |
| Infrastructure layer total | USD 18 billion | Menlo Ventures |
| Foundation model APIs | USD 12.5 billion | Menlo Ventures |
| Model training infrastructure | USD 4.0 billion | Menlo Ventures |
| Horizontal AI applications | USD 8.4 billion, up 5.3x | Menlo Ventures |
| Copilots within horizontal | USD 7.2 billion, 86% | Menlo Ventures |
| Agent platforms within horizontal | USD 750 million, 10% | Menlo Ventures |
| Departmental AI | USD 7.3 billion, up 4.1x | Menlo Ventures |
| Coding within departmental | USD 4.0 billion, 55% | Menlo Ventures |
| Vertical AI | USD 3.5 billion, up nearly 3x | Menlo Ventures |
| Healthcare within vertical | USD 1.5 billion, 43% | Menlo Ventures |
| Legal within vertical | USD 650 million | Menlo Ventures |
Coding at 55% of departmental spend makes software engineering the only function where AI purchasing has reached genuine scale; marketing, customer success, and IT operations each sit under USD 1 billion. Inference economics context sits in our AI inference cost statistics. Source: Menlo Ventures enterprise AI report PDF.
3. Model Market Share Inverted
The provider ranking that held through 2023 has completely reversed. Anthropic reached 40% of enterprise LLM usage, up from 24% a year earlier and 12% in 2023, while OpenAI fell to 27% from 50% in 2023. Google’s climb from 7% to 21% is the third leg of the same shift. The concentration is severe: those three hold 88% between them.
| Metric | Value | Source |
|---|---|---|
| Anthropic enterprise share, 2025 | 40% | Menlo Ventures |
| Anthropic share, 2024 | 24% | Menlo Ventures |
| Anthropic share, 2023 | 12% | Menlo Ventures |
| OpenAI enterprise share, 2025 | 27% | Menlo Ventures |
| OpenAI share, 2023 | 50% | Menlo Ventures |
| Google enterprise share, 2025 | 21% | Menlo Ventures |
| Google share, 2023 | 7% | Menlo Ventures |
| Combined top three | 88% | Menlo Ventures |
| Anthropic share in coding workloads | 54% | Menlo Ventures |
| OpenAI share in coding workloads | 21% | Menlo Ventures |
| Open-source model share | 11%, down from 19% | Menlo Ventures |
| Chinese open-source models | 1% of total LLM usage | Menlo Ventures |
Open-source falling from 19% to 11% cuts against the prevailing narrative, and the 1% figure for Chinese open-source models is strikingly low given their benchmark performance. Open-source context sits in our open-source AI statistics. Source: Menlo Ventures, 2025 State of Generative AI in the Enterprise.
4. Buying Beat Building
The build-versus-buy question was genuinely open in 2024 and is now closed. 76% of use cases were purchased in 2025 against 24% built internally, reversing a near-even 53 to 47 split the prior year. Conversion data explains why vendors won: AI deals reach production 47% of the time against roughly 25% for traditional SaaS, so buyers are getting better outcomes from purchasing than the software industry’s historical baseline.
| Metric | Value | Source |
|---|---|---|
| Use cases purchased, 2025 | 76% | Menlo Ventures |
| Use cases built internally, 2025 | 24% | Menlo Ventures |
| Use cases purchased, 2024 | 53% | Menlo Ventures |
| Use cases built internally, 2024 | 47% | Menlo Ventures |
| AI deal conversion to production | 47% | Menlo Ventures |
| Traditional SaaS conversion to production | 25% | Menlo Ventures |
| Product-led growth share of application spend | 27% | Menlo Ventures |
| Same figure in traditional software | 7% | Menlo Ventures |
| Startup share of applications layer | 63%, up from 36% | Menlo Ventures |
| Startup share in finance and operations | 91% | Menlo Ventures |
| Startup share in sales | 78% | Menlo Ventures |
| Incumbent share of infrastructure layer | 56% | Menlo Ventures |
The split by layer is the useful read: startups are winning applications while incumbents hold infrastructure, which is the opposite of how the last platform shift played out. Source: Menlo Ventures 2025 report announcement.
5. Adoption Is Not Value Capture
This is where the enterprise AI story gets uncomfortable. 88% of organizations use AI and 79% use generative AI in at least one function, but only 39% attribute any EBIT impact to it, and most of those put the figure below 5%. Roughly 5.5% report EBIT impact of 5% or more. Nearly two-thirds have not begun scaling AI across the enterprise, which means the adoption headline is measuring pilots rather than transformation.
| Metric | Value | Source |
|---|---|---|
| Organizations using AI in any form | 88% | McKinsey |
| Using generative AI in at least one function | 79% | McKinsey |
| Organizational adoption, independent measure | 88% | Stanford HAI, 2026 |
| Attribute any EBIT impact to AI | 39% | McKinsey |
| Report EBIT impact of 5% or more | approx. 5.5% | McKinsey |
| Qualify as AI high performers | approx. 6% | McKinsey |
| Have not begun scaling across the enterprise | nearly two-thirds | McKinsey |
| McKinsey survey sample | 1,993 organizations | McKinsey |
| Countries covered | 105 | McKinsey |
| University students using generative AI | 4 in 5 | Stanford HAI, 2026 |
| SWE-bench Verified performance change in one year | 60% to near 100% | Stanford HAI, 2026 |
Two independent surveys landing on 88% organizational adoption is unusually strong corroboration for a headline AI statistic, and it makes the EBIT gap harder to dismiss as a sampling artefact. Sources: McKinsey, The State of AI and Stanford HAI 2026 AI Index.
6. Agents in Production Are Rarer Than Advertised
The word “agent” is doing an enormous amount of marketing work relative to what is deployed. Only 16% of enterprise production systems are true agent deployments, rising to 27% among startups, with most systems remaining fixed-sequence or routing-based workflows built around a single model call. Menlo’s own spend data agrees: agent platforms took USD 750 million against USD 7.2 billion for copilots.
| Metric | Value | Source |
|---|---|---|
| True agent deployments, enterprises | 16% | Menlo Ventures |
| True agent deployments, startups | 27% | Menlo Ventures |
| Agent platform spend | USD 750 million | Menlo Ventures |
| Copilot spend for comparison | USD 7.2 billion | Menlo Ventures |
| Internal-facing use cases | 59% | Menlo Ventures |
| Customer-facing use cases | 41% | Menlo Ventures |
| Developers using AI coding tools daily | 50% | Menlo Ventures |
| Same figure in top-quartile organizations | 65% | Menlo Ventures |
| Reported velocity gains from AI coding tools | 15% or more | Menlo Ventures |
| Ambient clinical scribe market | USD 600 million, up 2.4x | Menlo Ventures |
| Documentation time reduction from scribes | more than 50% | Menlo Ventures |
Ambient scribes are the clearest case of AI producing a measurable operational result rather than a projected one, which is likely why healthcare outspends the next four verticals combined. Voice-interface adoption specifically is covered in our AI voice agents statistics, and the labour-market effects in our tech layoffs statistics. Source: Stanford HAI, 12 takeaways from the 2026 AI Index.
Summary: Enterprise AI by the Numbers
| Metric | Value | Source |
|---|---|---|
| Enterprise generative AI spend, 2025 | USD 37 billion | Menlo Ventures |
| Growth versus 2024 | 3.2x | Menlo Ventures |
| Spend in 2023 | USD 1.7 billion | Menlo Ventures |
| Share of global SaaS | approx. 6% | Menlo Ventures |
| Applications layer | USD 19 billion | Menlo Ventures |
| Infrastructure layer | USD 18 billion | Menlo Ventures |
| Foundation model APIs | USD 12.5 billion | Menlo Ventures |
| Copilot spend | USD 7.2 billion | Menlo Ventures |
| Coding spend | USD 4.0 billion | Menlo Ventures |
| Healthcare vertical spend | USD 1.5 billion | Menlo Ventures |
| Anthropic enterprise LLM share | 40% | Menlo Ventures |
| OpenAI enterprise LLM share | 27% | Menlo Ventures |
| Google enterprise LLM share | 21% | Menlo Ventures |
| Anthropic share in coding | 54% | Menlo Ventures |
| Open-source model share | 11% | Menlo Ventures |
| Use cases purchased versus built | 76% versus 24% | Menlo Ventures |
| AI deal conversion to production | 47% | Menlo Ventures |
| Startup share of applications layer | 63% | Menlo Ventures |
| Organizations using AI | 88% | McKinsey, Stanford HAI |
| Using generative AI in a function | 79% | McKinsey |
| Attribute any EBIT impact | 39% | McKinsey |
| AI high performers | approx. 6% | McKinsey |
| True agent deployments in enterprise | 16% | Menlo Ventures |
| Global corporate AI investment | USD 581.7 billion | Stanford HAI |
Methodology and Sources
- Spend totals, category breakdowns, model market share, build-versus-buy splits, conversion rates, startup share, and agent deployment rates come from Menlo Ventures’ 2025 State of Generative AI in the Enterprise, its third annual edition, based on a survey of approximately 500 US enterprise decision-makers fielded November 7 to 25, 2025 (report, full PDF, announcement).
- Organizational adoption rates, EBIT attribution, high-performer share, and scaling status come from McKinsey’s State of AI global survey of 1,993 organizations across 105 countries (McKinsey), with agentic-era framing from its 2026 follow-up (McKinsey).
- Global investment figures, independent adoption measurement, benchmark performance, and student usage come from the Stanford HAI 2026 AI Index, published April 2026 (report, takeaways, AI Index home).
- Data watch: Menlo’s survey covers US enterprises only and its revenue estimates for private companies are modelled from public data and industry analysis rather than reported, which the firm discloses. Menlo is an active investor in this category, so treat its market-share figures as directionally useful rather than audited. Menlo’s spend figures and Stanford’s investment figures measure different things and must not be combined. McKinsey’s EBIT attribution is self-reported by respondents rather than derived from financial statements. Model market share reflects enterprise API usage patterns among surveyed buyers, not revenue or token volume, and shifts quickly enough that a figure more than two quarters old should be treated as stale.
- Last updated: August 1, 2026. We update this roundup quarterly as Menlo, McKinsey, and Stanford HAI publish new survey waves.