Worldwide AI spending reached $1.5 trillion in 2025. Enterprise GenAI applications grew 3.2x year-over-year to $37 billion — the fastest-scaling software category in history. The headline figures measure different layers: total AI covers hardware, infrastructure, and services; the $37 billion reflects what enterprises are actually running in production. Only 16% of enterprises have reached true agent deployment. The spending curve is accelerating; the production gap is the story behind the numbers.
Enterprise GenAI spending grew 3.2x in a single year — from $11.5 billion in 2024 to $37 billion in 2025. That is not a trend line. That is a compression event. Software categories that grew this fast this quickly include cloud SaaS in 2010–2012 and mobile app spending in 2011–2013. Both of those became infrastructure. The 2025 AI spending data suggests the same trajectory is already in motion.
The headline figures require context. Not all AI spending is the same, and the number most often cited — Gartner's $1.5 trillion — covers a different layer than the $37 billion Menlo Ventures measured. Understanding the difference tells you where AI investment is concentrated and what the production deployment gap actually means.
What the $1.5 trillion AI spending figure covers
Gartner forecasts worldwide AI spending at $1.5 trillion for 2025.[¹] This figure covers the full AI investment stack: hardware (servers, GPUs, specialized AI chips), software (AI platforms, models, APIs), professional services (implementation, consulting, training), and infrastructure (data centers, networking, cloud capacity).
The $1.5 trillion is a supply-side figure. It measures how much the global economy is investing to build AI capability — the infrastructure layer, the foundation model training runs, the enterprise software integrations.
Gartner separately forecasts worldwide generative AI spending at $644 billion for 2025.[²] This is a narrower measure — software and services that specifically incorporate generative AI capabilities, from consumer applications to enterprise deployments.
The Menlo Ventures figure of $37 billion measures something more specific: what enterprises are spending on GenAI applications in production — tools and workflows actually running in their operations, not in pilots or evaluation.[³]
| Figure | Source | What it measures |
|---|---|---|
| $1.5 trillion | Gartner 2025 | Total global AI spending (hardware + software + services + infrastructure) |
| $644 billion | Gartner 2025 | Worldwide generative AI software and services spending |
| $37 billion | Menlo Ventures 2025 | Enterprise GenAI applications in production |
| $12.5 billion | Menlo Ventures 2025 | Foundation model API access (subset of the $37B) |
The gap between the $644 billion and $37 billion is significant. It reflects how much of the GenAI software spend goes to infrastructure, model training, and tools not yet in operational use. The $37 billion is the running tab — what enterprises are paying to run AI in their workflows today.
The $1.5 trillion and the $37 billion measure different things. Total AI spending covers the full infrastructure buildout — chips, data centers, model training. Enterprise GenAI applications measure what businesses are actually running in production. Both are accurate. Neither describes the other.
Where the $37 billion in enterprise GenAI went
The $37 billion breaks into two roughly equal halves: applications ($19 billion, 51%) and infrastructure ($18 billion, 49%).[³]
Applications layer — $19 billion. Enterprise GenAI application spending covers tools teams use in daily work. The largest subcategory is coding tools at $4 billion, reflecting how broadly developer productivity tooling has been adopted. Horizontal AI — tools that apply across multiple functions — accounts for $8.4 billion. Departmental AI (tools built for a specific function like marketing, HR, or customer success) accounts for $7.3 billion. Vertical AI (tools built for specific industries) accounts for $3.5 billion, led by healthcare at $1.5 billion.
Infrastructure layer — $18 billion. Foundation model API access is the dominant category at $12.5 billion — the cost of calling Claude, GPT-4o, Gemini, and other models through API endpoints to power enterprise applications. Model training infrastructure accounts for $4 billion; AI-specific infrastructure for $1.5 billion.
The applications-to-infrastructure ratio — roughly 51/49 — reveals where enterprise investment is at in 2025. The infrastructure layer is still consuming nearly half the total. As foundation model costs continue to fall and access commoditizes, more of the spending will shift toward applications. The current balance reflects a market still building the plumbing.
One structural shift the Menlo Ventures data captures: 76% of enterprise AI use cases are now purchased rather than built internally, up from 53% in 2024.[³] The direction of travel is toward buying implementation rather than building it. In 2024, nearly half of enterprises were building AI internally. In 2025, three-quarters are buying it. Implementation partner selection has become more consequential than technology selection.
How model market share shifted in a single year
The enterprise model landscape moved significantly in 2025. Anthropic holds 40% of enterprise GenAI market share, up from 24% in 2024. OpenAI holds 27%, down from 50% in 2023. Google holds 21%, up from 7% in 2023.[³]
The shift from OpenAI dominance to a multi-model enterprise environment happened in roughly 18 months. In the coding category specifically — the largest single application subcategory at $4 billion — Anthropic's share reaches 54%, with OpenAI at 21%.
Foundation model API access represents $12.5 billion of the $37 billion total — 34% of all enterprise GenAI spending goes directly to model inference. That share is expected to compress as competition increases and per-token costs continue to fall. In 2024, foundation model API costs fell approximately 10x from 2023 levels. The infrastructure cost drop is what enabled the $37 billion application market to emerge.
For the broader context on where AI is being deployed across industries, see which industries use AI agents.
From budget to production — what the conversion data shows
AI deployments reach production at 47% — nearly twice the 25% conversion rate of traditional SaaS implementations, according to Menlo Ventures.[³] The higher conversion rate reflects AI's adaptability to specific workflows rather than requiring the business to adapt to the tool.
But the production conversion data does not tell the whole deployment picture.
Within the organizations that have reached production, most deployments are fixed-sequence workflows or prompt-based customization rather than true agent systems. True agent deployments — where the system takes multi-step autonomous actions across tools — exist in only 16% of enterprises.[³] Startups are ahead: 27% of startups have reached true agent deployment versus 16% of enterprises.
The production deployment gap is the most strategically significant number in the 2025 spending data. $37 billion in enterprise GenAI spending, and only 16% of enterprises have deployed agents in production. The investment is flowing. The operational deployment is early.
Enterprise GenAI spend grew 3.2x in a single year. The floor is rising, not the ceiling.
The product-led growth dynamic amplifies the adoption picture. Menlo Ventures found that 27% of AI application spend occurs through product-led growth — individuals adopting tools within organizations without centralized IT procurement, at 4x the rate of traditional software.[³] Shadow adoption — AI tools in use that IT has not inventoried — may represent 40% of total enterprise AI usage. The reported figures likely undercount actual deployment.
What the spending curve means for a business deciding now
The $1.5 trillion headline and the $37 billion production figure tell a specific story about timing.
Infrastructure investment is accelerating and costs are dropping simultaneously. Foundation model API costs fell 10x between 2023 and 2024. The per-unit cost of running AI workflows continues to fall as model providers compete on price. This dynamic — high and rising investment, falling per-unit cost — characterizes a technology moving from early adoption to mainstream.
For a service business evaluating whether to implement AI agents now or in 12 months, the data suggests a few things. The cost of waiting is not standing still — other firms in the same market are implementing while 16% of enterprises have true agents deployed. That gap will close. The businesses that build operational agent workflows now are building institutional knowledge about what works, where the edge cases are, and how to extend the system — knowledge that is not transferable from watching the market from the outside.
The ROI window for early implementation concentrates in the next 12–18 months, when agent deployment is still differentiated. See AI agent ROI statistics for the research on what businesses are actually measuring from production deployments.
For a practical first step, see what is an AI agent for a grounded overview of what these systems actually do before evaluating whether the investment timing makes sense for your business.
Frequently asked questions
How much is spent on AI globally in 2025? Gartner forecasts worldwide AI spending at $1.5 trillion for 2025, covering hardware, software, services, and infrastructure. Generative AI accounts for $644 billion of that total. Enterprise GenAI applications in production reached $37 billion, up 3.2x from $11.5 billion in 2024 — the fastest-scaling software category in history.
What percentage of AI spending reaches production? AI deployments convert to production at 47%, nearly twice the 25% rate for traditional SaaS. However, only 16% of enterprises have reached true agent deployments. Most production AI use is prompt-based customization or fixed-sequence workflows. The conversion advantage is real; the agent deployment gap is also real.
How is enterprise AI spending allocated? Enterprise GenAI spending of $37 billion splits roughly evenly between applications (51%, $19 billion) and infrastructure (49%, $18 billion). Within infrastructure, foundation model API access dominates at $12.5 billion. Within applications, horizontal AI at $8.4 billion and coding tools at $4 billion are the largest categories.
Which AI models have the largest enterprise market share? Anthropic holds 40% of enterprise GenAI market share in 2025 (up from 24% in 2024), OpenAI holds 27% (down from 50% in 2023), and Google holds 21% (up from 7% in 2023). In coding specifically, Anthropic's enterprise share reaches 54%. The market shifted substantially from OpenAI dominance to a multi-model environment within 18 months.
Notes
- Gartner. (September 2025). "Gartner Says Worldwide AI Spending Will Total $1.5 Trillion in 2025." Gartner Newsroom. https://www.gartner.com/en/newsroom/press-releases/2025-09-17-gartner-says-worldwide-ai-spending-will-total-1-point-5-trillion-in-2025 — source for: $1.5 trillion worldwide AI spending forecast for 2025.
- Gartner. (January 2025). "Gartner Forecasts Worldwide GenAI Spending to Reach $644 Billion in 2025." Gartner Newsroom. https://www.gartner.com/en/newsroom/press-releases/2025-01-22-gartner-forecasts-worldwide-genai-spending-to-reach-644-billion-in-2025 — source for: $644 billion worldwide generative AI software and services spending forecast for 2025.
- Menlo Ventures. (2025). "2025: The State of Generative AI in the Enterprise." Menlo Ventures. https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/ — source for: $37 billion enterprise GenAI spending (up 3.2x from $11.5 billion in 2024); applications/infrastructure split; departmental AI breakdowns; foundation model API spending ($12.5 billion); model market share (Anthropic 40%, OpenAI 27%, Google 21%); 76% buy vs. build (up from 53%); 47% AI conversion to production vs. 25% SaaS; 16% true agent deployments in enterprises; 27% PLG share; shadow adoption estimate.