Though estimates of enterprise and other spending on artificial intelligence often are substantial, we sometimes do not stop to evaluate the magnitude of spending by customers, as opposed to suppliers.
That matters as customer spending represents the revenue upside from investing in generative AI infrastructure. Spending on AI infra by cloud computing giants to serve those customer needs arguably represents a “cost.” So does AI infra created by businesses.
To put it another way, if AI infrastructure investments (graphical processing units, tensor processing units, other acceleration chips) do not ultimately produce sufficient revenue to justify the investments, AI spending will fall.
Which is just another way of saying that, ultimately, customers pay for everything.
For example, analysts at IDC estimate $632 billion in 2028 AI spending. AI spending in the United States will reach $336 billion in 2028, IDC predicts. But AI is a large field, and generative AI investments in the U.S. market are forecast to be $108 billion in 2028, representing less than half of total AI investments.
The issue is what percentage of that investment will represent enterprise spending on AI software, hardware and services, compared to investments by suppliers to meet that need.
Consider that Amazon, Google, Meta, and Microsoft are expected to invest about $300 billion in capital expenditures in 2025 alone, reaching $336.5 billion for these four companies combined, in 2026. So something like half of all AI investments arguably will be made by a handful of “AI as a service” suppliers, rather than “buyers.”
And much of the other AI infra investment by businesses likewise is made in hopes of earning a financial return from customer purchases.
With the caveat that all spending anywhere in a value chain is cost to one participant and revenue to another participant, most AI spending today is arguably for creating AI capabilities that contribute in some way to product value, sales and revenue, or reduce cost in some way.
Relatively little direct AI end user customer revenue yet is obvious. Right now, perhaps 10 percent of “AI revenue” is generated by mostly-direct end user subscriptions or purchases. If we include indirect contributions (products that use AI), revenue could reach 20 percent.
Category of AI Spending | Purpose | Percentage of AI Spending | Examples |
AI Infrastructure | Investments in foundational AI tech, including computing power, cloud storage, and AI model training. | 20-30% | Cloud AI platforms (AWS, Google Cloud AI), GPUs, data storage solutions. |
AI Enterprise Software Apps and Features | Spending on AI-driven software for enterprise use cases in operations, marketing, finance, etc. | 25-35% | CRM systems with AI (Salesforce Einstein), ERP with AI, HR and analytics tools. |
AI Consumer Apps and Features | Development of AI-driven consumer applications and features enhancing user experience in consumer apps. | 15-25% | AI in search engines, recommendation algorithms in streaming, AI photo editing. |
End-User Product Revenues | Direct revenue from products where AI is a core feature or differentiator, including subscription models. | 10-20% | Autonomous vehicles, AI-driven personal assistants, generative AI tools. |
R&D and Experimental Projects | Investment in long-term or experimental AI research to fuel innovation and future capabilities. | 5-10% | Research in AGI, quantum computing for AI, human-AI interaction studies. |
Still, it might be fair to note that, at present, most AI investment is a cost incurred to generate revenue or improve efficiency. AI model subscriptions are among the best examples of consumer end user revenues being generated, at the moment.
AI’s value for business-to-business operations (commerce, advertising) might become the bigger revenue source soon.
