AI Spending Is the Honey Badger of IT Budgets

Estimating how much money is flowing into AI infrastructure and services is becoming increasingly difficult. The current investment cycle may be the most dramatic in information technology's history, comparable to the arrival of commercial computing systems six decades ago, which made IBM and several other companies extremely wealthy at the time.

AI represents a distinct architectural approach. Its hardware foundations are familiar to supercomputing specialists, but the architecture will remain distinctive until it becomes dominant in the data center. That transition is advancing rapidly, and AI spending estimates continue to move upward despite inflation, concerns about an AI bubble, power constraints, supply-chain problems, political instability, and uncertain national and global economies. AI is proving unusually resistant to pressure on IT budgets, with forecasts repeatedly revised upward.

As AI budgets accelerate, traditional data center IT spending is contracting. That decline began in 2023, when the generative AI boom gathered momentum, and current projections indicate that conventional IT spending could remain in recession for the foreseeable future.

This trend is reflected in the latest AI spending forecast from Gartner, including analysis from John-David Lovelock, a distinguished vice president analyst and economist at the company. The forecast also provides a basis for considering the revenue streams that may support the debt being used by hyperscalers, cloud builders, and AI model developers to fund an estimated $8.1 trillion investment in AI-related infrastructure and services from 2025 through 2027.

The contrast between AI and non-AI spending is not immediately obvious, so the AI forecast can be considered alongside Gartner's broader IT spending projections. The combined figures make the traditional IT recession more apparent.

If the figures are calculated together, total IT spending rose 10.7 percent to $5.58 trillion in 2025. It is expected to increase by another 14.2 percent to $6.37 trillion in 2026. Gartner has not yet published a publicly available forecast for total IT spending in 2027, so an estimate of $6.98 trillion, representing 9.6 percent growth, is used here.

Over the same period, aggregate AI spending increased by 2.6 times in 2025, reaching $1.79 trillion. It is expected to grow 49.7 percent to $2.67 trillion in 2026, followed by 36 percent growth to $3.64 trillion in 2027.

If the estimate for total IT spending is close to the actual figure, 2027 will be the first year in which AI spending exceeds traditional IT spending. Traditional IT spending fell 12.5 percent in 2025 to $3.79 trillion, is expected to decline another 2.5 percent in 2026 to $3.69 trillion, and could fall 9.5 percent in 2027 to $3.43 trillion.

This interpretation excludes AI smartphones and AI PCs from the AI infrastructure category. Although these devices increasingly include local AI capabilities, consumers generally purchase them as new phones or PCs, with AI features included as part of the broader product. Their AI functions are therefore treated as incidental rather than as evidence of spending specifically directed toward AI infrastructure.

Changes in Gartner's AI categories

Gartner's latest forecast separates licenses for generative AI models and sales of AI agents and assistants from AI data science and machine learning platforms, as well as from more general AI development platforms. These categories are distinct from the broader AI software category, which includes enterprise software with AI functionality built into it.

As with AI PCs and AI smartphones, it is not always clear whether customers are purchasing enterprise software specifically for its AI features or receiving those features as part of routine upgrades. Some organizations may be upgrading applications specifically to obtain AI capabilities, while others are upgrading their software more generally. AI is becoming another set of functions in the application stack, and over time it may simply become a standard part of what software does.

AI infrastructure spending

The area receiving the most attention is AI infrastructure spending on servers, storage, and networking equipment. In its review of 2025, Gartner slightly increased its estimate for AI infrastructure spending, from $975.6 billion in its May forecast to $981.9 billion in its September forecast. Gartner also added $52.9 billion to its 2026 estimate and $87.4 billion to its 2027 estimate.

A significant portion of this increase may come from custom XPU designs created by hyperscalers and cloud builders. These processors are used internally and sold to major AI model developers such as OpenAI and Anthropic. Additional spending is also expected from Nvidia and AMD systems using GPU acceleration.

“The buildout of AI data center capacity is the largest infrastructure project humanity has even undertaken,” Lovelock said in the statement accompanying the updated AI spending forecast. “The capacity growth from hyperscalers and service providers purchasing AI-optimized servers will continue to be the largest single area of spending.”

Revenue supporting the investment

Synergy Research Group's latest forecast offers a view of the revenue that could support this investment. It covers cloud computing services, cloud software services, and major consumer-oriented digital services. The companies in these categories are adopting generative AI software and purchasing infrastructure to operate those systems in production.

The largest participants include Google, Microsoft, Amazon Web Services, Meta Platforms, and ByteDance.

From 2026 onward, Synergy Research principal analyst John Dinsdale expects revenue growth among these service providers to exceed the growth recorded from 2020 through 2025. Adding the projected revenue from 2025, 2026, and 2027 produces approximately $6.2 trillion across infrastructure as a service, platform as a service, software as a service, search, and social media services. The figure is approximate because it is derived from a chart rather than a table of exact values.

That revenue must also cover the cost of operating these services, including employee salaries, health insurance, and other benefits. If hyperscalers, cloud builders, and AI model developers account for 75 percent of AI-related spending, their share would amount to approximately $6.1 trillion. Under that assumption, they would need to commit nearly all of their revenue during the period to fulfill their AI infrastructure plans.

This helps explain why major technology companies are using debt financing, vendor financing from Nvidia, AMD, and Broadcom, round-tripping arrangements, and other sources of capital to fund their IT spending objectives. Without these mechanisms, they would not be able to invest in AI at the rate implied by Gartner's forecast.

Neocloud providers, sovereign clouds, and commercial enterprises face similar issues, although the scale may be smaller. In every case, AI will eventually need to help reduce operating costs or generate sufficient returns to support additional AI investment.

The alternative is continued borrowing against future revenue. Synergy Research's projections for 2028 through 2031 show those revenue streams more than doubling. If the projections are realized, that future revenue provides a large base against which companies can borrow, and the borrowing can help create the infrastructure needed to produce that future.

The outcome will become clearer as these investments develop.