News

The $2.7 Trillion AI Boom Is Mostly an Infrastructure Boom

• By John K. Waters
• 09/17/2026

The artificial intelligence industry is expected to generate an extraordinary $2.67 trillion in worldwide spending this year. But relatively little of that money will be spent on the generative AI models dominating the industry's headlines.

The much bigger business is everything required to make them run.

Worldwide spending on AI infrastructure will reach nearly $1.5 trillion in 2026, according to a new forecast from Gartner, accounting for about 56% of all AI spending. Gartner expects spending on generative AI models themselves to total just $28.3 billion. Put another way, for every dollar Gartner expects to be spent on generative AI models this year, more than $52 will be spent on AI infrastructure.

It is a striking measure of how the economics of the AI boom are changing. ChatGPT, Claude, Gemini, and other models may be the most visible products of the generative AI era, but much of the money is going to the enormous buildout of servers, semiconductors, networks, cloud capacity, and other infrastructure required to support them.

"The buildout of AI data center capacity is the largest infrastructure project humanity has ever undertaken," John-David Lovelock, distinguished vice president analyst at Gartner, said in announcing the forecast.

$2.67 Trillion and Counting
Gartner now expects worldwide AI spending to increase 49.5% this year, from about $1.79 trillion in 2025 to $2.67 trillion in 2026. Infrastructure represents $1.484 trillion of that total. The next-largest categories are AI services, at $576.5 billion, and AI software, at $461.6 billion. Spending on AI agents and assistants is expected to reach $29.2 billion, only slightly more than the $28.3 billion Gartner forecasts for generative AI models.

The numbers also show just how quickly expectations are changing. In January, Gartner forecast $2.53 trillion in total AI spending for 2026, including $1.37 trillion for infrastructure. By May, those forecasts had risen to $2.60 trillion and $1.43 trillion, respectively. The latest forecast raises them again, to $2.67 trillion overall and $1.48 trillion for infrastructure.

Since January, Gartner has added roughly $143 billion to its 2026 AI spending estimate. About $118 billion of that increase, or roughly 83%, comes from a higher infrastructure forecast.

That is significant because infrastructure was already, by far, the largest component of Gartner's AI spending estimate. As the firm's expectations for the AI market have risen during 2026, most of the additional money has accumulated there.

The AI Factory Gets Bigger
The infrastructure category is broad. Gartner includes AI-optimized cloud infrastructure, servers, networking, AI processors, and devices. Demand for those systems remains strong despite rising memory prices, Lovelock said.

"The capacity growth from hyperscalers and service providers purchasing AI-optimized servers will continue to be the largest single area of spending," he said.

That helps explain why the infrastructure surrounding AI models has become such an important part of the industry's competitive landscape. Nvidia, AMD, Intel, cloud providers, networking companies, memory manufacturers, and data center operators are all competing for pieces of a market shaped by AI's increasingly formidable physical requirements. And those requirements are changing as AI moves from training models to running them.

Gartner forecast last month that spending on AI-optimized infrastructure as a service will reach $42.3 billion this year, up 96% from 2025.

More significantly, inference is overtaking training. Gartner expects $23.3 billion of AI-optimized IaaS spending to support inference this year, compared with $19 billion for training. Inference will account for 55% of spending in the category in 2026 and 59% in 2027.

That matters because inference is the computing required every time someone actually uses an AI model.

Training a frontier model is enormously expensive, but it is a bounded event. A widely deployed model, by contrast, must perform inference every time a user submits a prompt, an application calls the model, or an AI agent performs another step in a task.

Agentic AI potentially amplifies that demand because an agent may execute numerous model calls, use external tools, retrieve information, and repeatedly reason through a problem before completing a task.

"As organizations shift from model development to production-scale deployment, fine-tuned and domain-specific models are increasingly integrated into customer-facing and operational systems, requiring continuous, real-time execution rather than periodic training," Gartner senior principal research analyst Hardeep Singh said.

The Models Are Still Growing
None of this means the model business is shrinking. Quite the opposite. Gartner expects generative AI model spending to increase 117% this year, from $13 billion in 2025 to $28.3 billion in 2026. In a separate July forecast, Gartner projected particularly rapid growth for domain-specific and specialized generative AI models.

The distinction is scale. Even after more than doubling, generative AI model spending represents only about 1% of Gartner's $2.67 trillion estimate for the overall AI market. Infrastructure represents roughly 56%.

That gap helps explain why so much of the AI industry's attention is moving toward problems that have little to do with making models smarter: increasing memory capacity, moving data faster, reducing inference costs, supplying electricity, and building enough data centers to house the equipment.

It also puts the extraordinary capital spending by hyperscalers and AI companies into perspective. Gartner forecasts overall worldwide IT spending of $6.37 trillion in 2026. Data center systems alone are expected to reach $822 billion, up 62.5% from 2025.

Those figures provide useful context for the scale of the infrastructure buildout, although Gartner's AI and overall IT spending categories should not be treated as separate, directly comparable buckets. AI spending runs through many of the broader hardware, software, and services categories that make up worldwide IT spending.

Follow the Money
For the first few years of the generative AI boom, it was easy to think of the model as the product and the infrastructure as the machinery needed to produce it. Gartner's numbers suggest that distinction is becoming harder to maintain.

The models are still improving rapidly, and spending on them is growing. But deploying those models at global scale requires an expanding physical and cloud infrastructure beneath them. And increasingly, that infrastructure is being built not merely to train the next generation of models, but to handle the enormous amount of inference generated by models already entering everyday applications.

That changes the economics of the AI race.

The companies building the models may continue to capture the public's attention. But if Gartner's forecast is right, the biggest part of the AI economy this year will be the massive industrial system required to keep those models running.