top of page

The Infrastructure Behind AI: Where $2.7 Trillion Is Going

Writer: Merlin @GovernanceCentral
Merlin @GovernanceCentral
Sep 24
7 min read

Gartner forecasts worldwide AI spending will reach $2.67 trillion in 2026. More than half is expected to go toward infrastructure, illustrating how much of the current AI investment cycle is occurring behind the applications most people see.


For most people, artificial intelligence is experienced as software. We use an AI assistant, generate an image, summarize a document, write code or ask a model a question. The interaction takes place on a screen, making AI appear to be primarily another generation of software.


The spending behind AI tells a broader story. Gartner forecasts worldwide AI spending will reach approximately $2.67 trillion in 2026, a 49.5% increase from 2025. Of that amount, approximately $1.48 trillion is expected to go toward AI infrastructure, making infrastructure roughly 56% of total AI spending. That means approximately 56 cents of every dollar Gartner classifies as AI spending this year is associated with infrastructure. [thejournal.com], [techstrong.ai]


The distinction matters because the $2.67 trillion figure should not be interpreted as companies collectively spending $2.67 trillion on their own AI initiatives. A substantial portion is being spent by technology companies and infrastructure providers building the computing capacity on which AI applications depend. [thejournal.com], [techstrong.ai]


What Is Included in AI Infrastructure?


Gartner’s infrastructure category includes AI-optimized infrastructure-as-a-service, servers designed for AI workloads, AI networking, processing semiconductors and devices. Hyperscalers and technology service providers are expanding this capacity to support existing and anticipated future workloads. [cloudnews.tech], [techstrong.ai]


But the infrastructure required for AI extends beyond computing equipment. Modern data centers require buildings, electrical systems, cooling, networking and connections to substantial sources of electricity. In practical terms, the infrastructure supporting AI extends across a chain that looks something like this:


Semiconductors → servers → networking → data centers → cooling → electricity → grid capacity


This helps explain why the development of AI is affecting industries well beyond software and semiconductors. Data-center developers, utilities, electrical-equipment manufacturers, construction companies, cooling-system providers and network operators can all become part of the infrastructure required to provide additional AI computing capacity.


The Infrastructure Forecast Is Growing Quickly


Another notable feature of Gartner’s forecast is how much it has changed during 2026. At the beginning of the year, Gartner expected total 2026 AI spending of approximately $2.53 trillion, including about $1.37 trillion of infrastructure spending. Its latest forecast puts those figures at approximately $2.67 trillion and $1.48 trillion, respectively. [thejournal.com], [campustechnology.com]


That represents an increase of approximately $143 billion in Gartner’s overall 2026 forecast. Approximately $118 billion of the increase came from the higher infrastructure estimate, meaning infrastructure accounts for roughly 83% of the upward revision. The 83% figure is calculated from Gartner’s published forecasts rather than being a Gartner-reported percentage. [thejournal.com], [campustechnology.com]


Gartner expects infrastructure spending to continue growing, from approximately $982 billion in 2025 to $1.48 trillion in 2026 and nearly $1.98 trillion in 2027. The revisions provide useful context for the overall spending forecast: higher infrastructure expectations have accounted for most of the increase in Gartner’s 2026 estimate. [cloudnews.tech], [techstrong.ai]


Technology Providers Are Major Buyers


Who is spending the money is just as important as what is being purchased. Technology providers account for approximately 35% of worldwide AI spending, according to reporting based on Gartner’s research, as they purchase AI-optimized servers, semiconductors, networking equipment and cloud infrastructure. [techstrong.ai]


That spending is economically different from a manufacturer, retailer or bank purchasing an AI application. A technology provider building computing capacity is investing in the infrastructure used to supply AI services. A business subscribing to an AI application is purchasing the capability to use those services. Both can appear within a broad measure of AI spending, but they represent different activities and investment decisions.


This is why the $2.67 trillion figure needs context. It is a measure of activity across a broad AI ecosystem, not simply a measure of corporate spending on internal AI projects.


AI Is Increasing Electricity Demand


Electricity consumption provides one of the clearest indications that AI has also become an infrastructure issue. The International Energy Agency estimates that global data centers consumed approximately 415 terawatt-hours of electricity in 2024, or about 1.5% of worldwide electricity consumption. The IEA projects data-center electricity consumption to more than double by 2030, with AI identified as an important driver of that increase. [iea.org], [techxplore.com]


The IEA’s base case projects electricity generation required to supply data centers increasing from approximately 460 TWh in 2024 to more than 1,000 TWh in 2030. Servers account for around 60% of electricity consumption in modern data centers on average, while cooling accounts for roughly 7% in efficient hyperscale facilities and can exceed 30% in less-efficient enterprise data centers. [iea.org], [iea.org]


The consequences extend into the broader energy system. The IEA expects renewables to provide nearly half of the additional electricity required by data centers through 2030, while natural gas and coal together are expected to provide more than 40% of the additional demand over that period. [iea.org]


It is important not to equate all data-center electricity consumption with AI. Data centers support many other digital services. The IEA, however, identifies AI as the most important driver of projected data-center electricity-demand growth through 2030. [iea.org], [techxplore.com]


Is There Demand for the Capacity?


High infrastructure spending does not by itself tell us whether the capacity being built will eventually be fully utilized or what financial returns it will generate. There is, however, evidence of substantial current demand.


Alphabet reported in July that Google Cloud revenue grew 82% year over year to $24.8 billion in the second quarter of 2026, driven in part by demand for enterprise AI infrastructure and AI solutions. Cloud backlog increased to approximately $514 billion. Those figures are broader than AI alone because Alphabet also identifies core Google Cloud Platform services as contributing to the business, but they provide evidence of substantial demand for cloud and AI infrastructure. [s206.q4cdn.com], [abc.xyz]


Gartner likewise describes infrastructure investment as supporting anticipated future workloads. Infrastructure providers therefore have to consider both existing customer requirements and expectations of future demand when determining how much capacity to construct. [cloudnews.tech], [techstrong.ai]


That is characteristic of infrastructure investment generally. Data centers, power connections, cooling systems, servers and networking capacity cannot be constructed instantly when demand appears. Capacity has to be planned and built in advance, which means infrastructure investment necessarily incorporates expectations about future usage.


The Less Visible Part of AI


One useful way to understand the AI market is as a series of interconnected layers. At the top are the models, assistants and applications that people see. Underneath them are the semiconductors, servers, networking and cloud capacity required to provide the computing. Beneath that computing layer are data centers, cooling systems, electricity supplies and grid connections.


The farther down this stack one goes, the less visible the infrastructure becomes to the person actually using AI. Gartner’s numbers demonstrate how economically significant those less-visible layers have become, with infrastructure representing approximately 56% of its worldwide AI spending forecast for 2026. [thejournal.com], [techstrong.ai]


The spending forecast does not establish whether every infrastructure investment will generate an adequate financial return, nor does it tell us whether the industry will ultimately construct too much capacity, too little or approximately the right amount. It is a forecast of spending, not a forecast of investment returns. What it does illustrate is the scale of the computing and physical infrastructure being constructed to support AI.


Why Does This Matter to Boards?


For some boards, the connection is direct. Boards overseeing cloud providers, semiconductor companies, data-center operators, utilities, electrical-equipment manufacturers and other businesses participating in the infrastructure expansion face decisions involving capital allocation, capacity, demand, financing and investment returns.


For boards outside those industries, the relevance is more likely to be indirect. AI infrastructure may become a board-level issue if it materially affects the availability or price of computing services, major technology suppliers, electricity costs or availability, capital investment requirements, supply chains, or the economics of the company’s industry.


That does not mean every board needs an “AI infrastructure strategy.” Materiality is the better test. Does the infrastructure developing underneath AI materially affect the company’s strategy, economics, suppliers, capital requirements or risks? If the answer is yes, directors should understand it. If the answer is no, the infrastructure buildout may remain an important economic and technological development without becoming a significant board-level issue.


This distinction is also why Gartner’s $2.67 trillion forecast should not be used to suggest that every company needs to increase its own AI spending. The forecast covers a much broader ecosystem, and more than half of the amount is associated with infrastructure. [thejournal.com], [techstrong.ai]


Understanding What the $2.7 Trillion Represents


The headline that worldwide AI spending is approaching $2.7 trillion is significant, but the composition of the spending tells the more informative story. Approximately $1.48 trillion is expected to go toward AI infrastructure in 2026, and Gartner forecasts that figure approaching $2 trillion in 2027. Technology providers are major participants in that spending, while the infrastructure itself increasingly reaches beyond computing equipment into data centers and the electricity system. [cloudnews.tech], [techstrong.ai]


At the same time, major cloud providers are reporting substantial customer demand, and the IEA expects data-center electricity requirements to increase significantly through the end of the decade. These developments describe an AI economy in which the visible applications depend on a large and rapidly expanding infrastructure base. [iea.org], [s206.q4cdn.com]


For most people, AI will continue to be experienced primarily as software. But much of the investment taking place today is occurring somewhere users rarely see it: in the computing, physical and energy infrastructure underneath AI.


For boards, the question is not whether every company should participate in that spending. It is whether the infrastructure transformation occurring underneath AI is material to the business they oversee.


Sources and Further Reading

Comments

bottom of page