loading
loading
AI Spending Is The Honey Badger Of Budgets Trying to figure out how much money is sloshing around for AI infrastructure and services is becoming a sport. We are living through the most dramatic IT investment cycle in the history of information technology, rivaling the debut of commercial computing systems six decades ago that made IBM and several other companies fabulously wealthy (relatively speaking) at that time. AI represents a new architecture – not so new in terms of hardware if you are a supercomputing geek, of course – and is a thing that is distinct in its own right and will be so until it becomes the dominant architecture in the datacenter. And that new architecture is growing like crazy, and the estimates for AI spending keep get revised upwards despite inflation, worries about the AI bubble bursting, power constraints, supply chain woes, political instabilities, and uncertain national and global economies. AI is the honey badger of IT budgets, and nothing is stopping forecasts revised more up as they move to the right. And, as you may surmise, as the AI budgets are accelerating, the traditional datacenter IT spending is in recession, as it has been since 2023 when the GenAI boom took off in earnest. And it looks like traditional IT will be in recession for the foreseeable future. This prognostication is reinforced by the latest AI spending forecasts coming out of Gartner, and specifically from John-David Lovelock, who is a distinguished vice president analyst and economist the company. As usual, I am doing to add some perspective to this forecast, and also talk about the revenue streams that support the debt that the hyperscalers, cloud builders, and AI model builders are deploying to support the $8.1 trillion investment in AI stuff from 2025 through 2027. The contrast between AI spending and non-AI spending in the IT market is not obvious, which is why we have paired the AI forecast from Gartner with its overall IT spending forecasts. Here’s a chart that makes the traditional IT recession obvious: If you do the math, as I did, total IT spending rose by 10.7 percent to $5.58 trillion in 2025 and is expected to rise by an even larger 14.2 percent to $6.37 trillion in 2027. Gartner has not yet provided a publicly available IT spending forecast for 2027, so I plugged in my estimate of $6.98 trillion, up 9.6 percent. Over that time, AI spending in the aggregate rose by 2.6X in 2025 to $1.79 trillion and is anticipated to rise by 49.7 percent in 2026 to $2.67 trillion and by 36 percent to $3.64 trillion in 2027. Next year will be, if my estimate for overall IT spending is in the right ballpark, the first year where AI spending will be larger than traditional IT spending, which shrunk by 12.5 percent in 2025 to $3.79 trillion, and is expected to shrink further by 2.5 percent in 2026 to $3.69 trillion, and will shrink by 9.5 percent in 2027 to $3.43 trillion. In this interpretation of the data, I am ignoring the fact that people talk about AI smartphone or AI PCs and try to put them in the bucket. AI capabilities in these machines are largely incidental. People are buying a new phone or a new PC and it not surprisingly has features to run AI locally on these devices. That does not, in my mind, make them part of the AI infrastructure. Here is a table that brings all the latest Gartner numbers and forecasts together in one place: What is new with the latest forecast is that Gartner has broken licenses for GenAI models and sales of AI agents and assistants separate from sales of AI data science and machine learning platforms and more generic AI development platforms. This is distinct from the broader AI software category, which means enterprise software that has AI functionality built into it. Just like the AI PC or the AI smartphone, I am not sure if people are buying this functionality specifically, but are getting it as part of their upgrades. Many customers may be intentionally upgrading their apps to get that AI functionality, but I think it is more likely that they are upgrading capabilities even where there is not AI functionality – as is constantly happening in enterprise computing. AI is just another set of functions in the application stack, and at some point, this really is just what “software” will do. What everyone is most keen on understanding, for the sake of the GenAI bubble expanding even more, is how AI infrastructure spending – servers, storage, and switching – is holding up. In its review of 2025, Gartner actually raised AI infrastructure spending for 2025 a bit, from $975.6 billion in its May forecast to $981.9 billion in the September forecast. For 2026, Gartner added an additional $52.9 billion, and for 2027, it added $87.4 billion. I strongly suspect this increased spending is largely due to XPUs designs created by hyperscalers and cloud builders that are being used internally and also sold to the big AI model builders like OpenAI and Anthropic. Some of it is incremental Nvidia and AMD GPU-accelerated system sales, of course. “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.” So just how much money will these companies generate to fuel all of this investment? Synergy Research Group provides this answer in its latest forecast of spending for cloud computing services, cloud software services, and leading consumer-oriented digital services – the companies adopting GenAI software and making infrastructure purchases to run that GenAI in production. The biggies here are Google, Microsoft, Amazon Web Services, Meta Platforms, and ByteDance. From 2026 onwards, Synergy Research’s principle analyst, John Dinsdale, is expecting for revenue growth among these service providers to grow faster than they did from 2020 through 2025. If you add up the revenues from all of these service providers for 2025, 2026, and 2027 together (and the data is a little wiggly here because we are eyeballing it from a chart, not reading a table of actual numbers), there is $6.2 trillion in revenues for IaaS, PaaS, SaaS, search, and social media searches. Now, that revenue has to cover the costs of providing those services, which includes paying human beings salaries, health insurance, and other benefits. If you assume that 75 percent of the spending on AI stuff is done by the hyperscalers, cloud builders, and AI model builders, that works out to $6.1 trillion. The would have to spend all of their money – every penny of revenue stream – to fulfill their AI infrastructure dreams. This is why the tech titans are turning to debt financing, vendor financing from Nvidia, AMD, and Broadcom, roundtripping deals, and other ways of getting capital to allocate to their IT spending goals. If they don’t do this, they cannot afford to invest in the future of AI at the rate they want to, which is expressed by the Gartner AI spending forecast. Neoclouds, sovereign clouds, and commercial enterprises will face the same issues to a lesser or equal degree. In all cases, AI will have to start helping to remove costs to support further AI investments. There is no other way to do this except to keep borrowing from the future. And as that chart from Synergy Research shows for the years 2028 through 2031, the revenue streams are going to more than double. So, if this all pans out as expected, there is a big future to borrow against and that borrowing is what can make that future actually happen. We shall see. As I have said many times, the only way to predict the future is to actually live it.
document fetched 2026-10-06 02:55 · canonical https://nextplatform.com/ai/2026/10/05/ai-spending-is-the-honey-badger-of-budgets/5301184