IBM's latest earnings gave a clean snapshot of what is happening inside company technology budgets everywhere. Sales of its older mainframe computers fell sharply, while the part of its business that sells servers, storage and memory grew fast. Customers are not necessarily spending less. They are spending differently, moving money toward the machines that power AI and away from equipment that does not.
This is not just an IBM trend. Gartner now expects worldwide technology spending to reach 6.37 trillion dollars in 2026, up more than 14 percent from the year before. Spending on data center equipment, the physical machines that run AI systems, is expected to jump more than 60 percent on its own, making it the fastest growing part of the entire technology market.
That sounds like good news for anyone selling technology, but it hides a harder story for the people buying it. A Gartner analyst put it plainly: this is not a rising tide that lifts every part of a company's technology spending. Inflation, hardware costs and AI funding pressure are all pulling on the same limited budget at once.
Here is the part that matters most for any business leader watching this from the outside. A recent survey of over 140 chief information officers found that nearly half of all new AI spending is not new money. It is money pulled out of existing software and technology budgets and relabeled as an AI project, because the AI label makes it easier to get approved. One fractional technology executive we found described it as old money with a more fashionable job description.
That kind of budget shuffling might work for a quarter or two, but it creates a real risk. The unglamorous work behind AI, things like clean data, proper governance and basic security, is exactly what gets cut when budgets are tight, because it does not carry the AI label that gets funding approved. And that work is not optional. It is what determines whether an AI project actually works.
The evidence is already showing up. Separate research tracking AI projects found failure rates jumped from 17 percent to 42 percent in just one year, largely because companies skipped the groundwork of getting their data and systems ready before rushing to deploy AI tools.
The lesson for any business leader, whether running a manufacturing plant, an insurance consultancy or a retail chain, is simple. Do not judge a technology investment by whether it has the AI label attached. Judge it by whether it builds something the business will actually depend on. The companies that keep funding their data quality, security and basic IT plumbing without waiting for a label will be the ones whose AI projects actually work, while their competitors chase the label and wonder why the results never show up.