86% of CIOs plan to repatriate cloud workloads as costs exceed expectations, according to a Barclays survey. Hardware spending for AI rises, echoing earlier cycles.
Companies are buying more servers and storage for artificial intelligence, even as cloud spending disappoints. The move mirrors earlier patterns in mobile device management and cloud repatriation – and carries echoes of the optical-infrastructure boom that ended in a crash.
A Barclays CIO survey found 86% of CIOs planning to move some workloads from public cloud back to private or on-premises environments, the highest reading in the survey's history. An IDC survey reported that 69% of IT decision-makers said actual cloud costs exceeded original expectations. The response is not to leave the cloud entirely – public cloud spending is still climbing – but to re-own hardware when the rented promise falls short.
IBM's latest quarter offers a single data point inside that shift. Customers reprioritized capital toward servers and storage, straining IBM's revenue mix and weighing on its earnings. The stock carries an Alpha Score of 37 out of 100, flagged as Mixed. The pattern is consistent: when uncertainty about where AI value will land grows, the money moves toward the thing a company can put on a purchase order. IBM stock page
The same reflex appeared in the mobile era. Tanya Seda, an industry analyst who covered enterprise mobility, compared it to the years when unmanaged bring-your-own-device growth forced organizations to bolt on Mobile Device Management. "The lesson was never 'build better MDM,'" she said. "It was 'design systems so management isn't constantly playing catch-up.'" MDM managed the consequences of sprawl; it did not remove the conditions that produced them.
Cloud repatriation follows the same logic. It is a response to the disappointment of expected value, not a cure for the underlying operating-model misalignment. An IDC forecast projected a roughly 10% rise in hardware infrastructure sales driven by AI. Enterprises are increasingly choosing to build AI systems on their own hardware rather than rent from hyperscalers.
The scale of the buildout is massive. Gartner projects the average large enterprise will run more than 150,000 AI agents by 2028, while only 13% of organizations believe they have the governance in place to manage them. The four largest hyperscalers are guiding to roughly $725 billion in combined AI infrastructure spending in 2026, up 77% in a single year – a figure that, by many analysts' own admission, outpaces demonstrable AI revenue.
A historical precedent exists. In early 2000, Nortel Networks saw its market value fall more than 99% from a peak that had briefly made it one of the most valuable companies in the world. JDS Uniphase wrote down some $45 billion in value paid for acquisitions made ahead of demand that never fully arrived. What broke those companies was not a shortage of internet – it was infrastructure built far ahead of realized value.
Buying hardware during a boom is ambiguous. It can signal confidence that the technology is working so well that owning the infrastructure beats renting it. Or it can be displacement – reaching for the tangible because the expected value has not shown up, and hardware is easier to fund than the harder question of why. From the outside, the two look identical. Only the outcome tells which one was in play.
Barclays, whose CIO survey captured the shift, holds an Alpha Score of 59 out of 100, labeled Moderate. The bank's own analysis of enterprise spending patterns may inform how investors read the trend. BCS stock page
The deciding variable is not technology. It is whether organizations can absorb and operationalize what they are buying at the pace their investment assumes. That variable is the one most easily skipped, because operating-model coherence does not appear on a balance sheet the way a data center does.
For investors, the next differentiator may not be backing the best model or the largest pipeline. It may be assessing whether a company's customers can operationalize AI at the scale the valuation assumes. Readiness as diligence. The dot-com money that survived was not the capital chasing the promise; it was the capital that backed businesses which could execute. In this wave, execution is absorption.
The question remains open: three reversions to the tangible, one unresolved reading, one precedent that remembers how this goes. What the data shows is that the reflex is real. Whether it ends in correction or rational optimization depends on absorption – and that is not yet settled.
Drafted by a large language model from the source reporting linked above, then screened by automated publishing checks. It is not read by a journalist before publication. Some articles cite our Alpha Score. Verify prices and figures against the original source. Educational coverage, not personalized advice.