Incident sheet · /infrastructure
- System
- Infrastructure & Cloud
- Filed
- Read
- 5 min · 1,051 words
- Severity
- Routine
66% of Enterprises Move or Plan to Move Workloads Off Public Cloud
Foundry's 2026 Cloud Computing Study: 66% of enterprises have moved or plan to move workloads off public cloud, with security (60%) leading and AI workloads increasingly homeward bound.

Event log
- 66% of IT decision-makers say their organizations have moved or plan to move workloads out of public cloud, per Foundry's 2026 Cloud Computing Study
- Organizations have repatriated 22% of workloads on average and plan to move another 24% within 12 months
- Security tops repatriation reasons at 60%, ahead of cost and performance (42% each) and sovereignty (40%)
- 34% of organizations have already moved AI/ML workloads off public cloud; another 38% plan to
- Only 13% of IT decision-makers plan to run AI primarily in public cloud
Two-thirds of enterprises — 66% — have already moved workloads out of public cloud or plan to, even as 74% of IT decision-makers say their organizations accelerated cloud migration over the past 12 months. Those two figures come from the same dataset, Foundry's 2026 Cloud Computing Study, and together they describe an enterprise estate in which workloads are being sorted, not abandoned.
"Repatriation is one part of a wider rethink of where workloads belong," says Mukesh Ranjan, vice president at research and advisory firm Everest Group. That rethink now runs workload by workload, with AI adding new variables — GPU costs, where models run, where prompts get processed.
How big are the moves, really?
The numbers do not show a mass exodus. On average, organizations have moved 22% of their workloads back from public cloud and plan to move another 24% within 12 months. One-third of organizations have repatriated at least a quarter of their workloads, down slightly from 37% in 2024.
Many of those organizations are pulling back only two or three workloads, Ranjan notes — hardly a rejection of public cloud.
For clients of global consulting firm Protiviti, "cloud first" remains the default. "Repatriation is still the exception, driven by a thorough analysis of business value and technical capability," says Randy Armknecht, managing director at Protiviti. The takeaway: this is selective correction, not reversal.
What is driving the pull-back?
Security leads the stated reasons at 60%, followed by cost and performance at 42% each, and sovereignty at 40%.
"Outside manufacturing, AI is a passenger on the repatriation trend driven by cost and security, not driving it directly," Armknecht says. Ranjan agrees AI drives some repatriation, but not most of it.
Even so, AI workloads rank among the most frequently homeward bound — nearly as often as storage and databases. Foundry found 34% of organizations have already moved AI or ML workloads out of public cloud, and another 38% plan to. Just 13% of IT decision-makers plan to run AI primarily in public cloud.
Much of what moves is not pure AI. "Roughly half of what I've seen has 'an AI component,' and very few are pure 'AI systems,'" Armknecht says. Manufacturing is the exception: several of Protiviti's industrial clients tested models in the cloud, then moved them onto local hardware at the edge to improve efficiency and cut error rates.
Sovereignty pressures have sharpened, especially in Europe. "A year ago, sovereignty was still often discussed as data residency or a compliance checkbox. Today, particularly in Germany and the EU, customers are asking much broader questions: who operates the service, where the dependencies sit, which law applies, and whether critical processes can continue if a major service is unavailable," says Uwe Geuss, CTO of German enterprise communications provider Retarus.
AI compounds the question, he adds: "not just where data is stored, but where models run, who operates the inference environment, where prompts and documents are processed, and whether sensitive data is being sent into an external AI platform."
At visual effects studio Framestore, data control comes first, followed closely by cost, says CTO Lincoln Wallen. The studio builds its own models and fine-tunes open-weight ones, runs them on its own infrastructure or via infrastructure-as-a-service, and taps major cloud providers for extra capacity — so it can adopt new models without rebuilding its systems each time.
When does owning beat renting?
Utilization decides it. High-volume inference favors private infrastructure when usage runs steady; intermittent demand favors public cloud, says Ranjan.
"Predictable usage is what makes local hardware pay off, and the need for predictability in spend is another driver for local versus cloud-based AI," Armknecht says. Mistakes also cost more in-house: "In the cloud, a sizing mistake costs you a month; on-prem, you live with it for the depreciation cycle."
Cloud savings that never materialized add fuel — idle resources, hard-to-calculate exit fees, and lift-and-shift applications never adapted to cloud-native capabilities all erode expected returns, Geuss says. Repatriation, in Ranjan's framing, is a "selective correction of earlier cloud decisions."
The pull is strongest where sensitive data, predictable demand, and scale converge: banking, healthcare, government, and some industrial workloads. Large enterprises absorb dedicated infrastructure costs more easily; smaller firms can reach similar outcomes through managed private cloud.
Which AI models can actually come home?
Many frontier models are available only as cloud services. Open-weight models closing in on frontier quality are what make AI repatriation practical, Armknecht says — and Framestore already fine-tunes them on its own gear.
Software testing firm Tricentis took the opposite path. It runs generative AI workloads on major cloud platforms and keeps only small classification models, such as vision models, on its own hardware.
"Maintaining this infrastructure in production only makes sense for small LLMs that you can assign specific tasks," says David Colwell, Tricentis's vice president of AI and machine learning. GPUs are scarce and expensive, and at the current pace of improvement, the next general-purpose model from an AI provider would quickly outperform anything Tricentis customized itself. The company keeps its datasets current so it can revisit the choice if base-model progress levels off.
Trust also shapes provider choice. Tricentis buys AI through large cloud providers rather than the AI labs directly, because the cloud providers have long track records protecting its data. "The frontier AI labs are still building that reputation," Colwell says.
What comes next?
Expect no stampede. Armknecht anticipates repatriation continuing "at a measured and business-value-driven pace, not a rush or hot trend." Ranjan expects placement decisions to grow more deliberate across public, private, sovereign, and edge environments, with some AI workloads moving back and others starting outside public cloud altogether.
"The challenge for CIOs will be maintaining consistent governance and economics across that estate," he says.
Framestore's Wallen advises CIOs to plan for AI in nearly every business process — and to weigh what it will cost to keep paying outside providers for it: "Many of today's AI services are excellent at demonstrating what's possible, but the cost of relying on them for every task could become difficult to sustain as adoption grows."
The estate is fragmenting by design; the governance bill is what comes due next.
via cio.com/feed/ (Source)
More from Amara Osei
Show full bio
Correspondent covering regulation, policy and their impact on enterprise information technology and the leaders who run at CIO Monitor.
188 articles
The newsroom digest
One email a day with the stories that matter. No noise.