Ninety-three percent of enterprises have repatriated AI workloads from public cloud, are in the process of doing so, or are actively evaluating it. 1 Source 1 Cloudian, “Enterprise AI Infrastructure Survey 2026,” March 2026. Not a typo. Not a fringe survey of cloud skeptics. A cross-industry poll of 203 IT decision-makers, two-thirds of whom already have AI in production.
The cloud-first era isn’t ending. It’s being renegotiated. And the terms are shifting decisively toward ownership.
Why are enterprises moving data out of the cloud?
The promise of public cloud was simple: let someone else handle the infrastructure. For a decade, that trade worked. Companies poured workloads into AWS, Azure, and Google Cloud. The migration felt irreversible.
Then the bills arrived.
Forty percent of enterprises report that actual cloud AI spending exceeds initial projections (Cloudian, 2026). The pricing model rewards unpredictability. Storage tiers, API call charges, cross-region data transfer, and consumption-based AI credits add up in ways that are nearly impossible to forecast for data-intensive workloads.
These aren’t startups experimenting with hybrid. They’re large enterprises that ran the numbers, found the math no longer worked, and moved.
The examples are no longer anecdotal — large, sophisticated companies have run the numbers and walked:
- 37signals (the company behind Basecamp) left a $3.2 million annual AWS bill behind, projecting over $7 million in savings across five years. 2 Source 2 David Heinemeier Hansson, “We have left the cloud,” October 2024.
- GEICO spent a decade migrating to the cloud, watched costs climb 2.5x to over $300 million a year, and is now building private cloud infrastructure. 3 Source 3 The Stack, “Warren Buffett’s GEICO repatriates work from the cloud,” 2025.
- Broadcom moved critical database workloads off public cloud and saved over $10 million. 4 Source 4 VMware/Broadcom, “The CFO’s Case for On-Premises DBaaS Repatriation and Cost Control,” January 2026.
of enterprises have repatriated or are evaluating repatriation of AI workloads
Cloudian, 2026
of CIOs planned to move public cloud workloads back to private infrastructure
Barclays, 2024
of enterprises expect to repatriate some compute or storage workloads
IDC, 2024
What’s behind the repatriation wave?
Three forces are reshaping where enterprises keep their data.
Data sovereignty is no longer optional. When Gartner coins a term for what you’re doing, the trend is real. They call it “geopatriation” — moving data and applications out of global public clouds and into local options due to geopolitical risk. 5 Source 5 Gartner, Top Strategic Technology Trends for 2026, October 2025. Their prediction: by 2030, more than 75% of European and Middle Eastern enterprises will geopatriate their workloads.
Regulations are tightening. The EU AI Act’s high-risk system obligations became fully applicable in August 2026. 6 Source 6 European Commission, EU AI Act (Regulation 2024/1689). GDPR enforcement continues to intensify. The U.S. CLOUD Act gives American authorities the theoretical ability to compel access to data stored by American companies anywhere in the world, a scenario that European CTOs increasingly view as a non-starter for compliance.
“An American company needs to give access to the United States government on request,” said Sergio Samarelli, CTO of Italian satellite company Planetek. “This process is simply not compliant with European regulation.” 7 Source 7 Nutanix/The Forecast, “Clouds With Borders: IT Teams Design for Geopatriation,” February 2026.
Cloud costs are structurally unpredictable. The cloud pricing model was designed for variable, bursty workloads. AI workloads are the opposite. They’re data-intensive, GPU-hungry, and they run around the clock. IDC found that 59% of organizations spent more than budgeted on cloud in 2024. 8 Source 8 IDC, “Storm Clouds Ahead: Missed Expectations in Cloud Computing,” October 2024. Flexera reported that 84% of organizations cite managing cloud spend as their single biggest challenge. 9 Source 9 Flexera, “2025 State of the Cloud Report.”
For steady-state workloads, the economics have flipped. Broadcom’s internal analysis found that modern private cloud delivers 40–50% lower total cost of ownership compared to public cloud for predictable workloads. Deloitte’s analysis found that on-premise AI delivers 50% or more cost savings over three years compared to cloud API alternatives, once token volume crosses a certain threshold. 10 Source 10 Deloitte Tech Trends 2026.
AI demands control that cloud can’t deliver. Every AI deployment involving proprietary data forces a question: where does that data get processed? Cloud-based AI means your data leaves your environment. KPMG’s AI Pulse Survey found that data privacy as a barrier to AI adoption jumped from 53% to 77% between Q1 and Q4 of 2025. 11 Source 11 KPMG, “AI Pulse Survey,” Q4 2025. Cybersecurity concerns are even higher, at 80%.
When the data is sensitive, decision-makers overwhelmingly want it kept close. Ninety-one percent of Cloudian’s survey respondents would choose on-premises, private cloud, or hybrid infrastructure over public cloud when deploying AI involving sensitive company data. This isn’t cloud skepticism. It’s a rational response to the reality that AI amplifies every data governance risk by orders of magnitude.
How big is the repatriation wave?
Sovereign cloud infrastructure spending will hit $80 billion in 2026, growing 35.6% year-over-year. 12 Source 12 Gartner, “Worldwide Sovereign Cloud IaaS Spending Forecast,” February 2026. Europe alone is seeing an 83% increase in sovereign cloud spending. Gartner attributes the growth to rising geopolitical tensions and a desire for digital and technological independence.
The hyperscalers — the three giants that dominate public cloud, AWS, Microsoft Azure, and Google Cloud — see the trend and are scrambling to respond. Microsoft is deploying isolated Azure regions in European markets, operating separately from the global Azure backbone. In France, they partnered with Capgemini and Orange to launch the “Bleu” national cloud joint venture. In Germany, they’re partnering with SAP and Arvato Systems to create a sovereign cloud for the public sector. 13 Source 13 Brad Smith, Microsoft President, blog post on European digital commitments, April 2025. Google is expanding sovereign cloud partnerships with European service providers. AWS is working on sovereign cloud frameworks that limit operator access.
But as 3Cloud principal architect Joey D’Antoni noted: “Most of what Microsoft has done so far feels like compliance positioning. If they were to legitimately partner with someone and stand up a new data center, I would buy it a lot more.”
The hyperscalers are trying to have it both ways: global infrastructure with sovereign branding. The enterprises pulling data back aren’t buying it.
What does cloud repatriation mean for your SaaS stack?
Here’s where the conversation matters for the mid-market company spending $100,000 to $300,000 a year on SaaS.
Enterprise cloud repatriation is about infrastructure: compute, storage, networking. But the exact same forces apply to your marketing, sales, and operations software. Every SaaS vendor is a cloud you don’t control. Your CRM data lives in a vendor’s infrastructure. Your website lives in a vendor’s infrastructure. Your customer behavioral data, your automation logic, your audience segments — all hosted in vendor environments, governed by vendor terms, accessible through vendor APIs.
The math is exactly the same:
| Enterprise Cloud Problem | Mid-Market SaaS Problem |
|---|---|
| Data processed in jurisdictions you don’t control | Customer data stored in vendor clouds you can’t audit |
| Unpredictable consumption-based pricing | Per-seat pricing that scales with headcount, not value |
| AI workloads require data to leave your environment | AI features process your data through vendor systems |
| Vendor lock-in makes switching prohibitively expensive | The Five Locks make leaving your SaaS vendor an engineering project |
| ”Managed” doesn’t mean “owned" | "All-in-one” doesn’t mean “yours” |
The enterprise CIO who repatriates AI workloads from AWS is making the same calculation as the mid-market VP of Marketing who migrates off a legacy CRM: the cost of renting exceeds the cost of owning, and ownership delivers capabilities that renting can’t match.
We’ve spent 16 years inside the SaaS ecosystem building the integrations, extending the platforms, and watching companies hit the exact ceilings that the repatriation wave is now exposing at the infrastructure level. The vendor lock-in mechanisms we documented in The Five Locks: Code Lock, Data Lock, Logic Lock, Audience Lock, Dependency Lock are the SaaS equivalent of the infrastructure lock-in that enterprises are now paying billions to escape.
The lesson from the enterprise repatriation wave is clear. If the largest, most resource-rich companies in the world have concluded that ownership beats renting for their most critical workloads, the mid-market shouldn’t be paying rent on its CRM, CMS, and marketing infrastructure either.
What’s the alternative to SaaS for mid-market companies?
The Sovereign Stack is the mid-market answer to the same impulse driving enterprise cloud repatriation. Five layers of open-source, composable infrastructure — front-end, CRM, data warehouse, automation, and analytics — that a company owns outright.
The architecture addresses every repatriation driver:
Data sovereignty. Your customer data lives in a database you own, in a jurisdiction you choose, with access controls you define. Not scattered across vendor silos with incompatible export formats and competing claims on your data.
Cost predictability. No per-seat pricing. No consumption-based credits. No surprise invoices when your team grows or your AI usage spikes. The infrastructure costs what it costs, and the cost curve trends down with open-source maturity and commodity hosting.
AI-native by design. When your data isn’t locked behind a vendor’s API, AI agents can operate on it natively. Read any record, write any field, trigger any workflow, without paying API fees to access your own customer data and without your data leaving your environment.
Deployment flexibility. Run it on your own cloud, a private VPS, a managed hosting provider, or a combination. If your hosting provider doubles their price, you move. Your code runs on open standards, not a vendor’s proprietary runtime. You hold the leverage.
The Sovereign Stack isn’t a theoretical exercise. It’s the architecture that the largest companies in the world already run on. The difference in 2026 is that AI has collapsed the cost and complexity of building and maintaining these systems to the point where mid-market companies can deploy them without enterprise budgets.
How should mid-market companies evaluate their own repatriation?
The enterprises getting this right aren’t ripping everything out at once. They’re making workload-by-workload decisions. The same approach applies to your SaaS stack.
1. Audit your actual SaaS spend against value delivered. Start with the line-item breakdown. Tool, annual cost, contract renewal date, what it does, and whether the value justifies the price. Where are you overpaying for features you don’t use? Where are per-seat costs scaling faster than revenue?
2. Map your data residency exposure. Where does your customer data actually live? Which vendor clouds store it? What are the export options if you leave? If you can’t answer these questions for every major tool in your stack, that’s the first problem to solve.
3. Identify the weakest link. Not every tool needs replacing today. But there’s almost always one that costs too much and delivers too little. For most companies, that’s the CMS, the most visible layer with the clearest migration path and the most immediate performance improvement.
4. Assess your AI readiness. If you’re planning to deploy AI across marketing, sales, or operations, where will your data be processed? If the answer is “through our vendor’s AI features,” ask: what data are you sending them, what do they do with it, and what happens when the pricing changes?
5. Evaluate switching costs honestly. The Five Locks framework gives you a diagnostic for this. Rate your lock-in across Code, Data, Logic, Audience, and Dependency. A total score above 20 means migration requires a structured extraction plan — but every lock can be broken.
The repatriation wave isn’t coming. It’s here. Ninety-three percent of enterprises are already moving. The mid-market is next.