For most of the past decade, the conventional wisdom in IT has been a one-way street: everything moves to the public cloud. On-premise infrastructure was treated as a legacy holdover, something you migrated away from, not toward. I’ve spent a fair number of these columns pushing back, gently, on that assumption where Nonstop is concerned. It turns out I’m no longer the lonely voice.
The data now backs this up! A Barclays CIO survey found 86% of CIOs plan to move at least some workloads back to on-premises or private infrastructure. IDC puts the number of enterprises already shifting workloads to on-prem or hybrid environments at 70%. A 2025 sovereignty survey found 97% of mid-market organizations are planning to move workloads off public cloud altogether. Wow. However you slice the methodology, the direction is unmistakable: the cloud-only era is over, and “repatriation” has become one of the most-used words in enterprise IT planning.
So what’s driving this?
Cost is the biggest issue, and it isn’t subtle. Flexera’s State of the Cloud report found 84% of organizations cite managing cloud spend as their top challenge, and 27% of cloud infrastructure spend goes to waste on underutilized resources. Andreessen Horowitz has estimated that public cloud costs can erode a software company’s gross margins by 50% or more — and that repatriating workloads can, in some cases, meaningfully change how a company’s own financials look to investors. The examples are becoming a genre unto themselves: Dropbox saved an estimated $75 million over two years moving storage in-house. Basecamp’s parent company, 37signals, is on pace for $7 million in savings over five years after leaving the public cloud for owned hardware. GEICO reported cutting compute cost per core in half. These are not small, ‘cloud-skeptical’ companies making a nostalgic choice. They are sophisticated companies doing the math and reaching decisions based on a spreadsheet, not a philosophy.
Regulation is the second driver, and it’s one I suspect will grow. Data sovereignty laws — GDPR, CPRA, a growing list of country-specific data localization rules, the EU AI Act — are pushing companies, especially in financial services, toward infrastructure they fully control and for which they can prove the location. Gartner has forecast that 60% of financial firms outside the U.S. will adopt sovereign infrastructure deployments by 2028. If you’ve read my earlier column on AI and Financial Services (issue 17), this will sound familiar: compliance has a way of becoming an accelerator for exactly the kind of infrastructure decisions that used to be optional.
The third driver is newer, and it’s AI itself. Surprise. Training and running AI models, especially at any real scale, is turning out to favor infrastructure that sits close to the data and is provisioned for a known, sustained workload — not infrastructure billed by the API call in an environment you don’t fully control. In the latest issue of Nonstop Insider Issue 10-4 (AI’s Power Problem — and Why Boring Infrastructure Still Wins) I wrote about the strain AI is putting on power grids and data centers more broadly. That same pressure is quietly pushing some of the same companies toward owning more of their own AI infrastructure rather than renting all of it.
None of this means public cloud is going away, and I’d be doing my own argument a disservice if I implied otherwise. Cloud remains an excellent answer for bursty, unpredictable, or non-critical workloads, and for companies that genuinely don’t want to run their own data centers. What’s changed is the assumption that ‘The Cloud’ is the default answer for everything, especially including the workloads that are always on, always critical, and too important to be someone else’s shared-tenancy problem.
That’s the story I told in a recent column about a Nonstop customer (issue 28) who was courted hard by a major public cloud provider promising a 3-month migration with better availability and modernization at a lower cost. On closer inspection, the honest estimate was 36 months, and the SLAs the customer actually needed couldn’t be met. The customer stayed on Nonstop, upgraded to current systems, and saved $100 million+ over five years in the process. At the time, that felt like a Nonstop-specific story. It increasingly looks like it was an early data point in a much broader trend.
I don’t think this is really a story about cloud failing. It’s a story about enterprises getting past the hype cycle and doing the arithmetic – on cost, on control, on regulatory exposure, on what “always on” actually requires – workload by workload, instead of by default. That’s a maturity Nonstop customers arrived at a long time ago, because the platform never let you pretend availability and control were optional. It’s good company to be in, and it’s a trend worth watching closely over the next few years.
Repatriation adoption rates
- Cloud Repatriation: Trends and Statistics of Enterprises Leaving Cloud — Barclays survey (83% planning repatriation), IDC (70-80% repatriating annually), Flexera (42% of workloads moving)
- The Great Repatriation: Why Enterprises Are Moving Workloads Back On-Premises — Barclays CIO Survey (86%), IDC 2024 (70%), 2025 Sovereignty Survey (97% of mid-market orgs)
As always, I welcome your thoughts and reactions. Email me at justin.simonds@hpe.com.
Be the first to comment