86% of CIOs now plan to move workloads from public cloud back to on-premises infrastructure—the highest repatriation rate ever recorded. After a decade of “cloud-first” dogma where questioning AWS meant career suicide, the pendulum is swinging hard. This isn’t cloud failure. It’s cloud pragmatism finally defeating cloud religion.
Companies that migrated everything to AWS, Azure, and GCP in the 2010s are now running the numbers and discovering what 37signals learned the hard way: for mature, predictable workloads running 24/7, the cloud is bleeding them dry. Cloud spending hit $669 billion in 2024 and will reach $840 billion by 2026, but underneath that growth, 83% of enterprises are planning to bring at least some workloads home.
The 37signals Reality Check: $7 Million Saved
When David Heinemeier Hansson announced in October 2022 that 37signals—makers of Basecamp and Hey—was leaving the cloud, the reaction was predictable. Cloud advocates called it regression. DHH called cloud providers “merchants of complexity.” Two years later, the numbers prove DHH was onto something.
37signals was spending $3.2 million annually on cloud services, nearly $1 million on AWS alone. They bought $600,000 of Dell servers. The result? They’ve saved $2 million so far and project $7 million in savings over five years. Their monthly cloud spend dropped 60%, from $180,000 to under $80,000. They still use some cloud services—this isn’t a binary exit—but 90% of their infrastructure now runs on hardware they own.
The math is brutal for cloud providers: a $600,000 upfront investment beats $3.2 million in annual rent. Breakeven came in under two years. For Dropbox, the economics were even starker: they saved $75 million over two years after building their own infrastructure. When your workloads are mature, predictable, and running constantly, cloud’s “infinite scale” premium stops making sense.
The Cost Reality Nobody Talks About
Here’s what the cloud-first evangelists didn’t mention: 43% of IT leaders found cloud migration more expensive than expected. Not a little more expensive—significantly. The promise was OpEx flexibility and automatic savings. The reality is 28-35% of cloud spending gets wasted on over-provisioned resources and idle instances. 54% of that waste stems from lack of cost visibility, and 50% of IT leaders cite complex pricing models as the core problem.
Cloud costs sneak up on you. Data egress fees charge $0.08-$0.12 per gigabyte just to move your own data out. Reserved instances require year-long commitments. Auto-scaling sounds great until you realize it’s billing you for every spike. The companies now repatriating aren’t failing at cloud—they’re succeeding at math.
The breakeven analysis is straightforward: for AI workloads on GPUs, on-premises infrastructure pays for itself in under four months of 24/7 usage. For traditional enterprise workloads, breakeven hits around 18-24 months. Below 20% utilization, cloud wins. Above that, you’re paying a premium for flexibility you don’t need.
It’s Not Just About Money
Cost drives the headline, but 51% of organizations cite data security and privacy as their primary repatriation motivation. Compliance requirements like HIPAA and GDPR favor direct control. Some cloud providers can’t even guarantee your data stays within a specific jurisdiction during transfer—a non-starter for regulated industries. High-profile breaches in 2025 raised awareness fast. Financial services and healthcare are leading the retreat.
AI economics accelerate the trend. Cloud GPU rentals run $2-$4 per hour for NVIDIA H100s. Buy one for $25,000, and it pays for itself in four months of continuous use. When you’re training models daily, ownership beats rental every time. Dropbox didn’t leave cloud just to save money—they needed control to build Magic Pocket, their custom multi-exabyte storage system. Cloud’s abstraction layers become constraints when you’re pushing the limits.
The Death of Cloud Dogma
For a decade, “cloud-first” was gospel. Migrations were success stories. Staying on-premises meant you were behind. CapEx to OpEx shifts were automatically good. Vendors pushed cloud as the solution to everything, and questioning it was heresy.
That era is over. In 2026, 93% of IT leaders have been involved in cloud repatriation projects over the past three years. This isn’t edge cases—it’s mainstream strategy. The new term is “cloud-smart” or “cloud-pragmatic.” Workload placement is driven by economics and technical requirements, not ideology. Hybrid architecture is the norm, not the exception.
Cloud repatriation isn’t regression. It’s maturity. The 2010s taught us cloud can work. The 2020s are teaching us when it should work. Mature workloads with predictable usage belong on-premises. Variable, experimental, bursty workloads belong in cloud. There’s no universal answer—it depends on the workload. The lesson cost billions to learn, but the industry is finally learning it.
What This Means for Developers
Skills are shifting. On-premises expertise is back in demand: Kubernetes on bare metal, hardware knowledge, data center operations, networking. “Cloud engineer” is evolving into “infrastructure engineer” who can design for both cloud and on-prem. Hybrid architecture expertise commands a premium. Understanding total cost of ownership matters as much as understanding code.
Architecture decisions are more nuanced now. Teams need to classify workloads: which stay in cloud, which come back, which never left. Portability becomes a design principle. Infrastructure-as-code applies to both environments. The era of “just use AWS” is over. The era of “use the right tool for the job” is here.
Cloud-first was expensive education. Cloud-smart is the graduation. Companies that migrated everything are now optimizing based on what they learned. That’s not failure—it’s progress. With 86% of CIOs planning repatriation, we’re witnessing cloud computing finally grow up.
