Cloud Repatriation 2026: Why the Trend Is Moving Back to Bare Metal
For years, "Cloud First" was the standard narrative: faster to start, less CapEx, more agility. In 2026, the debate has become much more sober. Cost control, predictability, and regulatory requirements are leading companies to selectively bring workloads back from the public cloud – not necessarily "everything," but enough to be strategically relevant.
This isn't just anecdotal: In the Flexera 2025 State of the Cloud Report, companies indicate that 21% of cloud workloads have already been repatriated. At the same time, public cloud continues to grow – so repatriation is more a sign of hybrid maturity than a "cloud ending." 1
In parallel, Gartner continues to forecast strongly rising public cloud spending (including $723.4 billion for 2025) and expects hybrid cloud to become mainstream by 2027. The message is clear: Cloud is here to stay, but its role is shifting. 2
The "Trillion-Dollar Paradox": When Cloud Costs Become a Margin Brake
Andreessen Horowitz (a16z) articulated the dilemma early: Cloud is ideal for getting started quickly – but at scale, it can become a structural margin brake. In "The Cost of Cloud, a Trillion Dollar Paradox," Sarah Wang and Martin Casado argue that cloud costs can measurably impact profitability and even market valuation at larger software companies. 3
"You're crazy if you don't start in the cloud; you're crazy if you stay on it." — Martin Casado (a16z)
Important: Even a16z emphasizes this isn't a blanket "get out of the cloud" message, but rather a framework to treat infrastructure costs as a first-class KPI – and choose the right operating model for each workload. 3
As a counterpoint (and good reality check), CloudZero's perspective is worth reading: Repatriation can make sense, but it's by no means automatically the best answer – often the real problem is "lift-and-shift without cloud-native architecture" or missing cost ownership in Engineering/FinOps. 4
Case Study 1: 37signals (Basecamp/HEY) – Cloud Exit with Published Cost Breakdown
37signals is one of the most transparent examples because CTO David Heinemeier Hansson (DHH) publicly documented the exit, and Basecamp maintains a curated overview page with relevant articles. 5
Starting Point (published):
- In 2022, cloud costs were around $3.2M (separately itemized without the later S3 exit phase). 6
Actions Taken (published):
- Investment in own servers (Dell) + operation in colocation/existing racks; initially calculated at ~$600k hardware, amortized over several years. 6
Results (published):
- By 2023, DHH reports a run-rate reduction and writes that the hardware purchase "pays for itself with current monthly savings in under six months." 7
- For 2024, he calls it "the first clean year of savings": Cloud bill reduced from $3.2M/year to $1.3M/year, savings of nearly $2M/year, and total forecast "over $10M in five years." 8
The key takeaway: For "steady-state" SaaS workloads with predictable growth, ownership/colocation can be massively cheaper – provided you can handle operations and migration. 6 8
Case Study 2: Ahrefs – "Not Going to the Cloud" as an Economic Decision
Ahrefs has also published detailed cost arguments – primarily through Efim Mirochnik on the Ahrefs Tech Blog.
The Known Headline Number: Ahrefs estimates the difference for "own colocation vs. AWS equivalent" in Singapore at ~$400M in additional costs (over about 30 months/"~3 years" in their presentation). 9
Important for Context: Ahrefs openly describes assumptions, simplifications, and deliberately "AWS-favoring" modeling (e.g., EBS as simplified storage substitute, pricing assumptions, region, performance non-equivalences). This makes the source strong – but the number remains an internal model calculation, not an audit. 9
Long-term Perspective (2017–2023): In a follow-up post, Ahrefs looks at historical spending: $122M on-prem/colocation spend since 2017 vs. > $1B hypothetically in AWS (depending on on-demand vs. reserved). 10
Concrete Monthly Costs per Server: The Register reports (referencing the Ahrefs post) ~$1,500/server/month (including acquisition/depreciation) vs. ~$17,557 in AWS for the approximate equivalent. 11
Why Are Companies Repatriating in 2026? The Three Key Drivers
1) Price/Performance and Predictability
In public clouds, you share infrastructure. This isn't a problem per se – but for certain workloads, performance variance becomes expensive (overprovisioning, more complex SLO assurance, more caching).
- AWS defines vCPU as "number of processor threads" (i.e., typically hyperthreads), not as "guaranteed physical core." 12
- The "Noisy Neighbor" effect is documented as an architecture antipattern: One tenant's activity can impact another tenant's performance/reliability. 13
Bare Metal (Dedicated Servers/Colocation) can score here: fewer unknowns, clearer capacity planning, often better price/performance for continuously utilized systems.
2) Regulatory and Provider Control (DORA)
Especially in regulated industries, requirements for resilience, risk management, outsourcing control, and third-party risks have increased.
- DORA applies as EU regulation and has been applicable since January 17, 2025 (Financial Sector + ICT Third-Party Providers). 14
- In November 2025, Reuters reported that EU regulators designated AWS, Google, and Microsoft as "critical" third-party providers for the financial sector. 15
This isn't automatic "get out of US clouds," but it increases pressure to formally manage exit plans, portability, and provider risk.
3) Jurisdiction & Data Access (US CLOUD Act)
With US providers, a key point is: US law can potentially be relevant even when data is physically located in the EU, if it falls under "possession, custody, or control" of a US provider.
- 18 U.S.C. § 2713 explicitly states that providers must fulfill disclosure obligations "regardless of whether … located within or outside of the United States." 16
- Eurojust summarizes the CLOUD Act from an EU perspective. 17
- EDPB/EDPS have also published a joint legal assessment on the CLOUD Act. 18
Important: Switching to EU providers can reduce jurisdiction/access risks, but doesn't automatically "eliminate" them in every scenario.
4) Portability Is Being Politically Enforced: EU Data Act
A practical catalyst: Europe is pushing for switchability. The cloud switching rules of the EU Data Act became applicable on September 12, 2025. 19
In parallel, hyperscalers have adjusted exit/transfer fees:
- Google has eliminated certain transfer fees in EU/UK. 20
- AWS has removed transfer fees for customers switching to competitors. 21
- Google Cloud offers "free network data transfer" for migrations off Google Cloud. 22
This means: Lock-in is becoming regulatorily less attractive – and repatriation/hybrid becomes organizationally easier to implement.
Enabler: "Cloud-like" Operations on Own Hardware Is Realistic Today
What used to be comfortable only with hyperscalers can now be mapped as a Private Cloud/Hybrid Stack:
- Proxmox VE offers a documented API (automation/IaC-friendly). 23
- Kubernetes is an established standard for portable workloads. 24
- Ceph provides distributed object/block/file storage as a building block for private storage layers. 25
- For IaC/automation, Terraform or OpenTofu are established. 26
This makes repatriation less of a "step backward" and more of an operating model that transfers cloud principles to your own infrastructure.
Conclusion: Cloud Is Not Dead – But "Steady State" Needs to Be Recalculated
The cleanest conclusion from the sources is not ideology, but a decision tree:
Public cloud remains strong, but repatriation is a valid, growing component of hybrid strategies in 2026. 1 2
Bare Metal/Colocation is particularly worthwhile when:
- your load is continuously high (steady-state),
- you need to stabilize costs per transaction/unit economics,
- performance variance becomes expensive (noisy neighbors, overprovisioning),
- regulatory/third-party risk requires formal exit and control mechanisms,
- you have the operational discipline (SRE/Infra, monitoring, patch/lifecycle management).
And the 37signals/Ahrefs cases show: When these conditions are met, savings in the millions can be realistic. 8 9 10
Cloud Exit with OutaCloud: Your Partner for Repatriation
Considering bringing workloads back from the public cloud? OutaCloud guides you from assessment to go-live – with experience from dozens of migration projects.
Our Cloud Exit Services:
- Exit AWS – Leave Amazon Web Services for cost-effective infrastructure
- Exit Azure – Move away from Microsoft Azure without vendor lock-in risks
- Exit Google Cloud – Migrate GCP workloads cleanly
- Exit VMware – Escape VMware licensing with open-source virtualization
Every exit starts with a free analysis of your current infrastructure. Get in touch – we'll show you what's possible.
Sources
Footnotes
-
DHH – We stand to save $7m over five years from our cloud exit ↩ ↩2 ↩3
-
DHH – Our cloud exit has already yielded $1m/year in savings ↩
-
DHH – Our cloud-exit savings will now top ten million over five years ↩ ↩2 ↩3
-
Ahrefs – How Ahrefs Saved US$400M in 3 Years by NOT Going to the Cloud ↩ ↩2 ↩3
-
Ahrefs – How Ahrefs Gets a Billion Dollar-Worth Infrastructure With a 90% Discount ↩ ↩2
-
Reuters – EU designates AWS/Google/Microsoft as critical providers under DORA (Nov 2025) ↩
-
Reuters – Google scraps some cloud data transfer fees in EU and UK (Sep 2025) ↩
-
Reuters – AWS removes data transfer fees for clients switching to rivals (Mar 2024) ↩
-
Google Cloud – Eliminating data transfer fees when migrating off Google Cloud ↩
Cloud Migration Service
Break free from the cloud cost trap - up to 90% savings.