86% of CIOs Are Leaving the Cloud — I Told You This Was Coming

86% of CIOs Are Leaving the Cloud — I Told You This Was Coming

Let me tell you something I've been watching for the last three years. The pendulum is swinging. And it's swinging hard.

I've been running hosting infrastructure for over a decade. I've watched the hyperscalers — AWS, Azure, Google Cloud — spend billions convincing the world that the only way to run a business is to hand them your servers and pray the bill doesn't ruin you. For a while, it worked. Every company with a credit card and a dream migrated everything to the cloud because that's what the consultants said to do. But here's what the consultants didn't tell you: the bill comes due. And in 2026, that bill is arriving with interest.

The Number That Changes Everything

A Barclays CIO survey dropped a number that should stop every founder and CTO reading this cold: 86% of CIOs now plan to move at least some workloads back from public cloud to private or on-premises infrastructure. That's the highest rate ever recorded. In 2020, that number was 43%. In four years, it doubled.

Now, let me be clear about what this number actually means. It doesn't mean 86% of companies are abandoning the cloud. The skeptics are right about one thing — public cloud spending is still growing. Gartner forecasts it hit $723 billion in 2025, up 21.5% year over year. IDC says it'll pass $1 trillion in 2026. AWS revenue is still growing 20% a quarter. The cloud isn't dying.

But here's what is dying: the blind faith that the cloud is always cheaper. The idea that you should put everything on AWS or Azure just because that's what the industry tells you to do. That orthodoxy is cracking, and I've got the receipts.

37signals, Dropbox, and GEICO — Three Proof Points That Should Terrify the Hyperscalers

Let me name names, because that's how I operate.

37signals — the company behind Basecamp and HEY. They fully exited AWS and are now saving roughly $2 million per year. Over five years, that's $10 million in savings. Not projected. Not hypothetical. Actual, audited savings from a company that moves hundreds of millions of dollars in revenue. DHH didn't do this as a PR stunt. He did it because the math didn't work. And he published every number so the rest of us could learn from it.

Dropbox did it even bigger. Between 2015 and 2016, they moved 90% of their customer data off AWS to their own custom-built data centers. The result? $75 million in cumulative savings over two years. Their gross margins jumped from 33% to 67%. Think about that. Going from cloud-reliant to owning your infrastructure more than doubled your margins. That's not theory. That's a public company's financial statements.

GEICO — Warren Buffett's insurance giant. They spent a decade moving 600+ applications to the cloud. Their costs ballooned to 2.5 times over budget. They launched a project called "Project Boomerang" to bring it all back home. The result? A 65% cost reduction and $325 million in projected annual savings. They paid back the entire repatriation investment in 14 months. Fourteen months for a 65% cost cut on a decade of cloud spending.

These aren't small companies. These aren't startups running on a $10 VPS. These are billion-dollar enterprises that did the math and decided the cloud wasn't worth it for their steady-state workloads. And they're not alone.

Why the Math Stops Working at Scale

Here's the thing about cloud pricing that nobody tells you during the migration hype. The cloud is expensive for one very specific reason: you're paying for someone else's flexibility. AWS isn't cheaper than buying your own hardware. It's more expensive. What you're paying for is the ability to spin up and tear down resources on demand. That flexibility is valuable — for the right workloads.

But here's the catch. Most workloads in most companies are steady-state. Your database doesn't double in size overnight. Your web servers don't need to scale to a million users at 3 AM. Your file storage doesn't need to be globally distributed with 99.999% durability unless you're Dropbox. For the vast majority of workloads, the 80/20 rule applies: 80% of your cloud spend is on resources that run 24/7/365 with predictable utilization.

Flexera's 2025 survey found that 27% of cloud spending is pure waste. Underutilized resources. Oversized instances. Forgotten development environments running in production accounts. That's over $195 billion dollars of waste in 2025 alone. And the hyperscalers are happy to collect it — they're not in the business of telling you to downsize.

When you own the hardware, every cost is visible. You know exactly what a server costs per month. You know your power cost. You know your bandwidth cost. The bill doesn't surprise you. The cloud turns fixed costs into variable costs, which sounds great in a boardroom presentation until your variable costs end up 2.5x your projection — ask GEICO how that feels.

The Repatriation Playbook — What Actually Works

Now, I'm not saying every company should go full 37signals and exit the cloud entirely. That's the headline, but it's bad advice for most businesses. The data shows only about 8% of enterprises plan full cloud exits. The rest are doing something smarter: targeted repatriation.

Here's what the smart companies are doing in 2026:

1. Identify your steady-state workloads. If a server, database, or storage volume has run at consistent utilization for 12+ months, it's a candidate for repatriation. The cloud's elasticity premium isn't buying you anything if you're never going to scale that workload up or down.

2. Start with storage. Storage is the easiest repatriation win. It's latency-tolerant, doesn't need complex orchestration, and the cost difference between AWS S3 and self-hosted object storage is massive. Dropbox proved this at billion-dollar scale. You can prove it at any scale.

3. Keep dev/test in the cloud. Development environments are the perfect cloud workload. They're ephemeral, unpredictable, and you don't care about their uptime. Move your staging environments back but keep your dev boxes where they are.

4. Don't try to replicate AWS in your data center. This is where most repatriation attempts fail. You don't need 200 microservices, load balancers, auto-scaling groups, and a dozen managed services in your own rack. You need a few well-specced servers, good connectivity, and competent sysadmins. The hyperscalers convinced you that complexity is normal. It's not.

5. Hedge your egress costs. This is the killer. Cloud providers charge obscene amounts for data leaving their network. Before you start moving workloads back, calculate your egress costs. Build a transition plan that minimizes data transfer fees during the migration. The hyperscalers designed their pricing to punish you for leaving — plan around it.

What This Means for Independent Hosting Providers

Here's where this gets personal for me. I've been running independent hosting infrastructure for over a decade. I've watched the cloud-first wave wash over the industry and drown dozens of hosting companies that couldn't compete with the marketing budgets of AWS and Azure. But the tide is turning.

Every workload that comes back from the cloud is an opportunity for independent providers. Companies that repatriate need colocation space, dedicated servers, managed hosting, and competent support. The hyperscalers can't offer a warm handoff to a real human being who knows your infrastructure. That's our advantage.

The numbers back this up. While cloud spending is growing at 21%, independent hosting providers have been quietly growing at 8-15% annually as the repatriation trend accelerates. The companies that left the cloud aren't going back to running servers in a closet — they're going to providers who offer better economics and real support.

If you're running an independent hosting business, 2026 is your moment. The "cloud or die" narrative is dead. Companies are looking for alternatives. Be ready when they call.

The Bottom Line — The Honeymoon Is Over

Look, I don't hate the cloud. I use cloud services for specific things. Object storage for backups. CDN for static assets. Compute for burst workloads that don't justify dedicated hardware. The cloud is a tool, and it's a great one for the right job.

But the idea that everything belongs in the cloud was always a sales pitch, not a technical truth. And the sales pitch is losing its power. When 86% of CIOs are actively planning to move workloads back, when billion-dollar companies are publishing their $10 million, $75 million, and $325 million savings from leaving the cloud, the honeymoon is officially over.

The question isn't whether repatriation is real. It is. The question is whether you'll be smart enough to figure out which workloads belong where before your cloud bill eats your margins alive. Because the hyperscalers aren't going to tell you. They never did.

Stop treating the cloud as a religion and start treating it as a procurement decision. Your P&L will thank you.

— Allan Ali, Founder