We recently asked a simple question on LinkedIn: what's your actual ROI on running in AWS, GCP, or Azure?
The responses were overwhelming—and overwhelmingly negative. Business leaders, indie hackers, and engineering teams running real workloads all reported the same thing: cloud costs have exploded over the past five years, and nobody can clearly explain why.
This isn't a new complaint. But the scale of overspend has reached a tipping point. Teams that moved to the cloud to save money are now spending multiples of what bare metal would cost—and getting less control in return.
The Hyperscaler Tax
The big three cloud providers built their business on a compelling pitch: don't manage servers, just deploy your code. Pay only for what you use. Scale infinitely.
The reality for most teams looks different:
Costs that grow faster than revenue. A startup spending $500/month on AWS at launch is often spending $15,000-$50,000/month within two years—without a proportional increase in traffic or complexity. The pricing models are designed to be cheap at entry and expensive at scale.
Egress fees that penalize growth. Want to serve data to your users? That costs extra. Want to move your data to another provider? That costs a lot extra. AWS charges up to $0.09/GB for data transfer out. For a platform serving terabytes monthly, egress alone can be a six-figure annual line item.
Complexity that requires specialists. The "don't manage servers" promise evolved into "manage 200+ services, each with its own pricing model, configuration surface, and failure modes." Most teams now need dedicated cloud engineers just to understand what they're paying for.
Reserved instances and savings plans that lock you in. The discounts are real—30-60% off on-demand pricing. But they require 1-3 year commitments, and they're the cloud equivalent of a gym membership: you're paying whether you use it or not.
The Numbers Nobody Wants to Publish
We've audited cloud spend across dozens of client projects. The patterns are consistent:
- 40-70% of provisioned compute is idle at any given time
- Database costs are 3-5x what the same workload would cost on managed bare metal
- Managed services (queues, caches, search) carry 5-10x markups over self-hosted equivalents
- Networking costs are often the fastest-growing line item—and the least understood
A mid-stage SaaS product spending $20,000/month on GCP is typically using $4,000-$6,000 worth of actual compute and storage. The rest is margin, managed-service markup, and architectural inefficiency encouraged by the platform itself.
The hyperscalers are not charities. Their gross margins—consistently above 60% for AWS and Azure—reflect what they're charging relative to cost. You're paying for convenience. The question is whether you're getting $14,000/month worth of convenience.
Want to know what your workload actually costs? We help teams audit cloud spend and model alternatives. No vendor pitch—just numbers.
The Sovereignty Problem
Cost is one dimension. Control is another.
If you're running on AWS, GCP, or Azure, your data lives in infrastructure governed by US law. That's not a political statement—it's a legal fact. The CLOUD Act gives US authorities the ability to compel disclosure of data stored by US companies, regardless of where the servers are physically located.
For European businesses, this creates a growing tension:
GDPR compliance gets murky. Your data might sit in eu-west-1, but the company operating that data center is headquartered in Seattle. European regulators are increasingly scrutinizing this arrangement.
Industry regulations are tightening. Financial services, healthcare, and public sector organizations in the EU are facing explicit requirements for sovereign infrastructure. "It's in an EU region" is no longer sufficient for many compliance frameworks.
Customer expectations are shifting. Enterprise buyers in Europe increasingly ask where data is hosted and under whose jurisdiction. "AWS Frankfurt" raises more follow-up questions than it used to.
Data residency isn't data sovereignty. Having servers in the EU doesn't mean EU law exclusively governs the data on them. True sovereignty requires that the infrastructure operator is also subject to EU jurisdiction—not just the hardware location.
This isn't hypothetical risk. Schrems II invalidated the EU-US Privacy Shield. Its successor, the EU-US Data Privacy Framework, faces ongoing legal challenges. Organizations building on US-controlled infrastructure are exposed to regulatory shifts they can't predict or control.
What the Alternative Actually Looks Like
The reason teams stay on hyperscalers isn't that alternatives don't exist. It's that alternatives historically meant going back to managing servers yourself—racking hardware, patching operating systems, being on-call at 3 AM.
That tradeoff made sense a decade ago. It doesn't anymore.
Modern self-hosted PaaS platforms give you the developer experience of a managed cloud—git push deploys, automatic scaling, managed databases—running on infrastructure you actually control. European infrastructure, operated by European companies, under European law.
The cost difference is dramatic. The same workloads that cost $20,000/month on GCP can run for $3,000-$4,000/month on properly architected European infrastructure. That's not a theoretical number—it's what we're seeing across real production workloads.
The savings come from:
- No hyperscaler margin. You're paying for compute, not for a brand.
- No egress fees. Your data moves freely.
- Right-sized infrastructure. Instead of picking from predefined instance types designed to upsell, you provision exactly what you need.
- Simpler architecture. When managed services don't cost 10x their open-source equivalents, you make better architectural decisions.
This is what we're building. NuPaaS is a self-hosted PaaS for Europe—sovereign infrastructure with the developer experience of a modern cloud platform. Up to 80% cost reduction compared to GCP, with full EU data sovereignty.
"But What About Reliability?"
This is the objection we hear most. And it's fair—the hyperscalers have invested billions in reliability engineering. Multi-region, multi-AZ, automatic failover. That's real.
But let's be honest about what most teams actually use:
- Most applications run in a single region
- Most databases have a single primary with one replica
- Most teams have never tested a cross-region failover
- Most outages are caused by application bugs, not infrastructure failures
You don't need three availability zones across two continents for a SaaS product serving European customers. You need solid infrastructure in one or two European locations, automated failover for your database, and proper backup procedures.
The hyperscaler reliability pitch sells you insurance you mostly don't need at a premium you definitely notice.
Who Should Actually Stay on AWS
To be fair, the hyperscalers are the right choice for some workloads:
Genuinely global products that need sub-50ms latency on every continent. The hyperscalers' global networks are hard to replicate.
Workloads with extreme burst requirements — going from 100 to 100,000 requests per second in minutes. The elastic capacity of the big three is unmatched.
Teams deeply invested in provider-specific services — DynamoDB, BigQuery, Cosmos DB. If your architecture depends on proprietary services, migration is a larger conversation.
Very early-stage startups burning through cloud credits. AWS Activate, GCP for Startups, and Azure credits are effectively free infrastructure. Use them—just plan your exit before they expire.
For everyone else—especially European companies running predictable workloads with predictable growth—the hyperscaler premium is increasingly hard to justify.
The Math That Should Keep CTOs Up at Night
Consider a typical B2B SaaS company:
- $30,000/month on cloud infrastructure
- Growing 20% year-over-year
- 3-year planning horizon
At current hyperscaler pricing, that's $1.08 million in cloud spend over three years—and that assumes zero price increases, which historically hasn't been the case.
The same workload on sovereign European infrastructure: roughly $216,000 over three years.
That's $864,000 in savings. Not in theoretical efficiency gains. Not in "up to" marketing numbers. In actual infrastructure cost difference.
That's runway. That's additional engineering hires. That's the difference between needing a bridge round and not.
Making the Move
Migrating off a hyperscaler isn't trivial. But it's also not the multi-year odyssey people imagine. For most applications:
-
Audit your actual usage. Most teams are surprised by how little of their cloud provider they actually use. Compute, a database, object storage, maybe a queue. That's it.
-
Identify provider-specific dependencies. Lambda functions, proprietary databases, managed ML services. These need alternatives or rewrites.
-
Plan a staged migration. Start with stateless services. Move databases last. Run in parallel until you're confident.
-
Measure the difference. Compare real costs, real performance, and real operational overhead. The numbers speak for themselves.
The teams that have made this move consistently report the same things: lower costs, simpler operations, and the unexpected benefit of actually understanding their infrastructure again.
Ready to see what your workload would cost on sovereign infrastructure? We'll model your current cloud spend against NuPaaS and show you the real numbers—no commitment, no sales pitch, just math.