Ditch the Cloud: Run a Few Linux Servers and Save 10x

The author exited AWS for Hetzner and cut infra costs by ~10x while improving performance, arguing most teams don’t need expensive cloud-managed services. He shows concrete pricing for bare metal and VPS that undercut AWS by wide margins, and notes that buying hardware can be cheaper long-term. With modern datacenters, simple Linux hardening, Cloudflare, and AI-assisted ops, running your own servers is practical and sufficient for most businesses.

Key Points

Cloud providers impose massive markups; renting or buying servers (e.g., Hetzner or modest on-prem hardware) is often 10x–100x cheaper for typical workloads.
Most businesses are small and don’t need enterprise-grade cloud features; a single or few well-sized Linux servers can handle large traffic reliably.
The “you’re using cloud wrong” rebuttal is gaslighting—after serious cost optimization, cloud remains expensive compared to running your own servers.
Backlash is fueled by misaligned incentives and sunk-cost/cult dynamics around cloud tooling and roles; the debate often becomes semantic or dogmatic.
Modern datacenters are resilient, server management isn’t a full-time burden, AI greatly lowers the ops barrier, and basic hardening plus Cloudflare is sufficient for many.

Sentiment

The community broadly agrees with the article's core premise that cloud is overpriced for most workloads, but pushes back firmly against its absolutist tone. Many commenters share their own successful cost-saving migrations away from major cloud providers. However, the discussion is tempered by experienced voices pointing out legitimate cloud use cases around compliance, scale, and team capabilities. The prevailing view is that the author is mostly right but paints with too broad a brush.

In Agreement

AWS and major cloud providers charge vastly more than dedicated server providers like Hetzner for equivalent compute, with concrete examples of significant savings widely shared across the thread
Modern server hardware (high-core-count CPUs, large NVMe drives, abundant RAM) has made vertical scaling on a single machine viable for the vast majority of workloads
Cloud disk I/O uses a burst-credit model that degrades badly under sustained load, making local NVMe storage dramatically faster for many use cases
The original advantages of cloud (instant provisioning, no procurement delays) have been largely replicated by dedicated server providers offering hourly billing and API-driven setup
Cloud vendors deliberately create vendor lock-in through free startup credits, proprietary services, and expensive egress pricing
Most businesses vastly overestimate their need for auto-scaling, multi-region HA, and microservices architectures — simpler setups would suffice
AI tools have made Linux server administration significantly more accessible, reducing the sysadmin knowledge barrier

Opposed

Cloud providers offer pre-certified compliance (SOC2, ISO27001) that is extremely difficult and time-consuming to replicate with self-hosted infrastructure, especially in regulated industries like finance
Engineering time has real costs — teams without sysadmin expertise will spend far more setting up and maintaining self-hosted infrastructure than paying for managed services
For B2B companies, hosting on recognized cloud providers builds customer trust and simplifies SLA conversations and security audits
The article presents a false binary — the real answer depends on team skills, regulatory requirements, scale, and business needs; it is not universally one or the other
Cloud adoption was often driven by organizational dynamics (CapEx vs OpEx, procurement bureaucracy) that remain real factors in larger companies
At very large scale (tens of thousands of servers), cloud tooling and automation provide genuine efficiency advantages that are hard to replicate
Self-hosting email remains effectively impossible due to ISP restrictions and spam filtering requirements, illustrating that some managed services are worth paying for