1. Egress Bandwidth Economics & The Cloud Tax

The primary economic differentiator between hyperscaler clouds and dedicated hosting for email delivery is data transfer out (egress bandwidth). When sending 25 million transactional or marketing emails per month with an average payload size of 100 KB (including HTML, headers, tracking pixels, and inline attachments), your outbound transmission totals approximately 2,500 GB (2.5 TB) of egress monthly.

On Amazon Web Services, outbound data transfer over the Internet from Amazon EC2 instances to the public web is billed at a tiered rate starting at $0.09 per GB (after the first 100 GB free tier). At 2.5 TB, the monthly egress tax alone exceeds $216/month, completely separate from EC2 instance compute costs, EBS volume provisioned IOPS charges, and Elastic IP reservation fees. Scale that volume to 100 million messages per month (10 TB egress), and AWS charges $890/month just to push bytes out of the datacenter switch.

By contrast, Hetzner Online offers dedicated server lines (such as the AX42 or EX44 with AMD Ryzen or Intel Core processors and 64 GB DDR5 RAM) with unlimited or 20 TB+ free outgoing traffic at 1 Gbps or 2.5 Gbps unthrottled symmetric burst capacity. For email operations, Hetzner’s bandwidth cost is effectively $0.00, transforming a variable, open-ended cloud expense into a predictable, flat-rate monthly line item.

2. Port 25 Policies & SMTP Restriction Realities

Standard Mail Transfer Agents (MTAs) communicate peer-to-peer across the internet over TCP Port 25 (RFC 5321). However, major hyperscalers enforce severe restrictions on outbound Port 25 to protect their IP space from botnets and compromised credit card signups.

3. Disk Spool I/O & Queue Concurrency

When high-volume MTAs like PowerMTA, KumoMTA, or Postfix handle inbound spikes of hundreds of thousands of messages, the bottleneck is almost never CPU or RAM; it is disk queue serialization (spool I/O). Every message received by the MTA must be written to disk (fsync) before issuing the 250 2.0.0 Message accepted for delivery acknowledgment to prevent message loss during crashes.

# Hetzner Dedicated (2x 1TB NVMe PCIe Gen4 Software RAID 1):
fio --name=mta_spool --rw=randwrite --bs=4k --ioengine=libaio --iodepth=64 --size=4G --direct=1
Result: 142,500 IOPS | Latency: 0.18ms | Throughput: 570 MB/s

# AWS EC2 c6i.2xlarge with Standard gp3 EBS (3,000 IOPS baseline):
fio --name=mta_spool --rw=randwrite --bs=4k --ioengine=libaio --iodepth=64 --size=4G --direct=1
Result: 3,000 IOPS | Latency: 3.82ms | Throughput: 12 MB/s
* Upgrading to provisioned io2 EBS (32,000 IOPS) adds $450/month to AWS bill!

On Hetzner bare-metal servers, physical NVMe drives communicate directly across the motherboard’s PCIe bus with zero hypervisor serialization, delivering over 100,000 sustained random write IOPS. On AWS EC2, Elastic Block Store (EBS) is network-attached storage that routes disk traffic across an internal network fabric, causing latency spikes and queue backlogs under heavy campaign flushes unless provisioned at prohibitive costs.

4. Dedicated IPv4 Allocation & Subnet Isolation

Email deliverability depends on the reputation of the sender’s IP address. AWS Elastic IPs (EIPs) are heavily recycled across millions of dynamic cloud users, meaning a newly allocated AWS Elastic IP frequently carries a toxic history of Spamhaus SBL listings, Microsoft SNDS blocks, and Yahoo deferrals from prior tenants.

Furthermore, AWS assigns individual /32 IPs scattered across disparate classful subnets. If an adjacent IP in that AWS subnet gets blocked by Spamhaus, entire CIDR ranges often face collateral damage. With Hetzner Online, enterprise senders can purchase clean, contiguous IPv4 subnets (such as a /29, /28, or /27 block) directly assigned to their server MAC address, establishing a clean, unified IP reputation boundary that they exclusively control.

5. Hetzner vs AWS Architectural Matrix