Skip to main content
Dora Noda

Dora Noda

Software Engineer

2298 posts · View all authors

The 2026 DRAM Supercycle Is Permanent: What 'Not Cyclical' Means for Every Self-Hosted Hardware Bet
·Dora Noda·18 min

The 2026 DRAM Supercycle Is Permanent: What 'Not Cyclical' Means for Every Self-Hosted Hardware Bet

DRAM contract prices jumped 90-95% in Q1 2026 and another 58-63% in Q2 — and IDC says this isn't a cycle, it's a permanent reallocation of silicon toward AI. A worked cost table shows what the supercycle does to every 'self-host on a Hetzner box' comparison, and why the cost advantage over metered PaaS billing survives even as both sides climb.

self-hosting
PaaS
cost-optimization
infrastructure
+1
Fly.io Added Two Billing Lines in 2026: What Metering Creep Costs Against a Flat Hetzner Box
·Dora Noda·11 min

Fly.io Added Two Billing Lines in 2026: What Metering Creep Costs Against a Flat Hetzner Box

Fly.io started billing for volume snapshots and inter-region private networking in early 2026 — no sticker-price hike, but two new dimensions to forecast. A line-by-line invoice for a real multi-region app shows what they add and why a flat Hetzner box is immune to metering creep.

PaaS
self-hosting
cost-optimization
infrastructure
Fly.io Is Killing GPUs on August 1: What Two Retreats in Three Years Say About Renting vs Owning Your Accelerators
·Dora Noda·10 min

Fly.io Is Killing GPUs on August 1: What Two Retreats in Three Years Say About Renting vs Owning Your Accelerators

Fly.io's A10/L40S/A100 GPUs go unavailable August 1, 2026 — its second GPU retreat since 2024 — while its jobs page sits at zero. A concrete migration math of where inference workloads land on rented GPU clouds vs owned bare-metal nodes, and why an owned GPU pool can't be sunset.

self-hosting
PaaS
infrastructure
AI
+1
Fly.io Killed Its Subscription Tiers in 2024 and Kept Adding Billing Lines Through 2026: What Pure Usage-Based Pricing Actually Costs an Always-On App at Scale
·Dora Noda·14 min

Fly.io Killed Its Subscription Tiers in 2024 and Kept Adding Billing Lines Through 2026: What Pure Usage-Based Pricing Actually Costs an Always-On App at Scale

Fly.io cut its Hobby/Launch/Scale plans in October 2024 for per-second usage billing, then added snapshot and egress-IP charges in 2026. A line-by-line model prices the same always-on app on Fly.io versus a flat Hetzner CX23 and shows why the forecast matters more than the sticker price.

PaaS
self-hosting
cost-optimization
infrastructure
+1
Hetzner's 2026 Price Shocks: What Two Repricings Did to the 'Owning Hardware Is Cheaper' Pitch
·Dora Noda·13 min

Hetzner's 2026 Price Shocks: What Two Repricings Did to the 'Owning Hardware Is Cheaper' Pitch

Hetzner raised prices twice in five months — 30% in April, up to 176% on CPX/CCX in June — driven by a 171% DRAM surge. Here is what changed per instance family and what a real Cluster API fleet costs before and after both hikes.

self-hosting
PaaS
hosting
Kubernetes
Hetzner Raised Prices Twice in 10 Weeks: What Your Self-Hosted Fleet Really Costs Now
·Dora Noda·14 min

Hetzner Raised Prices Twice in 10 Weeks: What Your Self-Hosted Fleet Really Costs Now

Hetzner raised cloud prices 30–37% on April 1 and up to 204% more on June 15 for CPX/CCX instances. A recomputed fleet model shows post-June costs at $63–78 on CX/CAX and $136–184 on CPX/CCX versus $75–150 on Render, Railway, and Fly.io at their own 2026 prices.

self-hosting
PaaS
cost-optimization
infrastructure
+1
Two Incidents in Four Days: What Hetzner's Provisioning Delay Actually Costs When You Don't Multi-Cloud
·Dora Noda·12 min

Two Incidents in Four Days: What Hetzner's Provisioning Delay Actually Costs When You Don't Multi-Cloud

Hetzner logged two incidents four days apart in July 2026 — a rack network fault and a resource-creation delay. Only one would have taken your tenants down, and telling them apart is the real operator skill.

self-hosting
PaaS
Kubernetes
infrastructure
+1
Hetzner's €889 RTX PRO 6000: What Self-Hosted AI Inference Actually Costs When You Already Own the Fleet
·Dora Noda·15 min

Hetzner's €889 RTX PRO 6000: What Self-Hosted AI Inference Actually Costs When You Already Own the Fleet

Hetzner's GEX131 with a 96GB RTX PRO 6000 costs €889/month flat — no per-token meter, no pause button. A worked comparison of what that card actually serves via vLLM, where it undercuts hosted API pricing, and where bursty load makes the monthly commit the expensive choice.

self-hosting
PaaS
infrastructure
AI
+1
Kubernetes Is 80% Cheaper Than Heroku — Until You Pay the Ops Bill
·Dora Noda·15 min

Kubernetes Is 80% Cheaper Than Heroku — Until You Pay the Ops Bill

The viral 80% cheaper claim holds on raw compute — Hetzner beats Railway and Render by 50 to 83% in 2026 dollars — but a realistic ops burden flips the result for small teams. Three worked bills, post-hike Hetzner prices, and the break-even math.

self-hosting
PaaS
Kubernetes
cost-optimization
+1
Showing 46–54 of 2298 posts
Prev6 / 256Next