Northflank Now Runs 16 Managed Regions, Including Melbourne and Tokyo — What It Would Take a CAPH Fleet to Say the Same

Northflank's regions page currently lists sixteen managed locations spanning EMEA, the Americas, and Asia-Pacific — including australia-southeast (Melbourne) and asia-northeast (Tokyo), the latter with GPU support at launch. A self-hosted, Cluster-API-on-Hetzner fleet has six. None of them are in Japan or Australia. Before comparing what either side charges per gigabyte, that's the number that actually decides whether "we're in Tokyo too" is a real claim or marketing parity: a BYOC vendor can click a region into existence on infrastructure it leases from a hyperscaler-adjacent provider; a fleet that owns its own boxes can't stand up a region where the hardware vendor it depends on doesn't have a datacenter.

What Northflank Actually Shipped​

On December 11, 2025, Northflank cut network egress from $0.15/GB to $0.06/GB — a 60% reduction — dropped NVMe disk pricing from $0.30/GB/month to $0.15/GB/month, and removed per-request pricing entirely. The same announcement introduced Melbourne and Tokyo as new managed regions, with Tokyo shipping GPU support (L4, A100, H100, and later B200) from day one. Northflank framed it as a global price cut applied uniformly across its then-8 existing regions, with 8 more following in January 2026.

Seven months later, that expansion has landed. Northflank's current region list runs to sixteen:

Egress is now a flat $0.06/GB everywhere on that list, ingress is free, and zonal-redundancy transfer runs $0.02/GB — the same number in Melbourne as in Frankfurt. That uniformity is the actual product: a team can point a workload at Tokyo the same way it points one at Amsterdam, on the same billing line, with no separate negotiation, no separate contract, and no separate infrastructure to reason about.

The Catch: These Are Managed Regions, Not Owned Hardware​

Sixteen regions sounds like Northflank out-built every self-hosted alternative on this list by a wide margin. It didn't — it out-rented them. Northflank's managed cloud is a multi-tenant control plane running on capacity it leases across cloud and colocation providers; a customer gets a region, not a rack. That's the correct trade for a BYOC platform selling instant availability, and it's exactly why Northflank could add Melbourne and Tokyo in the same changelog entry as a price cut — no datacenter build, no hardware lead time, just a new placement target on infrastructure Northflank (or its upstream provider) already operates somewhere nearby.

A Cluster-API-on-Hetzner fleet doesn't get that shortcut, because the entire pitch — cheaper unit economics from owning the machines instead of renting a managed control plane — depends on the hardware actually existing under a Hetzner account first. And Hetzner's real footprint, unchanged by anything Northflank did in December, is six regions:

That's it. No Tokyo. No Melbourne. No presence anywhere in Japan or Australia, and Hetzner hasn't announced plans to add either. Singapore is the only APAC option a CAPH-based fleet has, and — per this list's own prior look at that region — it's a narrower one than the EU baseline: no Object Storage, a network zone that doesn't bridge back to eu-central over Hetzner's private networking, and a CAPH quickstart whose examples all still use fsn1, not sin1. Standing up Singapore today already means budgeting for a WireGuard overlay and an EU-routed object-storage plan that the managed-region model never makes a customer think about.

What "Tokyo Too" Would Actually Require​

For a CAPH fleet, saying "we're in Tokyo" isn't a config flag — it's one of two much bigger commitments:

1. Wait for Hetzner to build a Japanese or Australian datacenter. Hetzner has shipped exactly one new region-class location in the last four years (Singapore, 2024), following a roughly two-year buildout after the 2022 announcement. There's no public signal of a Tokyo or Melbourne facility on that roadmap, and nothing a platform layer can do accelerates a hardware vendor's own capex timeline.
2. Wire in a second Cluster API infrastructure provider for that geography. Cluster API is provider-agnostic by design — a fleet could run CAPH for its Hetzner-owned core and a different CAPI provider (targeting a bare-metal or colocation partner that does have Tokyo capacity) for an APAC extension. That's a real, executable path, but it means operating two infrastructure providers' worth of machine lifecycle, node images, and CAPI controller versions instead of one — a materially bigger operational surface than Northflank toggling a region dropdown, and not something to take on speculatively ahead of actual APAC tenant demand.

Neither option ships this quarter. Both are more honest answers than pretending a bex.yml region field could point at ap-northeast-1 tomorrow.

The Economics Still Favor Owning the Six You Have​

Here's the number that keeps this from being a one-sided story: per gigabyte, the region a self-hosted fleet can reach is still cheaper than the one Northflank just discounted. Hetzner's EU overage rate is €1.00/TB — €0.001/GB, roughly 60x cheaper than Northflank's new flat $0.06/GB. Even Hetzner's pricier Singapore overage tier, at €7.40/TB (€0.0074/GB), undercuts Northflank's post-cut global rate by close to 8x. Northflank's $0.06/GB is a genuine improvement over its old $0.15/GB — and free ingress plus zero per-request charges are real wins a self-hosted operator has to replicate by hand — but it's still pricing a managed, multi-tenant control plane, not a machine a customer owns outright. The 60% cut narrows the gap to owned Hetzner hardware; it doesn't close it.

That's the actual trade on the table, stated plainly: Northflank sells sixteen regions today, including two a Hetzner-based fleet structurally cannot reach without either a datacenter Hetzner hasn't built or a second infrastructure provider nobody has wired in yet — and it prices that convenience at a real premium per gigabyte over what owning the hardware costs in the six regions that do exist. For APAC-specific traffic today, that premium is often the right call regardless: it beats routing every Melbourne or Tokyo request back through Singapore or across the Pacific to Hillsboro, a distance no discounted per-GB rate fixes.

A Worked Example: 5TB/Month Out of Tokyo​

Put a number on what that premium actually costs. A tenant serving 5TB/month of outbound traffic to Japanese users has three real options today:

The Hetzner-Singapore path runs roughly 8x cheaper than Northflank's Tokyo egress bill even after eating its 7.4x-EU overage premium — but it's still serving Japanese users from Singapore, not Tokyo, because that's the only APAC region a CAPH fleet has. Northflank's $300 buys an actual point of presence in Japan, with sub-20ms latency to Tokyo and Osaka a Singapore-origin request can't match. Neither number is "wrong" — they're pricing two different things, and conflating them is exactly the marketing-parity trap this piece opened with.

The GPU Angle Nobody Should Skip​

Tokyo shipping with L4/A100/H100/B200 support at launch is a second, separate claim worth pulling apart from networking. A self-hosted PaaS betting on AI-agent sandbox workloads (this list's own repeated GPU-scheduling and DRA coverage) doesn't just need a Tokyo region — it needs GPU-capable Hetzner nodes to exist in whatever region it does have, and today that's a Falkenstein/Nuremberg-only story; Hetzner's GPU SKUs haven't reached Singapore, let alone a region that doesn't exist yet. So the honest gap here is two layers deep: no Hetzner presence in Japan or Australia at all, and even where Hetzner APAC capacity does exist, no GPU inventory on it yet. Northflank's Tokyo GPU support closes both gaps in one changelog entry a self-hosted fleet can't match by wiring in a second CAPI provider alone — that second provider would also need to source GPU capacity in-region, which is its own sourcing problem independent of the Kubernetes plumbing.

What This Means for a Self-Hosted Roadmap​

The honest scope for 2026 is Singapore, with its documented gaps priced in — not a promise to match Northflank's Melbourne and Tokyo footprint on a timeline nobody controls. A team whose actual users sit in Australia or Japan and who need that latency (or GPU capacity) now has a real, current reason to reach for a managed BYOC platform with a region already there, or to deliberately stand up a second CAPI provider rather than default to one. Everyone else is better served by the six regions already on owned hardware, at unit economics the worked example above shows still beat a rented region on a per-gigabyte basis — the fleet just needs to stop letting "sixteen regions" imply parity that a datacenter Hetzner hasn't built, and GPU inventory it doesn't stock, haven't actually delivered.

Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own. Star the repo on GitHub or deploy your first app today.