The 25 Gbit headline hides the real number: symmetric

What happened

Stefan Schueller's post on why Switzerland has 25 Gbit residential internet and the US does not hit 470 on Hacker News. The headline number is the one everyone repeated: Init7's Fiber7-X2 plan is 25 Gbps symmetric for ~71 CHF/month, no data cap, static IPv4 and a routed /48 of IPv6. Zurich and Bern get it today; the rollout is metro-level, not universal.

The comparable US plan — Comcast's Gigabit X2 in the handful of markets that have it — is $150 for 2 Gbps down and 200-300 Mbps up. AT&T Fiber's top consumer tier is 5 Gbps symmetric at $180 in a small number of neighborhoods. Verizon Fios caps at 2 Gbps symmetric. Everywhere else, the ceiling on the upload side of a residential connection is somewhere between 20 and 50 Mbps.

The article's framing — that this is a market-structure story, not a technology story — has been covered elsewhere. What has been undercovered is the specific number that makes Init7's plan interesting to a developer: the 25 in the *upload* column, not the download column, is what actually changes the shape of what you can build at home.

Why it matters

The download-vs-upload asymmetry in US consumer plans is not a physical constraint of fiber. Passive optical networks (GPON, XGS-PON) can be configured symmetric — Init7 runs XGS-PON to deliver 10 Gbps symmetric and Active Ethernet for the 25 Gbps tier. Comcast's asymmetry comes from DOCSIS on their hybrid coax plants, but their fiber-to-the-home builds are also being provisioned asymmetric. The 200 Mbps upload cap on a 2 Gbps down plan is a business decision: it protects the upstream capacity of the shared PON tree so the ISP can oversubscribe without visible degradation, and it makes it inconvenient to run anything a paid CDN or cloud provider would rather sell you.

Symmetric is the number that decides whether self-hosting is a hobby or a workflow. A 1 Gbps down / 35 Mbps up plan is a consumption pipe: you can pull Docker images and stream 4K, but you cannot practically host a git remote your team pushes to, you cannot back up a 500 GB Lightroom catalog to your own NAS overnight, and your self-hosted Immich instance is going to feel like a joke the first time your partner tries to sync a phone from work. Push 35 Mbps up to 1 Gbps up and every one of those workflows crosses from 'you'll regret this' to 'why did I ever pay for that.'

Community reaction on the HN thread tracked this. Multiple commenters pointed out that even France's Free, which sells 8 Gbps symmetric FTTH for €40/month, gets less attention than it should because the discussion always defaults to peak download numbers. A recurring line: "I moved from Comcast to Free and canceled Dropbox, Backblaze, iCloud+, and a Hetzner VPS in the same month. The delta paid for the fiber install." That is the practitioner-level consequence the policy debate keeps missing.

The policy detail worth naming precisely is local loop unbundling — the requirement that the incumbent that owns the last-mile physical layer lease it to competitors at a regulated wholesale rate. Switzerland has it via ComCom's rulings on Swisscom fiber. The UK has it via Ofcom's separation of Openreach from BT. France has it via ARCEP's dégroupage regime; Free built its business on it. Japan has NTT wholesale. All four markets have symmetric multi-gigabit consumer plans at prices that would look like a data-entry error on a US ISP bill. The US had unbundling from 1996 to 2005, then the FCC ruled in *NCTA v. Brand X* that fiber operators did not have to share, and every subsequent build has been vertically integrated.

What this means for your stack

If you are one of the small number of US developers who can get a symmetric fiber plan — AT&T Fiber in metro Texas or the Carolinas, Sonic in the Bay Area, Ziply in the Pacific Northwest, a municipal build in Chattanooga or Longmont — the honest recommendation is to price out what you are currently paying to work around asymmetric consumer internet. Object storage for photos and video, a small VPS for a git remote, a paid Tailscale plan you only need because your home IP is behind CGNAT, cloud backup for machines that could rsync to a Synology, a Plex Pass because Jellyfin at home would choke on outbound. It adds up faster than the fiber bill.

If you cannot get symmetric fiber, the practical adjustment is to stop treating self-hosting as a moral position and price it honestly. A $6/month Hetzner CX22 with 40 GB of NVMe hosts Forgejo, Umami, and a small Miniflux fine and does not care that your upload speed at home is 35 Mbps. The self-hosted-at-home dream is real; it is just gated on physical infrastructure most of us do not have.

And if you are building software that assumes symmetric bandwidth — real-time collaboration, WebRTC video, browser-based IDEs, sync engines, federated content — remember that your test conditions on a corporate 1 Gbps symmetric drop are not the conditions the median US consumer is on. Upload-heavy features fail silently and expensively when the pipe going the other way is 30x smaller than the one you designed against.

Looking ahead

The BEAD program is spending $42.5 billion on rural broadband over the next few years. The rules require symmetric 100/20 Mbps as a floor and prioritize fiber, but they do not reintroduce wholesale unbundling. That means the money is going to build another generation of vertically integrated last-mile networks, owned by incumbents whose business model depends on asymmetric consumer plans. The comparison the Schueller post makes is not really Switzerland vs. the US; it is *unbundled markets vs. integrated markets*, and every unbundled market has ended up with cheap symmetric gigabit at retail. Nothing on the current US policy roadmap changes which side of that line we are on.