Do the audit. Nineteen subscriptions is the median I keep finding when I ask technically literate people to count: notes, tasks, password vault, file sync, photo backup, VPN, email hosting, calendar, website builder, analytics, uptime monitor, link shortener, form backend, newsletter, media hosting, code forge, CI minutes, database hosting, Redis hosting, log retention. At a blended $21 a month each, that is $400 a month — $4,800 a year, every year, forever — to rent back access to your own data, one expiring session at a time.

Now price the alternative. A quiet desk node — mini-PC class, 32GB RAM, 2TB NVMe — costs about $600 once. Electricity to run it 24/7 runs roughly $8 a month. A domain is $15 a year. An edge tunnel for zero-egress remote access is free at personal scale. First-year total: about $710. Every year after: about $110. The break-even against the subscription pile arrives in month two, and from then on the node pays you roughly $4,700 a year to exist.

The objection used to be capability: surely nineteen venture-funded products do more than one box. That was true in 2015. It is not true now, because the primitives collapsed. SQLite handles millions of rows without breaking stride. Postgres runs forty sites on one instance without complaint. Static files, a reverse proxy, and a process manager cover hosting for nearly everything a person or small team actually runs. The SaaS layer above those primitives is now mostly billing infrastructure wearing a product costume.

The deeper objection is custody, and it cuts the other way. Every subscription is a counterparty that can raise prices, sunset features, suffer breaches, or lock you out over a billing dispute with your own work inside. Deletion is a request, not an action; export is a feature, not a right. On your own node, queries answer in milliseconds with no meter running, backups are files you can hold, and deletion actually deletes. Custody is not a vibe. It is a property of where the bytes physically sit.

What about reliability? Fair question, honestly answered: a single desk node has worse theoretical uptime than a hyperscaler region. But measure what actually fails in practice. The hyperscaler region stays up while your account gets flagged, your card expires, the vendor sunsets your tier, or the status page lies politely for six hours. A node with automated offsite backups fails rarely and recovers in minutes, with no support ticket and no one to blame but yourself — which, operationally, is a feature: the mean time to fix yourself is shorter than the mean time to convince a vendor.

The tunnel deserves its own paragraph because it is the piece most people have not priced. Inbound connectivity used to mean a static IP, port forwarding, dynamic DNS, and a prayer to your ISP. A modern edge tunnel reverses the connection — your node dials out, the edge answers — which means no open ports, no ISP cooperation, and TLS termination handled for you. Remote access that used to be a weekend project is now a config file. This single primitive is what makes the desk node a real server instead of a hobby.

None of this argues that cloud computing is dead. It argues that the rental model overshot for personal and small-team scale, and the pendulum is swinging back toward ownership the way it always does when a technology matures: mainframes to PCs, theaters to streaming boxes, now cloud dashboards to desk nodes. The pattern is the same every time — capability commoditizes, the interface simplifies, and the box moves closer to the human.

Start with one migration, not nineteen. Move the thing whose subscription annoys you most — usually notes or file sync — onto the node, prove the backup story by restoring once, and let the savings fund the rest. Within a year you will look at a $400 monthly subscription pile the way you now look at long-distance phone bills: a tax on an era that ended, which you simply stopped paying.