Research wiki
concepts · updated 2026-10-26 contested

Local hosting is not totalizing, and can be cheaper per user or community

Claim (restated 2026-10-26 per James: the point is that local hosting is not totalizing and can be cheaper per user or community, not that it is more efficient in aggregate): municipal, local or community hosting is not totalizing, and it can realistically be cheaper for each user and/or community than hosting on totalizing mega-cloud providers. The claim is not that it is more efficient in aggregate; scale economics can win on aggregate efficiency without refuting it. Still within his earlier correction: the route is existing and surplus hardware, not buying new (2026-10-11).

What tests this version: cost per user or per community for a realistic small operator against what the same users pay mega-cloud (in money, and in data, attention and lock-in); and what makes a provider totalizing (lock-in, radical monopoly in Illich's sense, see philosophy-of-technology) versus not. Fleet-level energy or utilisation figures bear on aggregate efficiency, so on their own they do not test it.

~~Was (2026-10-11 to 2026-10-26):~~ Claim (restated 2026-10-26 per James's correction of 2026-10-11: the route is existing and surplus hardware, not buying new; evidence about new-server prices does not test this version): the cost advantages of AWS-style mass-scale hosting have largely disappeared. "Maturity" now means deploying a stack that exists only to support scale. A municipal, local or federated model regains simplicity and saves cost while keeping two decades of cloud-driven virtualisation advances. This is genuinely affordable now, not after a theoretical bill comes due. The wiki should actively collect evidence for and against: where scale still wins (egress, GPUs, compliance, on-call), and real price series.

Contested: collect evidence for and against on this page; do not treat as established.

Provenance: James, in conversation with Claude, 30 Sept 2026 (unverified). Disposition: contested.

Related: convivial-computation, municipal-inference

Evidence for (read 2026-10-11, search-result text only; single-source each, stated by source)

Counter-evidence / cuts against (stated, weak)

Disposition unchanged (contested): evidence is mixed and thin. Trail: trail-2026-10-11.

Surplus-hardware evidence (added 2026-10-14, trail-2026-10-14)

For: a refurbished server measured as efficient as a new one of the same model (SERT 13.7 vs 13.8), and with the slowdown in per-generation gains (~16% energy from the latest step), refurbished kit at ~$1,200 against ~$2,800 new had the fastest payback when replacing servers older than ~6 years. See bashroush-refurbished-servers-2022 (read in full). Against, or a limit: servers older than five years were 40% of a European fleet, used 66% of IT energy and did 7% of the compute (Uptime/EURECA 2020, search text). So surplus works when it is a few years old, not when it is very old; energy price sets the cut-off (the study assumed $0.10/kWh). Who runs on surplus: Hetzner sells refurbished dedicated servers through a standing auction, now marketed as "the smart way to start self-hosting... Dare to own your stack" (164 offers on 2026-10-14, mostly i7-6700/7700 and Ryzen 5 3600 boxes; https://www.hetzner.com/sb/, read in the browser; prices did not render). Disposition unchanged (contested): this supports the surplus route within an age window and does not test the claim against scale economics.

Historical case: Community Memory (added 2026-10-23, trail-2026-10-23)

Who ran on surplus, 1972-75: Resource One and Community Memory in the Bay Area ran a free community computer on a donated off-lease SDS-940. It served local groups and the first public terminals, but the first version closed because the machine was "unreliable and irreplaceable" and could not be replicated elsewhere. Single participant source plus CHM material, stated. See community-memory. Disposition unchanged (contested): a historical precedent for the route and for its limit (maintainability and replicability, not price), not a test of present costs.

Replicated surplus labs: MetaReciclagem and Access Space (added 2026-10-26, trail-2026-10-26)

For: two networks show the surplus route solving the problem that killed Community Memory v1. Both took commodity office PCs, not one exotic machine, so any lab could be copied. Access Space (Sheffield, 2000 onward) has run on donated computers and Linux for a quarter-century; its founder's argument is that market value collapses around year four while use value barely falls, so the lab gets "regular free upgrades" with each generation thrown away (access-space-sheffield, search text). MetaReciclagem (Brazil, 2002-2012) turned it into a method of self-managed "spores" in many cities, gave the rebuilt machines to the people who rebuilt them, and fed the method into a federal programme (metareciclagem, read in full). Against, or a limit: a 2012 Brazilian government study says such projects "often end up accumulating considerable volumes of unusable material" and cannot deal with waste at scale. Both cases are about end-user desktops for access and learning, not servers for hosting, so they bear on who can run on surplus and how, not on the cost comparison with mass hosting. Disposition unchanged (contested): new evidence that the route is replicable when the hardware is commodity, which narrows Community Memory's limit to exotic kit; still no test of hosting costs.

Linked from (17)