Problem thread

Who pays for public data infrastructure?

Who pays to keep shared identifiers, query services and knowledge bases running after the launch money is spent?

issue
sustainability
also governance, scale & performance
status
open
steps
9, 1995–2025
threads
all problem threads

Shared identifiers, query services and knowledge bases are cheap to launch and expensive to keep running. PURL ran at OCLC for twenty years before moving to the Internet Archive; Freebase closed when Google withdrew; many public SPARQL endpoints were often down, which prompted cheaper interfaces. Wikidata, run by a donor-funded foundation, is the main survivor, and even its query service had to split its graph to stay up.

Lineage

  1. 1995
    PURL

    OCLC Research introduces a resolver so that a name can outlive the server it points to, and hosts purl.org for twenty years. In 2016 it hands the service to the Internet Archive for the sake of its future sustainability. [src]

  2. Dec 2006
    Understanding Knowledge as a Commons: From Theory to Practice

    Hess and Ostrom apply research on shared natural resources to knowledge: a commons lasts only if a community sets rules that keep it from being depleted or enclosed. [src]

  3. Mar 2007
    Freebase

    A start-up, Metaweb, runs a free, collaboratively edited knowledge base. When Google buys Metaweb in 2010 it says it plans to maintain Freebase as a free and open database for the world. [src]

  4. 29 Oct 2012
    Wikidata

    A knowledge base run as a Wikimedia project rather than a business. Its initial development is paid for by donations from the Allen Institute for AI, the Gordon and Betty Moore Foundation and Google. [src]

  5. Oct 2013
    SPARQL Web-Querying Infrastructure: Ready for Action?

    Monitoring of hundreds of public SPARQL endpoints finds many of them often unavailable, slow, or limited in what they will answer. [src]

  6. Mar 2016
    Triple Pattern Fragments: A Low-cost Knowledge Graph Interface for the Web

    Since full query endpoints are expensive to host at high availability, servers should answer only simple triple patterns and let clients do the rest of the work. Queries are slower but servers stay up. [src1][src2]

  7. 11 Apr 2016
    From Freebase to Wikidata: The Great Migration

    Google announced in 2014 that it would shut Freebase down; it went read-only on 31 March 2015 holding more than 3 billion facts. The paper reports how much of it a community project could take over. [src1][src2]

  8. 2016
    poses writing ★ seminal
    Roads and Bridges: The Unseen Labor Behind Our Digital Infrastructure

    Names the general pattern for open-source software: infrastructure everyone depends on, kept running by a few unpaid or underpaid people, with no one responsible for funding it. [src]

  9. 9 May 2025
    measures project
    Wikidata Query Service

    Wikidata's free public query service splits scholarly articles into a separate graph: its Blazegraph backend can no longer keep up with more than 16 billion triples growing by about a billion a year. A move to QLever follows. [src1][src2][src3]

Recent signals

Dated events that show the problem moving, newest last.

  1. 15 Jan 2026

    Five more AI and technology companies pay Wikimedia Enterprise for access to Wikipedia's content.[src]

  2. 12 Jun 2026

    NLnet pauses new NGI Zero calls while EU open-internet funding moves to a new Open Internet Stack programme.[src]

  3. 17 Oct 2026

    The Wikidata Query Service is migrating from the unmaintained Blazegraph to the QLever engine. Wikidata Query Service[src]

Open now

Under each question, the answers people have put forward or built. The catalogue gives no odds on them.

  • Who pays to keep a free public query service running as the data behind it keeps growing? [src1][src2]
    Wikidata's query service split its graph in 2025 and is moving to a new backend; the old Blazegraph endpoints may degrade from February 2027.
    • Split the data and change the engine underneath. Wikidata Query Service
      Wikidata split its graph in 2025 and is moving its query service to QLever.[src]
  • How much of a company's knowledge base survives when the company withdraws? [src]
    When Freebase closed, its data had to be mapped into Wikidata's model and reviewed by volunteers: about 360 of Freebase's roughly 37,700 properties were mapped early on.
  • Who funds the upkeep of open data infrastructure, rather than its launch? [src1][src2]
    Grants usually pay for new work. A few public funds now pay for maintenance of open digital infrastructure.
    • Public funds pay for maintenance.
      Funds such as the Sovereign Tech Agency and NLnet pay for upkeep of open infrastructure.[src1][src2]
    • The largest users pay.
      Wikimedia Enterprise charges companies that use Wikipedia's content at scale.[src]

Forces

Pressures that push this problem one way or another, as their sources argue them. All forces

Decision points

Dated moments when a decision or a deadline can change the outcome.

  1. 3 Nov 2026
    NLnet funding deadline funding

    Next deadline for grant proposals to NLnet's open funds for open internet software.[src]

  2. Feb 2027
    Old Wikidata query endpoints may degrade service

    From February 2027 the Blazegraph endpoints stay up but may degrade; users need to move to the new endpoints.[src]

How to contribute

  • Wikidata Query Service backend update forum · Wikidata Query Service

    Hub for moving the query service from Blazegraph to QLever: early access to the new endpoints, office hours, use-case collection and a Phabricator board. [src]

    checked 21 Oct 2026
  • Sovereign Tech Fund funding

    German public fund for open digital base technologies, including maintenance work. Projects apply through its portal; costs must exceed €50,000. [src]

    checked 21 Oct 2026
  • NLnet Foundation funding

    Dutch foundation that grants money to open internet technology. Its open funds take proposals at regular deadlines; the next is 3 November 2026. [src]

    checked 21 Oct 2026