Number of speakers per language in the world based on Wikidata (including dupes due to measuring speakers at different times / from different sources). Note that 6915 languages have iso 639-3 codes in wikidata, but only 1614 languages have data for number of speakers (presumably the languages with no data are at the lower end of the list)

speakersPerLanguage.sparql

SELECT DISTINCT ?item ?numberOfSpeakers ?speakersTime ?iso639_1 ?iso639_3 ?itemLabel ?itemLabel_en
WHERE {
  ?item wdt:P31 wd:Q1288568;      # Any instance of a modern (current use) language
      wdt:P220 ?iso639_3;         # that has an ISO 639-3 code

  # Require number of speakers, but as a statement so we can get the "timestamp" of the property:
  ?item p:P1098 ?numberOfSpeakersStatement .
  ?numberOfSpeakersStatement ps:P1098 ?numberOfSpeakers.
  optional {
    ?numberOfSpeakersStatement pq:P585 ?speakersTime.
    FILTER(datatype(?speakersTime) = xsd:dateTime). # exclude unknown values
  }

  # Get the native language label, if there is one, trying ISO 639-1 and -3:
  # (Can we match language of labels using wikidata identifiers?)
  optional {
    ?item wdt:P218 ?iso639_1.
    ?item rdfs:label ?itemLabel_1 . FILTER(lang(?itemLabel_1)=?iso639_1)
  }
  optional {
    ?item rdfs:label ?itemLabel_3 . FILTER(lang(?itemLabel_3)=?iso639_3)
  }
  bind(coalesce(?itemLabel_3, ?itemLabel_1) as ?itemLabel)

  # Get the label in English, if there is one:
  optional {
    ?item rdfs:label ?itemLabel_en . FILTER(lang(?itemLabel_en)='en')
  }

}
ORDER BY DESC(?numberOfSpeakers) DESC(?speakersTime)

speakersPerLanguage.tsv

speakersPerLanguageLatestNumber.tsv