ProdIPData/ip-locations

ProdIPData — Monthly IPv4 IP Geolocation Dataset

A free, monthly dataset of the routed IPv4 address space at /24 granularity — one row per /24 prefix — covering geolocation, network/ASN, ownership, registry and risk attributes. Maintained by ProdIPData and mirrored from https://prodipdata.com. Licensed CC BY 4.0 (free to use, including commercially, with attribution).

What's in each row (/24 prefix)

Group	Fields
Geolocation	CountryCode, ContinentCode, IsInEU, RegionName (ISO 3166-2), CityName, Latitude, Longitude, AccuracyRadius, TimeZone
Network / ASN	ASNID, ASNName, ASNOrg, ASNDomain, ASNType (ISP / hosting / business / education / government)
Ownership (WHOIS)	Company, CompanyCountry, WhoisNetName, AbuseContact
Registry (RIR)	RIR (ARIN/RIPE/APNIC/LACNIC/AFRINIC), RIRAllocationDate, RIRStatus
Risk signals	HasC2, C2Malwares, IsTor, Anonymization, bogon flags

Key column: IP24Prefix (e.g. 8.8.8.0/24).

Coverage

• 14M+ routed /24 prefixes · 240+ countries · ~3.7B IPv4 addresses represented
• Coordinates spatially validated against public-domain Natural Earth boundaries (~99.97% country-level match)
• Updated monthly; each release is tagged by month (YYYY-MM)

Load it

# pandas — read straight from the Hub
import pandas as pd
df = pd.read_parquet(
    "hf://datasets/ProdIPData/ip-locations/prefix_attributes_2026-07_iso-ALL.parquet"
)


# duckdb (query without loading it all into memory)
import duckdb
duckdb.sql("""
  SELECT CountryCode, COUNT(*) AS prefixes
  FROM 'prefix_attributes_2026-07_iso-ALL.parquet'
  GROUP BY 1 ORDER BY 2 DESC LIMIT 10
""").show()


# Hugging Face datasets
from datasets import load_dataset
ds = load_dataset("ProdIPData/ip-locations")

Formats

Also published as CSV and MMDB (MaxMind DB compatible format) at https://prodipdata.com/downloads.html. The site additionally offers an in-browser Prefix Lookup (resolves any IP/•/24 client-side, nothing sent to a server) and density / region maps.

Limitations

• Geolocation is derived from upstream sources including MaxMind and the RIRs, merged monthly — it is a snapshot, not real-time, and only as accurate as those sources.
• Low-confidence blocks fall back to country-level centroids. For example 8.8.8.8 resolves to the US geographic centre, not Mountain View. The AccuracyRadius field (in km) now makes this precision explicit, so country-level fallbacks are visible rather than implied.
• /24 granularity: attributes describe the block, not individual hosts.

Attribution

Please credit ProdIPData (prodipdata.com) under CC BY 4.0. Country/region boundaries © Natural Earth (public domain).

MaxMind and GeoIP are registered trademarks of MaxMind, Inc. ProdIPData is not affiliated with or endorsed by MaxMind.

• Website: https://prodipdata.com
• Downloads: https://prodipdata.com/downloads.html
• Methodology: https://prodipdata.com/methodology.html