Summary

  • In APNIC’s matched 29 October 2026 cumulative files, October-dated IPv4 rows total 53,760 addresses in transfer-apnic and 1,156,864 in transfer-all-apnic. The first file’s October rows exactly match the second file’s Unused rows.
  • The broader file contains one 16 October M&A row for 120.160.0.0/12, or 1,048,576 addresses, from PT. INDOSAT Tbk to Indosat Singapore Pte. Ltd. The row alone represents 90.64% of the broader file’s October-dated IPv4 address volume.

A registry can change the apparent size of a market without moving a single address. It only has to make two public files with different scopes and give both perfectly plausible names. APNIC does exactly that. Its 29 October transfer-apnic file contains 40 rows dated in October. Its same-day transfer-all-apnic file contains 88. The smaller file has not lost 48 rows through error. It was built to count a narrower class of movement.

APNIC’s older format specification describes a ten-column transfer log. The later specification adds transfer_type and covers Unused, M&A and HRT. Filter both cumulative files by transfer_date beginning 202610: the old feed’s 40 rows match the all-feed’s 40 Unused rows exactly once the extra type column is removed. The all-feed adds 45 M&A rows and three HRT rows. Those are rows, not 48 independently negotiated deals.

The address arithmetic is less polite than the filenames. The narrow file has 37 October-dated IPv4 rows covering 53,760 addresses. The all-transfer file has 61 covering 1,156,864: 53,760 Unused, 1,101,568 M&A and 1,536 HRT. The 61 IPv4 prefixes do not overlap in this snapshot, so the address totals can be added without counting a block twice. Open only the narrow feed and you see 4.65% of the broader file’s IPv4 volume. The missing 95.35% is not an undisclosed transfer. It is in APNIC’s other public file.

One row supplies the punchline. Dated 16 October and classified M&A, 120.160.0.0/12 moves in the log from PT. INDOSAT Tbk, with country code ID, to Indosat Singapore Pte. Ltd., with SG. Both RIR fields say APNIC. A /12 has 1,048,576 IPv4 addresses. That single row accounts for 90.64% of the broader file’s October-dated IPv4 volume and does not appear in the narrow one. The record gives no price, legal consideration or actual network deployment. Those facts would require transaction documents and operational evidence absent from the public log.

There is a second trap in the calendar. APNIC’s files are cumulative daily snapshots, and the latest October transfer date visible in these 29 October captures is 27 October. These are counts for October-dated rows visible on 29 October, not a completed month-end report. A later daily file can add or revise the visible cohort. The distinction matters when a reader turns a provisional sum into a chart of “October trading.”

APNIC’s publication design makes the narrow file easy to discover and easy to overread. Its accuracy does not rescue the implied denominator. If a board paper, market report or policy argument uses “APNIC transfer volume,” the writer has to name the file, the snapshot date, the resource type and the transfer classes. Otherwise a file-selection choice can make 1,103,104 addresses disappear from the sentence while remaining in the registry’s own records. The ledger is public. The expensive part is learning which drawer to open.