Summary

  • ARIN registered AS400900 on 2023-10-02 to SK Hynix Memory Solutions America Inc. at 3103 North 1st Street, San Jose, under the AS name SKHMSA-WEB, with one registration comment: "SK Hynix Memory Solution America site."
  • The registrant's ARIN network list contains one IPv4 /24 — 8.5.130.0/24, Net Type Reassigned, registered 2024-04-16 inside a block ARIN registered in 1992 — and returns no IPv6 netblock.
  • Public routing data shows one announced prefix, one origin, an empty IRR route-object list and no validating ROA; the group's Korean ASN, assigned in 2005, originates sixteen /24s and no IPv6.

The record is short enough to read in full, so read it in full. ARIN's autnum object declares the number active, its start and end both 400900, its AS name SKHMSA-WEB, and its registration and last-changed events on the same timestamp: 2023-10-02T15:25:44-04:00. The registrant is not a person but an organisation handle, SHMSA-1, whose vCard name is SK Hynix Memory Solutions America Inc. and whose address label reads 3103 North 1st Street, San Jose, CA 95134, United States. That organisation object has its own dates: registered 2023-09-06, last changed 2023-09-19.

Then the remark. Not a description of a network, not a peering policy, not a technical contact note: a purpose. Somebody at a memory-chip subsidiary sat in front of a registry form and typed what the number was for. A site.

The order the dates arrived in

The contact record arrives first. The single point-of-contact handle attached to this object was registered 2023-07-14 and last updated 2024-10-01, and its registered company is SK Hynix Memory Solutions America at the same San Jose street address (POC record). Fifty-four days later, by my subtraction, the organisation object appears. Twenty-six days after that, the autonomous system number. Then a pause: the address block is registered 2024-04-16, one hundred and ninety-seven days after the ASN, as a reassignment inside a parent block that ARIN registered on 1992-12-01 (organisation network list).

Six days after the block date, the prefix shows up in routing data with an origin that is not this ASN. RIPEstat's routing-status record for 8.5.130.0/24 puts the first sighting at 2024-04-22T16:00:00 with origin AS7018, and the current origin at 2026-11-20T16:00:00 as AS400900 (routing status). No document I obtained says what the earlier origin was doing with the address, and I am not going to guess at it. What the record does establish is that the sequence runs contact, organisation, number, address, announcement — which is the sequence of someone assembling a small network identity deliberately rather than inheriting one.

What a /24 is worth inside that parent

Here is the arithmetic, and it is mine. The parent object NET-8-0-0-0-1 prints NetRange 8.0.0.0 – 8.8.3.255 with CIDR 8.0.0.0/13 and 8.8.0.0/22, Net Name LVLT-ORG-8-8, Net Type Direct Allocation, held by Level 3 Parent, LLC (LPL-141), registration date 1992-12-01 (parent block). A /13 is 524,288 addresses; a /22 is 1,024; together 525,312, which is exactly what the printed range spans. Divide 524,288 by 256 and you get 2,048, so the subsidiary's /24 is one two-thousand-and-forty-eighth of one component of a block registered before the web existed.

That is what the Net Type field is telling you. It does not say Allocated, and it does not say Direct Assignment. It says Reassigned: a carve-out inside somebody else's direct allocation. The number AS400900 belongs to the San Jose company; the addresses are held under a title derived from a company in another state that registered the space in 1992. Two different kinds of custody, on one record.

One origin, and the fields that are empty

RIPEstat's overview returns the holder string "SKHMSA-WEB - SK Hynix Memory Solutions America Inc.", places the number in IANA's 400285–401308 block described as assigned by ARIN, and reports announced as true at 2026-11-20T16:00:00 (AS overview). The announced-prefix endpoint returns exactly one entry — 8.5.130.0/24, with a timeline from 2026-11-06T16:00:00 to 2026-11-20T16:00:00 — and no IPv6 prefix at all (announced prefixes). Routing status confirms one origin, AS400900, with an empty route_objects array, IPv4 visibility of 321 of 321 RIS peers, less-specific announcements 8.0.0.0/9 and 8.0.0.0/12 both originated by AS3356, and no more-specifics (routing status).

Now the part that matters more than the count. RPKI validation for this origin and this prefix returns status unknown, an empty validating_roas array, and the validator name routinator. There is no Route Origin Authorization saying who may originate 8.5.130.0/24. The IRR route_objects array is empty too, so no routing registry object links the prefix to AS400900 either. Both attestations that would let a stranger verify the announcement are absent, and the announcement works anyway, because BGP does not ask permission; it asks whether the neighbours accepted the path.

That is not a scandal. It is the ordinary state of a small edge, and a great many prefixes on the internet sit in exactly that condition. But it settles the question of what carries the prefix's authority. Not cryptography, and not an IRR object: the registry record, and whatever filters AS3356 applies when its customer hands it a route.

Six roles on one handle, and ARIN's own sentence

Everything the registry needs a human for, on this object, resolves to one handle. The POC record lists functions Organization Admin, DNS, Abuse, Tech, NOC and Routing, all against SHMSA-1 (POC record). There is no second contact to route an abuse report through, no separate routing engineer, no distinct DNS administrator. For a subsidiary network with one prefix that is proportionate rather than alarming — and it is the complete list of contact structure the record offers.

That same record carries ARIN's own note, and it is worth quoting rather than paraphrasing: "ARIN has attempted to validate the data for this POC, but has received no response from the POC since 2025-10-01." That is the registry describing its own validation attempt and dating it. It is a statement about a form and a reply, and I will not inflate it into a statement about whether anyone answers a telephone in San Jose.

The sibling in another registry

Now the comparison the record invites. In Korea, KRNIC's AS-number assignment list prints AS23581 against the registrant name 에스케이하이닉스(주) with assignment date 20051102, on a list stated as current to 2025.06.07. The English AS name and holder appear elsewhere: RIPEstat's search index describes AS23581 as "HYNIX-AS-KR-KR - Hynix Semiconductor Inc."

What that number originates is a different magnitude. Hurricane Electric's AS23581 page reports 16 IPv4 prefixes originated, 15 currently announced, 0 IPv6 originated and 0 IPv6 announced, and its visible table lists sixteen /24s running from 166.125.0.0/24 to 166.125.254.0/24. Sixteen times 256 is 4,096 addresses, which is sixteen times the American subsidiary's 256 — my multiplication, on both figures as printed. The page distinguishes what it originated from what it currently announces and does not name the withdrawn prefix, so both counts stand as printed and the difference stays unidentified.

Set the two dated records side by side and the shape is plain. Korea: a number assigned in November 2005, sixteen prefixes, thousands of addresses. America: a number registered in October 2023, one prefix, 256 addresses, a borrowed title, and a remarked purpose that reads "site." The gap between the two assignments is seventeen years and eleven months by my subtraction.

Three names in a routing table

A name search on the routing toolkit prints four rows worth reading together: AS400900 described as SK Hynix Memory Solutions America Inc.; AS23581 described as Hynix Semiconductor Inc.; the route 8.5.130.0/24 attributed to the memory-solutions name; and the routes 65.167.196.0/22 and 205.132.240.0/21 attributed to S.K. Hynix America Inc. Note what that page is: a routing-table search. It prints no ARIN handles and no registration dates, so those two larger prefixes are associated with a name in a routing table, not established by that page as ARIN registrations. I did not obtain a primary registry record for them, and the article does not claim one.

What the company's own side says is separate and narrower. SK hynix's worldwide-locations page lists two entries at adjacent building numbers: SK hynix America Inc. at 3101 North 1st Street, San Jose, and SK hynix memory solutions America Inc. at 3103 North 1st Street, San Jose, CA 95134 USA. The second matches the ARIN registrant address exactly, street number and all. Two subsidiaries, two doors on the same street, and the network identity registered to the one whose door number is 3103.

So the group is not one network with one number. It is at least three names in at least two registries, and the newest and smallest of them belongs to the memory-solutions subsidiary — which is precisely why the remark on it is worth quoting.

What the record attests, and what it cannot

A registry object attests custody of an identifier. It does not attest what the identifier carries, how much traffic crosses it, how many people maintain it, or whether the site named in the comment is served from the addresses in the block. Every sentence above about the record is a sentence about documents; none of them is a measurement of a network.

Here is what the documents let me say without reaching. The American subsidiary holds an ASN whose only stated purpose is a site. The address that goes with it is one /24 out of a block registered in 1992 by another company. The announcement exists, is seen by every RIS peer in the sample, and is attested by neither an ROA nor an IRR route object. The ARIN network list for this organisation shows one block and, in that response, no IPv6. And the group's older Korean number, assigned in 2005, originates sixteen /24s and no IPv6.

Here is what they do not let me say. They do not tell me the /24 is used only for a website, because no document read states what any address in it serves. They do not tell me the site is hosted there rather than somewhere else; hosting a site and registering an address for it are two different decisions made by two different people, often in two different years. They do not tell me the Korean sixteen prefixes carry the group's production traffic, only that they are originated. The purpose field is the registrant's own description, and the register records the description, not a verified fact about the wire.

A null result worth printing

One more document belongs in the record because a reader will meet it. APNIC's web WHOIS answers the string HYNIX with "%ERROR:101: no entries found" and "% No entries found in source APNIC,JPNIC,KRNIC,TWNIC,IDNIC,IRINN-GRS" — including, on its own list, the Korean source. Yet KRNIC's assignment list, RIPEstat's index and Hurricane Electric all print AS23581. A query surface returning nothing is a fact about the query surface.

It is not evidence that the Korean record is missing, and anybody who types HYNIX into the obvious window and concludes the group has no Korean number has been misled by a search box rather than by a registry.

Reading limits

The autnum object was read through a redirector that self-links to rdap.arin.net; a direct fetch of ARIN's RDAP and REST endpoints returned no readable content in my first pass, so every registry field above rests on that redirector's rendering of the same object. Absence of IPv6 is scoped to three responses — the organisation's ARIN network list, the RIPEstat announced-prefix window ending 2026-11-20T16:00:00, and Hurricane Electric's counters for AS23581 — and none of them is a census of the group's holdings.

RIPEstat excludes routes seen by fewer than ten full-feed RIS peers, and its announced-prefix window covers two weeks, not history. Hurricane Electric prints sixteen originated and fifteen announced without naming the withdrawn one. And no document I obtained identifies AS7018 or explains the first recorded origin of 8.5.130.0/24 in April 2024.

The purpose field

Registries are built to record who holds what. Almost incidentally, one field on one object records why. SK hynix's American subsidiary filled that field in, and what it wrote describes a website. Around that sentence sits the whole architecture of a small network identity: a number registered in October 2023, a borrowed /24 registered six months later, one origin, no ROA, no route object, one contact handle doing six jobs, and a registry note about an unanswered validation attempt.

The larger network is elsewhere — in a Korean assignment list from 2005, sixteen prefixes wide, described under a name the routing indexes still print as Hynix Semiconductor Inc. Both records are public. Both are current. And the one attached to the American memory-solutions subsidiary says, in the only free-text field it has, that the number is for a site.

That is the answer to the question I set out to ask, and it is not a small one. A group capable of running a sixteen-prefix edge chose, for its San Jose subsidiary, to register one number and one /24 and to write down what they were for. The registry did its job. The comment did the rest. Anyone who wants to know what the American ASN carries will have to read a document this register does not hold.