Summary

  • The Seattle Internet Exchange describes itself as a Washington State nonprofit corporation with IRS 501(c)(6) status, and states that its articles of incorporation, bylaws and minutes of official meetings are available on a documents page; its board and officers are on a separate roster page.
  • The one mechanism the retained record states precisely is commercial: the applicable port fee is the rate in force at the earlier of letter-of-authorization issuance or port configuration, charged once by default with a monthly option, while a $100 extension charge sits alongside further extension fees set by third-party operators.
  • No retained operator sentence in this record assigns the authority to set or amend that schedule to members, to a board, or to staff — so a reader deciding whether to buy a port there is looking at a published price, not a published procedure for changing it.

Few pieces of internet infrastructure are as easy to describe badly as an exchange point. The physical description is unglamorous: a switching fabric inside one or more carrier buildings, ports sold in speed classes, cross-connects run by landlords, and a set of routing sessions that carriers and content networks choose to establish. The institutional description is harder still, because the interesting question is rarely "does the exchange exist" but "who can change the terms on which it exists".

That second question is what this article tests. Not whether the Seattle Internet Exchange is a serious facility — the record gives no reason to doubt the fabric or its operating continuity — but whether the public record establishes who holds the pen on the terms of interconnection. The distinction matters to anyone sizing a port, renewing a contract, or deciding whether to build a PoP around a particular exchange's pricing.

A note on method before the evidence. Every operator passage below was captured from the exchange's own public pages on 21 October 2026 between 08:36 and 09:04 UTC. The captures came through a text-extraction path rather than by rendering the pages in a browser, and where a passage is a summary of page content rather than a continuous quotation, I say so at the point of use. That is not a formality. The difference between what a page says and what a summariser reports that it says is exactly the difference this article is about, one level up.

The fabric, as the exchange describes it

The front page gives the shortest version of the institution's self-description. "The SIX is a non-profit Internet Exchange Point in Seattle, Washington," it reads, offering "reliable and low-cost interconnection between member networks in the Northwest United States and beyond" (seattleix.net). That sentence does three things at once: it names a legal character (non-profit), a function (interconnection), and a constituency (member networks). Only the first two are developed anywhere else in the retained record.

A summary of the same page, which I am labelling as a summary rather than a quotation, reports that networks connect at speeds from 1 Gbps to Nx400 Gbps, that the fabric spans the Westin Building, KOMO Plaza and Sabey Intergate, that a single Arista 36-port 100G QSFP28 line card of the 7500R2-36CQ type is installed at the Westin Building, and that a 400GbE one-time port fee of $15,000 was introduced on the join page with availability at the Westin Building and KOMO Plaza.

These are the details a network engineer would want, and they are the details I trust least in this record, because they arrive as compressed paraphrase rather than as sentences read off a page. I use them here for texture and for one structural point only: the exchange presents itself as a multi-facility fabric selling a port ladder that runs to 400 gigabits per second, not as a single-room switch.

That structural point is not trivial, and it is the kind of thing my beat rewards noticing. A port at 400GbE is not a commodity a small network buys by accident. It implies optics, line cards, cross-connect capacity and a physical path between at least two carrier buildings.

The reported $15,000 one-time figure, if accurate, is small relative to the optics and colocation needed to use it — which tells you something about how this exchange positions itself: the port is cheap, the surrounding real estate is not, and the exchange's own join page is explicit that it takes no responsibility for the third-party extensions, colocation providers or consultants a participant will deal with.

The commercial rule

The join page is where the record stops being descriptive and becomes operative. As summarised from that page: port fees "are determined by the fees current on the date of LOA issuance or port configuration by SIX admins, whichever is earliest". The same page reports that the exchange does not charge recurring fees but offers a monthly recurring payment option, that a $100 one-time charge applies when connecting via an extension, and that additional port fees in an extension are determined by the extension operators (seattleix.net/join).

The page's own verbatim excerpt is narrower and safer: "The SIX offers 1GbE, 10GbE, 40GbE, 100GbE, and 400GbE port options. One-time port fees are detailed on our join page. The SIX does not charge recurring fees, but they are offered as a payment option." Read together, the excerpt and the summary describe a tariff with a trigger rather than a subscription. The price is not set at signature, not at invoice, and not at the start of a calendar period. It is set at the earlier of two administrative events: the issuance of a letter of authorization, or the configuration of the port by exchange administrators.

That is a precise piece of drafting. It also explains why the join page bothers to say "whichever is earliest". The two events can occur in either order, and the exchange closes the gap between them. If a participant's LOA is issued in March and its port is configured in May after a rate change, the March rate applies. If the port is configured first in March and the paperwork catches up in May, the March rate still applies. The trigger runs off the participant's own administrative progress, not off the exchange's billing cycle.

Three further features complete the commercial picture. First, the default is a one-time charge rather than a recurring one, which turns the port into a capital item with a possible financing choice attached: the monthly option exists, but as a payment mechanism, not as a different product. Second, the $100 extension charge is explicitly a charge for the extension itself, with the real costs after that point belonging to whichever operator runs the extension.

Third, and most consequential for anyone modelling total cost, the exchange states plainly that it is not responsible for the quality, liability or outcome of engagement with listed colocation providers or consultants. The exchange sells a port. It does not sell the path to the port.

What the trigger does to a participant

Here I move from what the page says to what the wording implies, and I mark the move because it is mine, not the exchange's. A rule that fixes price at authorisation creates cohorts. Two networks with identical ports can pay different rates for the life of those ports, not because either negotiated better, but because one authorised in a given month and the other in a later one. The tariff is published, the trigger is published, and the resulting price distribution is still not uniform.

For a buyer, the practical consequence is a timing decision with a real option value. If you believe a rate change is coming, the cheapest action is to secure the trigger — get the LOA out, get the port configured — before the schedule moves. The exchange's wording gives you that option and makes no attempt to close it. If you believe rates are stable or falling, waiting costs you nothing. What the wording does not tell you is who decides that a schedule should move, and on what notice, and with what consultation. That is the hole in the middle of the tariff.

For the exchange, the same rule produces a revenue profile worth naming carefully. Because the default is one-time, income tracks new authorisations rather than installed base. A port configured in 2019 contributes nothing recurring in 2026 unless its owner elected the monthly option. That is not a criticism; many exchanges have chosen exactly this structure to keep participation cheap and friction low. It is simply a fact about incentives: an institution whose revenue follows new connections has a structural interest in growth, and a structural indifference to the renewal cycle of ports already in service.

Again, that is my inference from a documented charging rule, not a statement the exchange makes.

There is one boundary I will not cross. The trigger freezes the exchange's own port charge for a triggered port. It does not freeze total delivered cost. The extension operator's fees are outside the exchange's tariff by the exchange's own words, and the complete join-page tariff was not read in full in this record, so any question about what happens to a monthly-electing participant after a rate change is unresolved here. A reader who needs that answer needs the full tariff and its effective dates, not this article.

The legal form, as stated

The about page supplies the institutional chronology. "The Seattle Internet Exchange (SIX) was originally established in April 1997," the retained passage begins, and continues through incorporation as a non-profit in 2001 and receipt of IRS section 501(c)(6) income tax exempt status in 2008 (seattleix.net/about). A second capture of the same page renders the milestones in timeline form and gives the months: "2001-05: Incorporated as a non-profit" and "2008-09: Received IRS section 501(c)(6) income tax exempt status."

Section 501(c)(6) is a specific instrument, and its specificity is worth a moment. It is the Internal Revenue Code provision for business leagues, chambers of commerce, real estate boards and boards of trade — organisations whose purpose is to improve business conditions in a line of commerce rather than to earn profit for themselves. For an exchange point, that characterisation is coherent: the members are competitors in some markets and customers in others, and the exempt purpose is the improvement of interconnection conditions in a region. It is not a certification of democratic internal governance.

A 501(c)(6) organisation can be governed by a self-perpetuating board with no member vote at all, and the record does not distinguish between that model and a member-reserved one.

That distinction is why I am careful with a word the exchange uses freely: "member". The front page speaks of "member networks" and of low-cost interconnection "between member networks". Nothing in the retained text defines what membership entails, whether it is acquired by connecting a port, or whether a member network holds any governance right whatsoever. The word appears in marketing context and in operational context. It does not appear in a rights context anywhere I can read.

The instruments named, and the gap

The FAQ page carries the most important sentence in this record, because it is the one place where the exchange answers the governance question directly rather than by implication. Asked "What is the organizational structure of the SIX?", it answers: "The Seattle Internet Exchange is a Washington State nonprofit corporation. Our articles of incorporation, bylaws, and minutes of official meetings are available on our documents page. The SIX is an IRS 501(c)(6) income tax exempt organization." Asked who is on the board and who the officers are, it answers in three words: "Check out the Who's Who." (seattleix.net/faq)

That reply is more useful than it first appears, and less conclusive than it might seem. It is useful because it names the exact instrument class that would settle the question. Articles of incorporation establish the corporation and its authorised purposes. Bylaws allocate power: who the members are, whether they vote, how directors are chosen, which decisions are reserved to the board, which are delegated to officers, and how any of that may be amended. Minutes of official meetings record the exercise of whatever authority the bylaws confer, dated and attributed.

Together those three document types form a complete chain from formal allocation to actual decision. A Who's Who page completes a fourth link by attaching names to offices.

It is inconclusive because none of those documents is in this record. Between 08:43 and 09:16 UTC on 21 October 2026 I made six attempts to retrieve them. Four ended in provider-side failures of one kind or another; two returned a fallback path that reported, in terms, that no web-search or URL-retrieval capability was exposed in that session and that no public-web request had been made. The last attempt returned zero source candidates and an explicit statement that no SIX page or governing document had been inspected.

I want to be exact about what that establishes, because both over-reading and under-reading it would be dishonest. It establishes non-retrieval in one session on one day. It does not establish that the documents are absent from the public web. It does not establish that they are withheld, secret, restricted, or that the exchange has anything to hide. The FAQ says they are available on a documents page; I have no evidence contradicting that, only evidence that I could not reach the page. A reader who wants the answer can do in ten minutes what this record could not do in thirty-three: open the documents page and read the bylaws.

The other half of the boundary is equally firm. Nothing in the retained operator text allocates decision rights. Not one sentence in the front page, the about page, the join page or the FAQ states who may set the port schedule, who may admit a new participant, who may change technical rules on the fabric, or whether participants have any vote in any of it. The exchange asserts a legal form and points to instruments. It does not, in any passage I hold, assert a member vote or describe a procedure for one.

What would settle it

A negative finding is only worth publishing if it comes with the test that would overturn it. Here is mine, stated as conditions rather than as complaints, and drawn from the structure of the instruments the exchange itself names.

First, demonstrating a member vote on fees would require current adopted bylaws that define an eligible voting class, distinguish legal members from operational participants, allocate the specific subject of the port schedule to that class, and state the notice, quorum and approval threshold that applies. A document that grants members the right to elect directors, without more, establishes indirect electoral influence and not a vote on each tariff. That distinction is not pedantry; it is the difference between a member who can replace the people who set a price and a member who can vote the price down.

Second, demonstrating that member authority produced a particular rule would require a dated record of its exercise: minutes, a resolution, or a ballot record showing that the tariff in force was approved by that body on that date. The instruments allocate; the minutes show use. An organisation can hold a formal right it never exercises, and a minute is the only thing that distinguishes a live procedure from a dormant one.

Third, demonstrating delegated management authority would require a resolution or bylaw provision delegating rate-setting to officers or administrators, with the delegation chain named. This is the most likely of the three models in a working nonprofit, and it would not be a scandal. It would simply mean the tariff reflects professional administration under board oversight, which is a perfectly ordinary and defensible arrangement — and one a participant would want to know about, because delegated pricing tends to move on operational logic and notice rather than on membership debate.

There is a fourth condition specific to the pricing trigger. If the complete join-page tariff limits the trigger to a base port charge while excluding monthly billing, later modifications or extension charges, then my reading above is too broad and should be narrowed to the base charge only. The complete text would settle that too.

And a fifth, which I owe to the directory record rather than to the operator. If any page or tariff notice states a 9000-byte and a 1500-byte peering fabric, then the matter discussed in the next section stops being an inference from a database and becomes a documented design decision.

The dual-MTU record and what it is not

The BTW directory holds three records that appear to describe this subject. One is the entity card for Seattle Internet Exchange. The other two are named "Seattle Internet Exchange (MTU 9000)" and "Seattle Internet Exchange (MTU 1500)", both with region United States, both with zero articles, and neither with a confirmed operator. Those two records differ in nothing except the maximum transmission unit in their names.

That is a suggestive pair of records for anyone who has worked with exchange fabrics. Maximum transmission unit is one of the earliest and most consequential design choices an exchange makes, because a fabric that carries 9000-byte frames can serve jumbo-frame applications across many hops while a 1500-byte fabric cannot, and the two cannot be trivially mixed at a shared port. Large exchanges have, historically, handled this by running separate peering LANs and letting participants choose, which is why a directory system that catalogues exchanges might end up with two rows distinguished only by MTU.

But a record in a directory is not evidence of a design. No page captured in this run — front page, about, join or FAQ — mentions jumbo frames, 9000 bytes, 1500 bytes, or a split peering LAN. The MTU distinction exists in this article only as what it is: two directory records with no confirmed operator, presented as an unresolved observation rather than as an exchange-stated architecture. If you want to know whether the Seattle fabric actually separates frame sizes, the answer is on the exchange's own technical documentation, not here, and I decline to manufacture the inference that the directory rows invite.

What BTW's own card records

Because this article relies on a directory, it owes the reader an accurate account of what that directory says. The entity card for the Seattle Internet Exchange, last updated on 16 June 2026 at 23:35:44 UTC, records the object as a company whose known role is "network infrastructure operator", with region Global, a legal type of "private company", a legal name matching the display name, a head office country and registration jurisdiction both recorded as "Global", and a confidence score of 0.95. It lists one contact, scoped to peering, held in a gated layer. It lists no relationships, no related events, no leadership contacts and no profile article. Its entire evidence basis is a single entry: the AS-number record at rdap.org/autnum/33108, labelled "Public registry record", carrying a confidence of 0.82.

Three observations follow, and I state them as observations about the internal record rather than as accusations about the operator.

The first is that a card scoring 0.95 rests on one evidence edge scoring 0.82. Whatever the scoring means internally, the two numbers describe different things: one is the card's standing, the other is the source behind it. A single AS-number registration is a thin basis for an entity record, and it explains why the card's data gaps include verifying the legal entity's country and its services.

The second is the legal-type conflict. The card says "private company" with jurisdiction "Global". The operator says, in its own words, that it is a Washington State nonprofit corporation. Both statements are in the record; they cannot both be right, and the operator's own statement is the more authoritative on its own legal form. I report the conflict rather than resolve it, because resolving it properly requires the Washington corporate record, which this run did not obtain.

The third is that the directory's uncertainties and the operator's silence point at the same missing file. The card lacks a legal-form document. The operator's pages lack an allocation of authority. Both would be answered by the same three instruments — articles, bylaws, minutes — that the FAQ names and this record could not reach. That convergence is the strongest argument for treating the retrieval gap as a gap in coverage rather than as a finding about the exchange.

Where this leaves a reader

Three things are now documented, and they are worth separating cleanly. The Seattle Internet Exchange presents itself as a Washington State nonprofit corporation with 501(c)(6) status, established in April 1997, incorporated as a non-profit in May 2001, and recognised as tax-exempt in 2008. It sells ports through a published tariff whose key feature is a trigger: the rate in force at the earlier of LOA issuance or port configuration, charged once by default, with a monthly payment option, a $100 extension charge, and further extension fees set outside its own price list.

And it names, on its own FAQ page, the instruments — articles of incorporation, bylaws and minutes of official meetings — that would show who holds the authority to change any of that.

One thing is not documented: which of those models is real. A member-reserved fee power, a board-set fee under member electoral influence, and a delegated administrative fee are all consistent with everything in this record, and they have materially different implications for a participant deciding what to expect at renewal. The difference between them is not knowable from a published price list, and I will not pretend otherwise.

The practical consequence is narrower than a governance verdict and more useful than one. If you are buying a port at this exchange, the price you will pay is knowable in advance and defensible in a budget; the procedure that might change it is not, and on the evidence here that procedure lives in a document you can read in a sitting. The exchange has published a tariff and pointed at its own constitution. The gap between those two acts is where a reader's diligence belongs — not in the directory record, not in this article, and not in any claim that the missing document does not exist.

For the record on novelty: two earlier BTW pieces examined this entity. A case file published on 17 October 2026 compared registry, operator and aggregator records and found a listing resting on a single autonomous-system entry; a briefing published on 19 October 2026 reconciled five published port classes against the exchange's own peak and capacity figures. Both interrogated what the exchange says about itself numerically. Neither asked who may change the price, which is the question this article puts to the same page set. The directory entry for the exchange preserves the record this article tested.