Summary

  • The London Internet Exchange's homepage, in the copy retrieved on 17 October 2026, states "13Tbps" of "peak traffic across the exchange" and "900+" "Members networks worldwide". Neither figure carries a measurement date, a definition, or a counting rule.
  • The exchange's one dated public declaration in this record is "a new global maximum peak traffic of 12.069 Tbps" across IXPs in the UK, US, Africa and the Middle East, published in January 2026. It is numerically lower than the undated homepage headline.
  • The two cannot be chained into growth. They differ in scope, publisher and date provenance, and no document in this record reconciles them. Nothing in the record shows either figure to be wrong.
  • Member count, port capacity and peak traffic measure three different things. Treating them as a single scale signal is the most consequential error this record supports.
  • The financing question stays open. The registered company name and number appear in retained page text; no account line, filing date, fee amount or statement of legal form does.

Two numbers, one narrator

Start with the sentence that does the work. The exchange's own site, in the copy retrieved for this report on 17 October 2026, presents two headline figures together: "900+" described as "Members networks worldwide", and "13Tbps" described as "Of peak traffic across the exchange" (linx.net).

Neither string carries a date. There is no "as of", no measurement window, no statement of whether the traffic figure sums both directions or one, no indication of whether the member count includes dormant networks, and no link in the retained text that would let a reader resolve any of it. The page also names two London fabrics — "London (LON1) The original and largest LINX exchange" and "London (LON2) Diverse from LON1 for dual-site resilience" — which is the closest the retained copy comes to describing physical plant.

The dated figure lives somewhere else. On 27 January 2026, trade coverage reported that "The London Internet Exchange (LINX) recorded a new global maximum peak traffic of 12.069 Tbps across their Internet Exchange Points (IXPs) in UK, US, Africa and the Middle East, fuelled by the recent UEFA Champions League, European League and Conference League games matches and AFCON finals" (Advanced Television).

That is the whole comparison, and it is worth stating precisely what it is not. It is not a trend. The homepage's 13 Tbps is higher than the January 2026 declaration of 12.069 Tbps, and a careless reader will see a 0.931 Tbps increment and call it growth. The record does not support that reading, and it does not support the opposite reading either. The larger number is undated and first-party. The smaller number is dated by publication and reported by a third party quoting the operator. Neither page states when its own measurement was taken.

A later event under a different aggregation rule, a different sampling interval, a different geographic boundary, simple display rounding, or a homepage that has not been touched since before January 2026 would each explain the ordering without any misreporting at all.

The honest formulation is narrow: an undated operator headline of 13 Tbps sits above a trade-reported global maximum of 12.069 Tbps published in January 2026, and no retained document reconciles them. Everything else is inference, and inference is precisely what a buyer or a valuer should refuse to accept from a headline.

Three scorecards, not one

The second thing the record establishes is structural. An exchange publishes at least three kinds of number, and they are not substitutes.

Participation asks who is counted. The retained copy gives "900+" and the phrase "Members networks worldwide". It does not give an as-of date, a rule for whether a network must be actively peering to count, a distinction between members and member networks, any churn or retention figure, or any concentration measure. A count of nine hundred members and a count of nine hundred member networks attached to four hundred organisations are different facts about an exchange, and the retained text cannot tell them apart.

Physical capacity asks what has been built. The retained copy names two London fabrics and describes one as "Diverse from LON1 for dual-site resilience". A syndicated report of a completed upgrade describes "a broad upgrade of its 17-site LON2 network in the capital, handing Nokia a deeper role in one of the UK's most important interconnection systems" (MSN). That sentence is the only retained text in this record that puts a site number against a named fabric, and it arrives through a syndication of someone else's report rather than from a page that authored it. No retained document lists the seventeen sites, says which are live, distinguishes a second site from a second independent path, or gives port counts, interface speeds, fabric limits, spare capacity or utilisation at any location.

Volume asks what actually flowed. The retained copy gives one undated exchange-wide figure and one dated four-region figure, with no site-level distribution, no average alongside the peak, and no statement of interval.

Keeping these apart is not pedantry. It changes what a number means. Nine hundred members describes potential reach. Two named London fabrics and a reported seventeen-site build describes installed plant. Thirteen or twelve terabits per second describes one observed load event. An operator can be large on one scorecard and modest on another, and a reader who collapses them will systematically overestimate what a single port at a single site will deliver.

The financing question the record does not answer

The commission behind this piece asked how a member-owned exchange finances capacity: what the filed accounts show for revenue composition, surplus, capital expenditure, member fees and rebates.

This record does not answer that question, and the gap is worth describing exactly rather than papering over with plausible inference.

What is retained is thin and precise. The Companies House filing-history page carries the string "LONDON INTERNET EXCHANGE LIMITED Company number 03137929" (Companies House). The operator's own footer renders a matching registration: "Registered in England and Wales: 3137929" (linx.net). Those two strings agree on the entity. That is the entire financial base of this article.

What is not retained matters as much. No account line — revenue, cost, surplus, capital expenditure, liabilities, cash — appears in retained page text. No filing date or filing type does. No statement of legal form does. No fee amount, tariff or rebate does.

The homepage names two fee pages, "Service Fees Port access, peering bandwidth and service pricing" and "Cross Connect & Installation Fees Cross connects, setup and other one-off fees", and a second retrieval of the site during this work returned no excerpt at all, with the retrieval record stating that the fee, traffic-statistics and member-list pages could not be obtained.

The correct reading of that pair of page titles is narrow: they establish that the operator publishes, or intends to publish, a service-fee schedule and a cross-connect schedule. They establish no price, no port availability, no bandwidth commitment and no service term. A page title is a disclosure surface, not a disclosure.

The financing question therefore remains a question. It is also the right question, because a member-owned exchange cannot distribute profit to outside shareholders, which means its funded capacity has to show up somewhere else: in port fees, in cross-connect charges, in member subscriptions, in a surplus retained and spent, or in some combination. Until those lines are read directly, anyone describing how the exchange pays for its fabric is describing a hypothesis.

Where the sites are, and what "diverse" is doing in that sentence

The most useful word in the retained operator copy is "diverse". LON2 is described as "Diverse from LON1 for dual-site resilience" (linx.net).

In interconnection practice, "diverse" can mean several different things, and the difference is commercial, not semantic. Two sites can be physically separate buildings fed from the same utility substation and the same fibre duct, in which case a single civil event takes both. They can be separate buildings on genuinely independent cable routes with independent power feeds. Or "diverse" can be a product description meaning that a customer who buys across both sites has a contractual path-diversity commitment, whether or not the underlying plant is diverse in the engineering sense.

Nothing in the retained text resolves which of those is meant. The phrase "dual-site resilience" establishes that there are two sites and that resilience is the claimed purpose. It does not establish independent path. A buyer who reads "diverse" as "independent route" and then discovers a shared duct has bought a different product than the one they believed they were buying, and the discovery usually arrives during the incident that the diversity was supposed to survive.

The seventeen-site figure sharpens the same point. If LON2 is a fabric spread across seventeen locations in London, then "LON2" is not one place and its resilience properties cannot be stated once for the whole fabric. Some of those seventeen sites will be more redundant than others. No retained document says which.

The second narrator who was not there

This is where the evidence base becomes uncomfortable, and it should be stated plainly rather than smoothed over.

Both traffic figures in this article trace back to the operator. The 13 Tbps headline is first-party by definition. The 12.069 Tbps declaration is trade press reporting what the exchange announced. Those are two publication events, not two independent measurements, and treating them as corroboration would be exactly the error of counting one narrator twice.

Efforts to find a genuinely independent narrator in this work did not succeed. The PeeringDB organisation record and the Euro-IX exchange profile for the operator were both identified by URL and both returned no excerpt; they are named but unretrieved, which is a different condition from absent, and a different condition again from contradictory. A final research pass returned no candidates at all after the retrieval provider reported that web search was unavailable. That is a dated fact about this record's limits, not a finding about the exchange.

The consequence is specific. This article can tell a reader what the operator says about itself, and what one trade outlet reported it saying on one date. It cannot tell a reader what an independent measurement body, an academic study, a regulator or a member's own telemetry says. Where a claim about scale matters, that absence should travel with it.

What a buyer of a port should actually ask for

The practical value of all this is a short list of questions, and the discipline of asking them before signing rather than after an outage.

Ask for port availability and interface speeds at the specific site, date-stamped. An exchange-wide peak does not describe a single fabric, and a four-region aggregate does not describe London.

Ask for site-specific peer and traffic data. Which networks are present at the site being bought, and what traffic is actually exchangeable there, is the question a port is supposed to answer.

Ask for cross-connect pricing and installation lead times in writing. The retained page titles establish that a cross-connect schedule exists; they say nothing about what a cross-connect costs or how long it takes.

Ask whether site diversity is physical route diversity, contractual commitment, or a marketing description — and get the answer in a form that survives the first incident.

Ask for maintenance and outage history, service levels, credits and continuity terms. A peak-traffic record is a statement about good days.

Treat the global headline as a screening signal and nothing more. It tells a buyer that the exchange is material. It does not tell a buyer what they are buying.

What a valuer should build instead of a multiple

The valuation temptation is to capitalise the headline: nine hundred members, thirteen terabits, apply a per-member or per-terabit multiple, produce a number. That exercise would be building a valuation on two figures whose definitions are unknown, whose dates are unknown, and which do not reconcile with each other.

The alternative is unglamorous and correct. Decide first what is being valued: operating cash flow, member utility, network effect, or transferable rights. Those are different assets with different buyers, and a member-owned structure usually makes the fourth one the binding constraint, since a member's interest may not be saleable at all.

Then build separate series, not one. Active-member retention, because a stock count says nothing about churn. Site and port capacity with utilisation, because installed plant that nobody uses is a cost, not an asset. Normalised traffic, meaning a series with a stated interval and scope rather than a peak. Pricing, revenue, margin and capital expenditure, once the filed accounts are actually read. Concentration, because an exchange whose traffic is dominated by a handful of members carries a different risk profile from one with a broad base.

And continuity terms, because interconnection value evaporates if service can be withdrawn on short notice.

Where the record lacks reconciled series and direct financials — as this one does — the right output is a scenario range with an explicit verification-risk discount, not a point value. The discount is the price of the gap. A valuer who omits it has not valued the asset more generously; they have valued a different asset, one that does not exist.

What would change this assessment

A short list of observations would move this article from a reconciliation problem to a measurement story, and each is checkable.

A date-stamped traffic-statistics page or archive that places the 13 Tbps figure on a timeline. A member count with a stated as-of date and a counting rule. A filed accounts document read directly, yielding revenue, surplus, capital expenditure and liability lines. A site list or port inventory for LON1 and LON2. An independent measurement of a London fabric by a party with no commercial interest in the answer. A fee schedule with actual amounts and lead times.

Any one of those would materially strengthen the record. None of them was available here, and the absence is the article's central finding rather than an apology for it.

The bounded conclusion

An undated 13 Tbps headline, a dated 12.069 Tbps declaration, a "900+" member count with no counting rule, two named London fabrics, a syndicated report of a seventeen-site build, two fee-page titles, and a company name and number on a public register. That is the entire public record this article could verify.

It is enough to establish that the London Internet Exchange is a material interconnection institution. It is not enough to establish how large, how fast-growing, how redundant, how financed, or how durable. Those are separate questions with separate evidence requirements, and the institution's own published numbers answer none of them fully.

The consequence for a reader is uncomfortable but simple. When an operator's headline and its dated declaration do not reconcile, and no independent narrator is available to arbitrate, the reader's obligation is to carry the ambiguity into the price and the contract rather than to resolve it in the operator's favour. That is not suspicion. It is the minimum discipline that a reconciliation gap of this kind requires.