Summary

  • The Saudi Arabian Internet Exchange reports three sites, 41 connected members, 157 connected ports, 222 total ASNs, 6.6 Tbps of provisioned capacity and an all-time traffic peak of 1.65 Tb/s — figures captured from its own pages on 30 October 2026.
  • Read together, those numbers say the fabric's best recorded hour used one quarter of the capacity it says is provisioned, and its current rate roughly a tenth. If you are deciding whether to buy transit or connect a content cache in the Kingdom, that is the number that matters, and nobody outside the operator has verified it.

There is a particular kind of confidence in a webpage that publishes its own scoreboard. The Saudi Arabian Internet Exchange does exactly that: a counter block at the top of the site, eight figures, all of them crisp. Three sites. 41 connected members. 157 connected ports. 6.6 Tbps of provisioned capacity. 222 total ASNs. More than 32,000 IPv4 and IPv6 prefixes. An all-time peak of 1.65 Tb/s. A current rate of 642.5 Gb/s.

The numbers were captured in a single retrieval on 30 October 2026, from the operator's own pages. They are the only measured values in this article, and every one of them is the exchange's own statement about itself. That provenance is not a quibble. It is the entire evidential situation, and it is what makes the scoreboard worth reading closely rather than quoting.

What the operator publishes

The homepage describes SAIX as "a non-profit exchange point brought by MCIT in cooperation with KACST" — the Ministry of Communications and Information Technology and the King Abdulaziz City for Science and Technology. That is the operator's own attribution of its origin, published on its front page. No founding document, licence or charter was retrieved here to confirm the legal arrangement, and this article does not treat a website sentence as a corporate record.

The same page states the exchange's purpose in one sentence: it "interconnects networks and optimize traffic performance in Saudi Arabia by reducing latency and improving localization of both content and traffic." Localisation of content. Localisation of traffic. That phrase is doing real work in this story, and I will come back to it.

The peering page adds the technical shape: services offered "on top of industry most advanced Layer 2 switching platforms," with public peering letting a member reach multiple networks — ISPs, carriers, content providers and CDNs — through a single connection, and private peering offering dedicated traffic exchange between groups of members over a private VLAN. Layer 2 switching, public and private peering, one port to many networks: this is the standard architecture of a modern regional exchange, described in the standard way.

What the site does not publish is at least as telling. The members page, as retained here, resolves to navigation text — a Member List link, a Become A Member link, a login. No member names, no network names, no facility names, no per-site breakdown. Three sites are claimed and none are named.

What the counters mean, and what they do not

Start with the pair that gets read as interchangeable and is not: 41 members and 222 ASNs. Members are organisations that have signed up and taken a port. ASNs are autonomous systems — routable identities. The ratio is about 5.4 ASNs per member. It follows that the exchange is not counting one ASN per member, and it also follows that 222 is not a customer count. It may be networks reachable across the fabric, networks peering at it, or some other convention the site does not state. Nobody reading the counter block can tell which.

The port count is similarly elastic. 157 ports against 41 members is about 3.8 ports per member. That could mean most members take redundant or multi-site connections — sensible for a carrier or a CDN with three sites to reach. It could also mean the port total includes infrastructure links that no member holds. The page does not separate them.

The prefix figure, published as "+32K", is the vaguest of the set and the most impressive-sounding. Divided by 222 ASNs it works out to roughly 144 prefixes per ASN, which is plausible for a fabric carrying content networks that announce many more-specific routes. But "+32K" is a rounded claim with no timestamp, and a rounded claim is a floor, not a measurement.

Then there is the traffic block, which at least orders itself honestly. Current 642.5 Gb/s sits below the average of 903.08 Gb/s, which sits below the 1.22 Tb/s peak, which sits below the 1.65 Tb/s all-time peak. A stale or carelessly assembled counter set would not normally produce that consistent ordering. Internal consistency is not verification — a single template can generate consistent numbers — but an inconsistent set would have been positive evidence that something was wrong, and this set is not inconsistent.

What none of it carries is a date. Not one counter says when it was last updated, over what window the average is taken, or how the peak was measured. A number without a window is an assertion about a moment nobody can locate.

The capacity gap in plain arithmetic

Here is where the published figures become genuinely useful, because they can be divided.

6,600 Gb/s of provisioned capacity against a 1,650 Gb/s all-time peak: the best hour this exchange has ever recorded filled one quarter of the capacity it says is provisioned. Against the current rate of 642.5 Gb/s, that is about 9.7 percent. Against the average of 903.08 Gb/s, about 13.7 percent.

Those ratios come with a definitional caveat large enough to drive a truck through. "Provisioned capacity" is not defined on the page. If it means the aggregate capacity of ports handed to members, then a quarter is a reasonable and even healthy utilisation figure for a fabric built to grow — nobody sizes an exchange to its current peak. If it means switch fabric capacity, then a quarter is still unremarkable for a young exchange. If it means something else, the ratios describe something else.

The operator publishes the numerator and the denominator without the definition that would let a reader compare them, and the four-times gap between provisioned capacity and best-ever hour is therefore a fact about the scoreboard, not a proved fact about the switches.

That distinction matters to anyone deciding whether to connect. What the figures do establish is that this is not an exchange straining at its limits. There is no published evidence of congestion, of capacity exhaustion, or of an operator racing to add fabric because the last terabit is full. The published picture is of headroom: a fabric provisioned well ahead of its measured traffic, in a country whose operator says its purpose is to keep content and traffic local.

Who connects, and the members page that does not say

An exchange is a room full of other people's decisions, and the decisions are the story: which carriers buy ports, which content networks cache inside the country, which internet service providers stop hauling traffic to Europe. SAIX publishes none of it.

The gap is not trivial. A 41-member exchange with 222 ASNs and three sites could be a genuinely national fabric serving the Kingdom's carriers and the global content networks that want a local foothold. It could equally be a smaller number of organisations holding many ports, with the ASN count reflecting reachable networks rather than connected ones. Both are consistent with the published counters. The two pictures imply very different things about resilience, about the balance of local versus foreign participants, and about whether the exchange is a genuine domestic meeting point or a landing strip for inbound content.

That question also decides how to read the three sites. Three sites is a resilience claim if the sites are separate facilities in separate cities with separate power and fibre paths. It is a marketing claim if they are three racks in one operator's campus. The site publishes the number and withholds the answer, and no facility document was retained here that would settle it. I am not going to merge three sites into one building, and I am not going to credit them with redundancy on the strength of a counter.

The policy frame the operator states

The tempting move here is to explain the capacity gap with policy: a state-directed build, provisioned ahead of demand because the Kingdom wants its traffic at home. The operator half-invites it. MCIT and KACST are named. The purpose sentence says localisation outright.

But an exchange built with government involvement and a mission statement about localising traffic is not the same thing as an exchange whose growth is caused by legal compulsion. That link needs documents: dated instruments that require or reward keeping data and traffic inside the Kingdom, and a dated membership record showing the counters moving with them. I went looking for both. The PeeringDB record for exchange identifier 926 — the obvious independent cross-check on a member count — did not return a record in this run. Two attempts at Saudi regulatory sources returned service-level failures rather than text. So the policy account stays where the evidence puts it: a purpose statement written by the operator, and nothing more.

What I will say is narrower and defensible. An exchange that advertises localisation as its purpose, reports a quarter-utilisation ceiling and withholds the composition of its membership is asking to be taken on its own account. For a carrier deciding whether to peer there, the localisation claim is the promise; the utilisation figure is the headroom; the missing membership list is the part that would tell you who you would actually be meeting.

What could not be verified

The honest inventory is short and it matters.

No independent measurement of SAIX's membership, traffic or facilities was retained. The PeeringDB record did not load, which is a retrieval failure in this run and not evidence about PeeringDB's coverage — a distinction worth keeping sharp, because a directory that failed to answer says nothing about the exchange, only about my access.

No member or network names were captured, so nothing here identifies who peers at SAIX or how the 41 members distribute across the three sites.

No regulatory text was retained, so nothing here states what Saudi instruments require of a hosting provider or a carrier.

No legal-form document was retained, so "non-profit" remains the operator's description of itself.

And no counter carried a timestamp, so the entire measured picture is a snapshot of an undated scoreboard on a page retrieved on 30 October 2026.

What to watch

Four things would move this assessment, and all of them are observable.

A dated member list, published by the operator or logged by a directory that actually answers, would test the 41 and the 222 and reveal whether ASNs mean reachability or connection. A definition of "provisioned capacity" would turn the headline ratios from arithmetic into measurement. A facility record naming the three sites would settle whether they add resilience or only add slides. And a published traffic series — even monthly — would show whether the fabric is filling, because a quarter-utilised exchange on the way to half is a different proposition from a quarter-utilised exchange that has plateaued.

Until one of those appears, the usable finding is this: the Kingdom's national exchange has built a fabric substantially ahead of its best measured hour, describes its purpose as keeping content and traffic at home, and publishes a scoreboard that is internally consistent, definitionally thin and entirely its own word. That is not a scandal. It is also not verification, and the difference between those two is precisely what a reader deciding whether to connect, buy or believe needs to hold on to.