Revised 28 September 2026. Earlier versions of this brief read the shift among new arrivals as an edge tilt - a CDN put in front of the same kind of hosting. The September readings show the visible-cloud share rising while the edge share held, so we now read it as two fronts of similar size. The figures below are from the current release; the argument is what changed.
Every hosting investor asks some version of the same question: are hyperscalers and edge platforms capturing the new websites? Installed market shares can't answer it - they average over a decade of old decisions. So we measured arrivals instead: every European domain that entered our web-active population over eight complete weeks - 1,878,174 new web-active domains across 17 European markets - against an installed base of 29,028,480, by hosting-front class.
The first cut looked dramatic. The second cut is the story - and it answers both halves of the question the same way: neither front is taking over, and both are gaining ground.
Naive momentum reads are set by single events, not by markets. In Italy, 51.3% of the entire new-domain cohort sits behind one nameserver: read raw, the market's "momentum" is one operator's registration book. 3 of the 17 markets we measure trip the same screen.
That is not tens of thousands of Italian businesses making an infrastructure choice; it is one portfolio. The screen drops a flagged market entirely - from the arrivals and from the installed base it is compared against - so it changes the country mix on both sides, not just the headline. Here that widens the gap it is measuring: read raw, new European domains are CDN-fronted at 16.4% against 14.7% installed, a gap of about two points; screened, 16.1% against 13%, about three. The markets it removes carry an unusually edge-heavy installed base, so leaving them in flattens the comparison rather than inflating it.
Single operators move portfolios. Markets move slowly. If you don't screen bulk events, you will mistake one for the other.
With bulk events screened out (1,268,868 organic new domains against 21,504,391 installed), all three classes move, and only one of them moves down.
Where does that ground go? To both of the other two fronts, in broadly similar size, and that is the part the earlier read of this brief got wrong. We argued an edge tilt: the new cohort putting a CDN in front of the same kind of hosting. The CDN/edge front is real, and across those same September readings it held steady. What has changed is the other side of the split - visible hyperscaler origins among new arrivals are now around one in eight (12.1%), against roughly one in twelve of the installed base (8.2%), and across those September readings that gap grew in every scope we measure while the edge gap did not. The two fronts are no longer one story and a footnote. They are two fronts of similar size, and in September it was the cloud one that moved.
That still is not a hyperscaler takeover, and the number says so plainly: about one new European website in eight starts on a visible AWS, Google or Azure origin. Read the installed base against it and the direction is unmistakable, but the level is a minority of a minority. The honest sentence is the boring one - the cloud is taking a growing slice of a shrinking traditional lead, from a small base, and there is no week in this data where it takes the market.
Three mechanics distort naive "the cloud is taking the new sites" reads - and the third is why the cloud side of the split above is a floor rather than an estimate:
1. Placeholder pollution. Fresh domains disproportionately sit on registrar forwarding and placeholder infrastructure - which runs on major-cloud IPs. To an ASN-level read, a parked new domain looks like a cloud-hosted new website. Our web-active tier excludes placeholder and forwarding infrastructure by reverse-DNS fingerprint, so that inflation never enters these numbers.
2. Bulk events. One operator, one window, a hundred thousand domains - screened here, headline-making if not. They can push a number either way, and in this cohort they push two at once: read raw, the edge gap looks smaller than it is and the visible-cloud gap looks larger.
3. The masking bound, stated. For about one in six organic new sites (16.1%) the origin sits behind a CDN and we cannot see it. That bound used to be the reason to discount this brief's headline, because the headline was the edge front - a rising CDN share does not say a site moved off traditional hosting, it says we stopped being able to see where it is served from. It cuts the other way now. The front that is moving is the visible one: a hyperscaler origin we can name, not an origin we have to bound. And the masked share held steady through September, so the movement cannot be an artefact of the bound widening. If anything, the masked sixth means the cloud side is a floor rather than an estimate - we report what we can see, and how much we cannot, and in September what we can see is the part that moved.
For operators: the feared leak of new customers to hyperscalers is no longer invisible in European arrival data, and it is not a takeover either. Visible hyperscaler origins run at about one new site in eight (12.1%) against one in twelve installed (8.2%), and that gap widened in each September reading. What that changes is the question to put to your own book: not "are we losing sites to the cloud" - at these levels you are not, at any speed that shows up in a renewal cohort this year - but "which segment of new business is choosing it", because a drift of a few points a quarter concentrated in one customer type is a segment leaving, and a drift spread evenly is a market moving. The front door of new websites still opens on traditional hosting about seven times in ten. The fight is still winnable at acquisition; it is no longer winnable by assuming the other side is not competing.
For investors and advisors: a momentum claim is only as good as the window and the screen behind it. The same eight weeks read three ways - raw Europe, screened Europe, global - can put a different front in first place, which is why we publish all three and lead on the screened one. Two questions for management, both answerable from their own data: what share of last quarter's new customers arrived on a cloud-origin stack, and is that share rising in the segments carrying the renewal margin. A group that cannot separate those two has a momentum story rather than a number.
Every figure here is queryable through the HostingBrain connector. In Claude or any MCP-compatible assistant, this is the whole brief in one prompt:
“Using HostingBrain, show new-site hosting momentum in Europe: among domains newly observed in the last eight weeks, which hosting classes (CDN-fronted, hyperscaler, traditional) are over- or under-represented versus the installed base - with bulk events screened out.”
Resolves to hosting_momentum (free). See definitions('bulk_event_screening').
Reproduce this analysis: the hosting_momentum
tool returns every number above - global, Europe raw vs screened, and per-market with bulk-event flags -
and it is available on the free tier.
Ask the follow-up yourself. HostingBrain answers questions like this - with the date, denominator and caveats attached - inside Claude and any MCP-compatible assistant.