How sticky is hosting, really? We measured every layer.

Hosting churn numbers are mostly folklore. Surveys ask providers what they believe. Raw "domain migration" feeds count tens of millions of moves a month. Neither tells you what an investor or an operator actually needs: how often do real businesses change each layer of their hosting stack?

So we measured it — every observed infrastructure transition on 35M+ confirmed-active business domains over 26 weeks, normalized, decomposed, and annualized.

The stickiness ladder

The provider relationship is the atom of hosting switching: the serving layer moves without it at 0.08% a year — 150× less often than the relationship itself.

Read the ladder from the bottom. Businesses essentially never re-home where their site is served while keeping the same provider relationship — when infrastructure moves, the relationship moves, as one unit. Above it: providers shuffle IPs internally (3.2%/yr) forty times more often than customers move serving — most IP-level "change" in the market is housekeeping, not churn. And at the top: the relationship itself changes for about one in eight European business domains a year — which means ~88% hold. Trust migrates slowly. Now it's measured.

Do not read 12.2% as a churn target. Switching is not total churn: domains are also lost to abandonment (1.9%/yr above) and to outright expiry, which sits on top of switching — so gross book attrition runs meaningfully higher. And the average hides mix: a business-dense ccTLD book should switch far less than a volume book, so the honest benchmark is your observed rate against your book's mix-adjusted expectation, not against the market average.

Why raw migration numbers mislead

Public feeds report domain migrations in the tens of millions per month. Over the same 26 weeks, our raw log contains ~33M nameserver-change events — yet organic switching by confirmed-active businesses amounts to ~2M domains. The gap is not measurement error; it's composition:

1. Shard rotation isn't switching. Large cloud DNS services rotate nameserver shards; compared as raw strings, that alone fabricated ~30% extra "switches" before normalization.

2. Lifecycle isn't switching. A domain moving from registrar-default onto an operator is onboarding; a domain falling back to default or parking is abandonment. Both are business signals — neither is a customer choosing a new provider.

3. The aftermarket isn't a customer. Parked and drop-catch portfolios flip nameservers constantly — high event volume, near-zero commercial meaning (the same phenomenon that fabricated the "edge surge" in new-site data).

What this means

For operators: the relationship is your moat, and it holds at ~88%/yr — the leak is not mass defection but slow abandonment (1.9%/yr) plus the wallet fragmenting around a relationship that stays. Retention economics should price that asymmetry.

For investors and advisors: relationship-churn is measurable per book — survey churn and raw migration feeds will mislead a model in opposite directions (surveys overstate stickiness of the wallet; feeds overstate movement of the relationship). Book-level switching, tenure and flow detail is in the product.