Email infrastructure changes slowly. Across 192 DNS snapshots of the Tranco top-1M between 2016 and the 2026-07-05 snapshot, the typical ESP trajectory is a gentle slope: Mailchimp's much-discussed decline took years to shave a quarter off its peak, and Mandrill still holds most of its share a decade after being sunset as a standalone product. Against that backdrop, one line on the chart looks like a different phenomenon entirely.
MailChannels — the relay service best known for sending mail on behalf of web-hosting platforms — peaked at 2.94% of SPF domains in March 2021. As of the 2026-07-05 snapshot it stands at 1.07%. That is a 64% collapse in share, unfolding over roughly five years and clearly visible as a cliff in the data.
What the data does and does not say
Let us be precise about the claim. Our scan classifies a domain as a MailChannels sender when its SPF record includes MailChannels' mechanisms. The chart therefore measures one thing: how many top-1M domains authorize MailChannels to send for them. That number rose steadily through 2020, peaked in March 2021, and has fallen ever since.
What the DNS cannot tell us is why. MailChannels' business went through publicly visible changes in those years — shifts in its platform partnerships and product focus that were widely discussed in hosting communities — but a DNS census does not observe contracts or pricing, and we will not present any single cause as fact. What we can say with confidence is structural: MailChannels' growth came largely through platforms — hosting providers and edge services that bundled it as the default outbound relay for their customers. Share acquired through a platform can leave through the platform, wholesale, when the bundling changes. Thousands of domains do not coordinate an exodus; one platform decision moves them all at once.
Bundled infrastructure is borrowed infrastructure
That is the actual lesson of the cliff, and it applies far beyond one vendor. A large share of senders never chose MailChannels — they chose a web host, and MailChannels came with it. The same is true today of senders on e-commerce platforms, site builders and CRM suites whose outbound mail rides a relay the customer has never evaluated. When your sending infrastructure is a side effect of another purchase, you inherit three risks:
- Repricing and repackaging risk. The relay's deal is with the platform, not with you. If that deal changes, your mail path changes on someone else's timeline.
- Shared-reputation risk. Bundled relays pool enormous numbers of small senders. You share IP reputation with whoever else the platform onboarded this week.
- Silent-migration risk. Platforms swap relays without asking. Your deliverability can change overnight with no notice, no changelog entry, and no action on your part.
A portability checklist for relay dependence
You cannot always avoid bundled sending — sometimes it is the right economic choice. You can make sure that a MailChannels-style cliff under your feet is an inconvenience, not a crisis:
- Authenticate on your own domain. DKIM keys on your domain, a DMARC policy you control, and an SPF record you can edit in minutes. If the platform only signs with its own domain, your reputation is non-portable by construction.
- Keep your data exportable. Suppression lists, bounce history and engagement data should live somewhere you can extract them from. Rebuilding a suppression list from scratch is how migrations turn into blocklist incidents.
- Know your switch time. Run the thought experiment: if your relay announced a change effective in 30 days, what would you do? If the answer involves discovering how your mail is sent, do that discovery now.
- Monitor the vendor's public trajectory. A relay losing domains at scale is visible in public DNS long before it shows up in your metrics. The MailChannels series — and every other major provider — updates daily in our daily email infrastructure report, an approach we describe in reading provider health from public DNS.
Slow decline is normal life for a mature ESP — Postmark drifted from 0.76% to 0.47% over nine years without drama. A fast, sustained drop like 2.94% → 1.07% is a different signal class: whatever its cause, it means the sender population you share infrastructure with is changing rapidly, and your deliverability assumptions should be re-tested rather than assumed.
If your relay is changing under you
Suppose you are on a platform whose relay is repricing, repackaging or losing share fast. The playbook is unglamorous: stand up authentication on a domain you control, shortlist a replacement relay, warm it with a slice of real traffic, and compare placement empirically before moving volume. The comparison step is the one senders skip — and the cheapest to do. Send the same message through the old and new path to seed mailboxes at the major providers and see where each lands. A one-hour placement test beats weeks of post-migration forensics, and it converts a platform-risk event into a controlled infrastructure change.