Email infrastructure
KumoMTA serversOpen-source MTA, Rust + Lua, no per-message feesPowerMTA serversLicensed, VirtualMTA pools, vendor supportDeliverabilityWarm-up, DMARC alignment, feedback loopsServers
Bare metal — AMD EPYCSingle-tenant Turin / Genoa, up to 192 coresGPU & AI serversNVIDIA Blackwell B200 / H200, PCIe Gen5VPSPer-core instances on the same EPYC hostsCloud & network
Cloud computeHourly instances, snapshots, NVMe Gen5DDoS protectionEdge absorption included on every serverNetwork & regionsAmsterdam · Frankfurt · London · AshburnTier-0 hosting · Email & compute
VV Internet Hosting runs single-tenant servers tuned for two demanding jobs: high-volume email delivery on KumoMTA or PowerMTA, and heavy compute on AMD EPYC bare metal and NVIDIA Blackwell GPUs. It is built for senders, platform teams, and engineers who need the machine to themselves.
The figures above are illustrative of platform-wide volume; your own delivery numbers live in your panel, measured per pool and per mailbox provider, so you watch the same signals our engineers do.
In short
The stack
Pay for what you run and subscribe to what you keep. Email hosts, bare metal, cloud instances, and GPU nodes are provisioned from the same place, billed in one invoice, and watched by the same team. One panel means a sending host and the database behind it are neighbours, not two accounts you reconcile at the end of the month, and one team means the engineer who reads your queue depth also knows your network.
Tuned bare metal for the open-source MTA that large senders use to leave PowerMTA and Momentum behind.
Millions of msgs/hr per node EmailHardened hosts sized for Port25/Bird's licensed MTA, VirtualMTA pools and per-ISP shaping.
VirtualMTA-ready ComputeSingle-tenant AMD EPYC Turin and Genoa, DDR5 and NVMe Gen5. No virtualisation tax, no noisy neighbours.
Up to 192 cores ComputeNVIDIA Blackwell B200 and H200 nodes with Gen5 lanes and local NVMe for training and inference.
B200 / H200 / MI300X CloudHourly instances on the same network, with snapshots and one panel across every product.
From €0.006/hr CloudPer-core VPS on EPYC hosts with NVMe Gen5 storage for general-purpose workloads.
NVMe Gen5 NetworkAlways-on absorption across the edge, included with every server. Nothing to configure.
Multi-Tbps edge EmailDedicated IP pools, reverse DNS, SPF/DKIM/DMARC alignment and feedback-loop wiring, done with you.
Guided IP warm-upEmail infrastructure
The right MTA depends on how you want to operate, not on which name you have heard most. KumoMTA suits teams that want source access, Lua scripting, and no per-message cost; PowerMTA suits teams that want a license, a support contract, and the VirtualMTA model they already know. Both push millions of messages per hour from a single host, so raw speed rarely decides it.
We host both and we operate both. If your volume is modest or your team is small, a managed email service may cost less once you count the hours; we would rather say that now than sell you a server you do not need. If your volume is large, or you send on behalf of your own customers, owning the MTA gives you control that a shared platform cannot: your own IPs, your own reputation, and policy you write rather than policy you inherit. The line between those two cases is usually clear within one sizing call.
Compare the full email stack →
-- VVIH baseline: listener + two egress pools
local kumo = require 'kumo'
kumo.on('init', function()
kumo.start_esmtp_listener {
listen = '0.0.0.0:25',
relay_hosts = { '10.0.0.0/24' },
}
kumo.define_spool { name = 'data', path = '/var/spool/kumomta/data' }
kumo.define_spool { name = 'meta', path = '/var/spool/kumomta/meta' }
end)
-- pick a pool by the campaign tag your app sets
kumo.on('smtp_server_message_received', function(msg)
local pool = msg:get_meta('tenant') == 'mkt'
and 'marketing' or 'transactional'
msg:set_meta('egress_pool', pool)
end)
-- shape traffic per destination, per pool
kumo.on('get_egress_path_config', function(domain, _, _)
return kumo.make_egress_path {
connection_limit = 12,
max_message_rate = '200/s',
enable_tls = 'Opportunistic',
}
end)Deliverability
Inbox placement is decided by reputation and authentication, not by raw sending speed. Mailbox providers score the IP and the domain you send from, check that SPF, DKIM, and DMARC agree, and weigh how recipients react over weeks. A fast MTA gives you the controls to manage all three. It does not earn the reputation on your behalf, which is why a powerful server with no plan behind it still lands in spam.
SPF, DKIM, and DMARC have to agree before a provider trusts the envelope. We set the records with you, sign at the MTA, and move DMARC from monitoring to enforcement once the reports come back clean, so mail spoofed in your name gets rejected instead of landing next to yours and eroding the trust you built.
Transactional receipts and marketing campaigns earn reputation at different rates, so they belong on different IPs. Splitting them keeps a promotional blast from dragging down the password-reset mail a customer is waiting on right now. KumoMTA and PowerMTA both pin a stream to its own pool, and we wire that separation on day one.
New IPs start with no history, so providers treat them with suspicion. We raise volume on a ramp matched to your daily numbers and the mailbox providers you reach, watching open and complaint rates as the ceiling rises. Most domains reach full volume in three to six weeks, and we hold the pace the moment a provider throttles.
Complaints and bounces are signals to act on, not noise to ignore. We connect the major feedback loops, classify hard and soft bounces, and suppress the addresses that hurt your standing, so the list you keep sending to stays the list that wants to hear from you. Reputation is mostly the sum of those small, daily decisions.
None of this ships as a product you buy on top. Authentication, pool layout, warm-up, and feedback wiring are part of standing up a sending host with us, because a server that delivers poorly is not worth hosting and reflects badly on the network it runs from. The goal is mail that lands where it should, measured in your own panel rather than promised in a brochure.
Compute
Dedicated servers are back because the cost of virtualisation moved. After Broadcom changed VMware to bundled subscriptions, many renewals jumped between two and twelve times their old price, so teams started counting cores again. A single AMD EPYC Turin host with up to 192 cores can replace three or four older machines, and on bare metal you pay for the silicon rather than a license that scales with it.
Single-tenant Turin and Genoa processors with DDR5 and NVMe Gen5. Full root, no neighbours, and no hypervisor between you and the cores. Sized for databases, mail hosts, and steady-state workloads that punish shared infrastructure, where a noisy tenant on the same box turns a fast query into an unpredictable one.
EPYC Turin / Genoa · up to 192 cores · DDR5 · NVMe Gen5
NVIDIA Blackwell B200 and H200 nodes with PCIe Gen5 lanes and local NVMe, so the cards are fed instead of idling on data. For training, fine-tuning, and inference you would rather run on your own metal than rent by the cloud-hour, especially once a job runs for weeks rather than minutes and the rental meter never stops.
B200 · H200 · NVLink · local NVMe scratch
When the job is bursty, hourly cloud instances and per-core VPS run on the same EPYC hosts and the same network. Snapshot, scale, and tear down without leaving the panel your servers already live in, and move a workload from a VPS to bare metal when it outgrows shared cores without changing providers.
Hourly billing · snapshots · NVMe Gen5
Operations
A server is only as good as the network and the people around it. These come with every machine, on email hosts and compute alike, instead of arriving as line items you discover at renewal.
Volumetric attacks are scrubbed before they reach your host, with capacity sized for the whole network rather than a single rack. Your SMTP and API traffic keeps flowing while the noise is dropped upstream.
Queue depth, delivery rates, bounce categories, and host health stream to Prometheus and Grafana. The numbers a provider uses to judge you are the numbers you watch, not a softened marketing dashboard.
Moving off Momentum, an aging PowerMTA box, or a managed ESP, we carry over your domains, signing keys, and suppression lists and run old and new in parallel until the fresh IPs are warm.
Support is staffed by engineers who operate mail and metal every day, with a median first reply under a minute. You reach someone who can read a Lua policy, not a tier-one script reader.
Network
Email and compute share one backbone across carrier-neutral facilities in Amsterdam, Frankfurt, London, and Ashburn, with edge points for absorption in front of them. A sending host and the metrics that judge it sit close together, so the feedback you act on arrives in seconds rather than minutes.
Running several regions on one network lets you place a host near the recipients it serves and keep a second region ready for failover, without stitching together separate providers and separate invoices. Routing is ours end to end, which is what lets us put a number on latency and absorption rather than pointing at a transit partner and hoping.
Clean IP ranges with full reverse-DNS control, the records mailbox providers expect, and pools you grow as your volume does.
Low-latency links between regions, private networking between your own hosts, and the bandwidth that GPU and database workloads actually consume.
How we sell
Most hosting sites push you toward the most expensive box. We would rather keep you for a decade. So the advice is shaped by your sending volume, your team size, and your workload — sometimes that means a single tuned server, and sometimes it means a managed service we do not even sell.
That stance has a cost: we talk some people out of buying. It also means the customers who stay tend to stay for years, and they send us the next team that needs a server. We would rather grow that way than on a sale someone regrets at the first renewal.
Steady, high volume; latency-sensitive delivery; data you want on hardware nobody else touches; budgets that core-based licenses would punish.
Low or spiky volume where a managed ESP costs less in hours; tiny lists where a plain Postfix relay is plenty; teams with no one to run an MTA at 3 a.m.
Getting started
Standing up a sending platform should feel like a planned project, with each step owned and dated. Here is the path most new senders follow with us, and where compute customers branch off after step two.
We start from what you send or run — daily volume, peak rate, message size, and the providers you reach — and size the host to it. If a managed service fits you better, this is the moment we say so, before any contract.
Your single-tenant host is built, the MTA is installed and tuned to your profile, and SPF, DKIM, and DMARC records go in. We confirm reverse DNS and TLS before a single message leaves the box.
Volume rises on a schedule matched to your traffic while we watch the feedback loops. Transactional and marketing pools warm in parallel, so neither stream waits on the other to reach full speed.
Once reputation holds at full volume, the panel, the metrics, and the policy files are yours. We stay on call for tuning, but the day-to-day runs without waiting on a ticket queue.
Questions
The things people email us about most, answered plainly.
KumoMTA is an open-source MTA written in Rust and configured in Lua, released under Apache 2.0, with no per-message licensing. PowerMTA is licensed commercial software from Bird (formerly Port25/SparkPost) with VirtualMTA pools and vendor support. Both move millions of messages per hour from a single node; the choice usually comes down to whether you want a license and a support contract or full source access and in-house operations.
You can run either on a VPS for testing or low volume, but serious senders move to single-tenant bare metal. Shared CPU and storage cap your connection concurrency and queue throughput, and a noisy neighbour can hurt the latency that deliverability depends on. We size dedicated hosts to your sending profile.
No. We provide and operate the servers; we do not resell PowerMTA licenses, and KumoMTA is free and open source. If a managed email service or a plain Postfix relay is the better fit for your volume, we will tell you so before you sign anything.
A standard warm-up runs three to six weeks, raising volume gradually so mailbox providers build a reputation for your IPs. The exact pace depends on your daily volume, list quality, and the providers you send to. We set the ramp schedule with you and watch the feedback loops as it runs.
Current bare metal runs on AMD EPYC Turin and Genoa processors with DDR5 memory and NVMe Gen5 storage, up to 192 cores per socket on the top tier. GPU nodes use NVIDIA Blackwell B200 and H200 cards with PCIe Gen5 lanes and local NVMe so the cards are not starved waiting for data.
Our network runs across carrier-neutral facilities in Amsterdam, Frankfurt, London, and Ashburn, with edge points for absorption in front. Email and compute share the same backbone, so a sending host and its analytics live close together.
Yes. We run NVIDIA Blackwell B200 and H200 nodes with PCIe Gen5 and local NVMe scratch for training, fine-tuning, and inference. For steady jobs, owning the metal usually costs less than renting cloud GPU hours once you compare a full month of runtime, and your data stays on hardware nobody else shares.
Volumetric traffic is absorbed and scrubbed at the network edge before it reaches your server, and that absorption is included on every machine rather than sold as an add-on. Your legitimate SMTP and API traffic keeps flowing while the attack traffic is dropped upstream.
Yes, and we run the old and new platforms side by side so nothing stops sending during the move. We carry over your domains, signing keys, and suppression lists, warm the new IPs while the old ones still carry production, and cut over only once the new reputation holds.
There is no single number, but the line usually appears somewhere between a few hundred thousand and a million messages a month, depending on how much control and reputation isolation you need. Below it, a managed ESP often wins on total cost once you price the operations time. Above it, or when you send on behalf of your own customers, owning the host and the IPs tends to pay for itself in both cost and deliverability. We work the math with your real numbers on the sizing call.
A short call, a sizing, and an honest answer — including when the answer is to spend less. No pressure, no resold licenses, and no commitment until you have seen the numbers for your own traffic.