200 Gbps guaranteed bandwidth

The Two Machines That Can Actually Fill 200Gbps

A 200Gbps port is 25 GB/s of payload crossing the bus. Only the 24-bay dual EPYC chassis have the PCIe lanes and the memory bandwidth to source it and store it at the same time.

X-ZoneServers 200 Gbps dedicated servers start at €12879 per month for a guaranteed 200 Gbps port — 2 bare-metal configurations with up to 64.8 PB of monthly transfer across 12 EU and US datacenter locations. The port is effectively unmetered: hold the full committed rate around the clock, with no overage charges and no fair-use throttling inside your tier.

12879
starting price per month
64.39
per Gbps per month
2
configurations
64.8 PB
transfer ceiling per month

A network port is the cheapest part of this configuration. The expensive part is everything standing behind it. Payload has to come off NVMe, cross the PCIe root complex, land in memory and leave again through the network card, and each of those hops is a budget that can run out long before the port does. That is why the line-up on this page is two machines long instead of eight. The Dual EPYC 9554 at €12879 and the Dual EPYC 9754 at €13439 are the only builds we offer with enough PCIe lanes and DDR5 memory channels to keep a 200Gbps port fed while the same silicon is also driving a full NVMe backplane. Everything else in the catalogue runs out of bus before it runs out of port.

Line Rate, in Payload Terms

At full utilisation the port passes 25000 MB/s. Sustain that across a 30-day month and 64.8 PB has left the chassis, which is a figure worth holding against your real egress before you shop at this level. In viewer terms it is 40000 concurrent 1080p streams at 5 Mbps, or 8000 concurrent 4K streams at 25 Mbps. A 1 TB dataset clears the port in 40 seconds. That last number is the one that reframes the whole build: at forty seconds a terabyte, the elapsed time of a transfer is decided by how many devices you can read from in parallel and how quickly memory can stage the result, not by the network. The port stops being the constraint, and the machine takes over as the constraint instead.

64.8 PB
Transfer per month
Sustained volume at full port utilisation across a 30-day month.
40,000
1080p viewers
Simultaneous streams at 5 Mbps each.
8,000
4K viewers
Simultaneous streams at 25 Mbps each.
40 seconds
Per terabyte
Wall-clock time to move 1 TB off the server at line rate.

Figures use decimal units (1 Gbps = 125 MB/s) across a 30-day month, at 5 Mbps per 1080p stream and 25 Mbps per 4K stream. They describe what the port can carry, not a guarantee of application throughput: disk, CPU and the far end all apply their own limits.

2 configurations at 200Gbps

Both configurations are dual-socket EPYC behind a 24-bay NVMe backplane, and that bay count is the visible end of the lane budget rather than a storage boast. Fill all twenty-four bays, then add a card fast enough for this tier, and you have committed more PCIe lanes than anything else in the catalogue carries. It is why the Xeon E5 platforms stop at 20Gbps, why the Gold 6148 goes no further than 100Gbps, and why even the 10-bay EPYC 9254 builds stop at 50Gbps. Between the two parts on this page, clock is the honest differentiator: the 9554 runs at 3.1GHz, the 9754 at 2.3GHz, both on DDR5-4800 across two sockets. Storage type and capacity, memory, bandwidth tier and operating system are all configurable, and anything outside the standard build is quoted on request.

Dual EPYC 9554

Ultimate Performance

12879
/month
128c / 256t @ 3.1GHz512GB DDR5 @ 4800MHz
24x NVMe Slots200Gbps Guaranteed
AI Optimized Routing Included
Order Now
ULTIMATE

Dual EPYC 9754

Maximum Performance

13439
/month
256c / 512t @ 2.3GHz1024GB DDR5 @ 4800MHz
24x NVMe Slots200Gbps Guaranteed
AI Optimized Routing Included
Order Now

What a 200Gbps server is for

Erasure-Coded Object Storage

Pools where a single read is fanned across the whole drive set. Storage and the network card draw on the same root complex, so the platform needs lanes held in reserve rather than lanes exactly counted.

In-Memory Analytics and Cache Tiers

Column stores, feature stores and hot caches that keep the working set resident in DDR5. Memory bandwidth answers the port directly here, so half a terabyte of RAM is the floor and not the option.

Streaming Origin and Packaging

Origin nodes that package, encrypt and push in the same pass. The CPU touches every byte on the way out, so cores and memory channels are spent alongside the port rather than after it.

Estate-Wide Backup and Replication

A landing zone for entire fleets checking in at once. Ingest is a write path, which puts the NVMe controllers and the inbound port inside the same PCIe budget at the same moment.

What the €560 Step Actually Buys

At entry the port works out at €64.39 per Gbps, against €44.99 per Gbps one tier down. That premium is not paid to the network. It is paid to the class of machine that can keep the network busy, because doubling the port means doubling the platform underneath it and there is no cheaper chassis that can stand in for the job. Inside the tier the arithmetic gets much friendlier. €560 a month separates the two configurations, and that single step takes you from 128 cores and 512GB of DDR5 to 256 cores and 1TB. Set against the monthly total it is a small move, so the real question is not whether the upgrade is affordable. It is whether your workload wants 128 fast cores or 256 patient ones, and whether the extra half terabyte of memory will actually be resident with data or simply sit there.

All guaranteed bandwidth tiers

The same hardware range is sold at nine committed port speeds. Compare the entry price of each.

200Gbps dedicated server questions

Why are only two servers offered at 200Gbps?
Because we will not sell a port that the machine behind it cannot use. The test is not whether a network card physically fits. It is whether the platform still has PCIe lanes and memory channels to spare once the drive bays are populated. Two builds in the catalogue pass that test, so two builds are listed. The rest hit their ceiling on the bus, not on the network.
What does the PCIe lane budget have to do with the number of NVMe bays?
Every NVMe drive draws lanes directly from the CPU, and so does a card of this speed. On a platform with a modest lane count you can have deep storage or a very fast port, but not both at full stretch at the same time. Dual EPYC gives enough headroom to populate the full backplane and still leave the network card an uncontended path, which is the entire reason this tier is restricted to that platform.
Which should I choose at this tier, the 9554 or the 9754?
Choose by thread behaviour rather than by price. The 9554 clocks higher, so request-latency work, single-threaded critical paths and per-core licensing all favour it. The 9754 doubles the thread count and doubles the memory ceiling, which suits container fleets, virtual machines and parallel batch work that scales sideways inside one box. Both carry the same port and the same backplane.
Does the memory subsystem matter as much as the lane count?
It matters just as much, and it is part of why both machines here are dual-socket. Payload rarely travels straight from drive to wire. It is buffered, checksummed and often encrypted, and every one of those passes crosses memory. A platform with too few DDR5 channels turns the memory controller into the choke point while the port and the drives both sit half idle.
What kind of operation needs 64.8 PB in a month?
Very few, and that is rather the point of quoting the figure. The realistic tenants are regional caching and streaming origins, research groups shifting instrument data between sites, backup providers absorbing a whole customer estate overnight, and platforms replicating object storage across data centres. Most of them run well under the ceiling and buy the tier for the peaks, not for the average.

200 Gbps dedicated servers by location

Same hardware and guaranteed port, deployed from the datacenter closest to your users.

Two Platforms, One Conversation

Tell us which of the two platforms fits your workload and which of the 12 locations you want it in, and it can be live in about an hour with up to 1Tbps of DDoS protection in front of it. Custom builds are quoted on request.

Up to 1Tbps DDoS protectionDeployed in about 1 hour99.9% uptime SLA