AMD Ryzen AI dedicated servers • Built to order in five locations

Two Ryzen AI machines, and nothing else on either

The Ryzen AI Max+ 395 at 126 TOPS across 16 cores, or the Ryzen AI 9 HX 370 with 128 GB of DDR5 behind it. Both single tenant, both with root on the metal and an unmetered gigabit port. From $249 a month, and that figure does not move.

Configure from $249/mo

Root on the metal Unmetered gigabit port Built to order in 4-24h No setup fee

16 cores 126 TOPS Radeon 8060S

Ryzen AI Max+ 395 — $399 a month

The faster of the two: sixteen Zen 5 cores, 126 TOPS across the NPU and the rest of the package, and the Radeon 8060S. 64 GB of DDR5 and a 1 TB NVMe come with it. Take this one when the model fits in 64 GB and you want it answering quickly.

12 cores 128 GB DDR5 Radeon 890M

Ryzen AI 9 HX 370 — $249 a month

The cheaper machine is the roomier one. Twelve cores, 80 TOPS with 50 of them on the NPU, and twice the memory of its faster sibling. Take this one when what you are loading is larger than 64 GB, which is the usual reason a run stops.

No setup fee Monthly terms 4h hardware swap

The price is the price

Billed monthly, cancellable monthly, with no setup charge and no introductory rate that lapses. Three, six and twelve-month cycles take 5, 10 and 15 percent off if you would rather commit. Basic management is included on both.

How the machine reaches you

We do not keep these on a shelf, and we would rather say so than pretend otherwise.

Built for the order

Each one is built for the order that paid for it, in whichever of our five halls you picked. The configurator quotes a 4 to 24 hour window and that is the window we work to.

Fedora 43 and root, nothing else

It arrives running a clean Fedora 43 with full root and no inference service sitting in front of it. What you run on it — vLLM, Ollama, llama.cpp, PyTorch pinned to the release your code was tested against — is yours to install and yours to version.

Your traffic is not metered

A gigabit port, unmetered. Not a large allowance, not a burst credit, not a fair-use figure that turns into an invoice. One IPv4 address comes with it and IPv6 is there for the asking.

What runs on them

An NPU and an integrated Radeon sharing one memory pool suit some jobs far better than a rented slice of a bigger box.

Extraction that cannot leave the building

Summarising contracts, pulling entities out of logs, classifying a support queue. The work is unglamorous and the reason it lands on a dedicated box is almost always that the corpus is not allowed near a shared endpoint.

Serving an AI model on your own hardware

Assistants, help-desk deflection, voice front-ends. One machine nobody else is scheduled onto, so the latency you measure on Tuesday is the latency you get on Friday.

Code completion, kept in-house

Code completion and review models serving a team, with the repository staying on hardware you rent outright. Several model versions can sit on the disk at once and you choose which is live.

Finding the surprises early

Getting a stack building and running against an NPU and ROCm before committing a budget to racks of anything. Cheaper to discover here what does not port.

Reference specification

The builds the configurator opens on. Everything here is included in the monthly figure.

Ryzen AI Max+ 395

Processor

AMD Ryzen AI Max+ 395, 16 cores

AI throughput

126 TOPS

Graphics

AMD Radeon 8060S

Memory

64 GB DDR5

Storage

1 TB NVMe SSD, one slot

Price

$399 per month, no setup fee

Ryzen AI 9 HX 370

Processor

AMD Ryzen AI 9 HX 370, 12 cores

AI throughput

80 TOPS, 50 of them on the NPU

Graphics

AMD Radeon 890M

Memory

128 GB DDR5

Storage

1 TB NVMe SSD, one slot

Price

$249 per month, no setup fee

Both machines

Network

1 Gbps unmetered, unshared

Addresses

1x IPv4 included, IPv6 available

Operating system

Fedora 43, preinstalled, full root

Storage options

1 TB, 2 TB or 3.84 TB NVMe

Management

Basic included: 12h response, 4h hardware replacement

Locations

New York, Miami, San Francisco, Bucharest, Amsterdam

Availability

Built to order, 4 to 24 hours

Prices are per month and exclude sales tax where it applies. Need more cores, more memory or a discrete card? Have one built to your specification.

What comes with the rental

The parts of the service that are included rather than sold back to you.

Basic management at no extra charge

Twelve-hour response and failed hardware replaced within four, on every machine at no extra charge. Bronze, Silver, Gold and Platinum tiers shorten the response to four hours, two, one and thirty minutes if you need it.

NVMe, not spinning disk

The included 1 TB is NVMe, so loading a checkpoint or streaming a training set is not the thing that holds the run up. 2 TB and 3.84 TB are available in the same single slot.

Engineers in the same hall

Engineers on live chat around the clock, working in the halls these machines sit in. Ask before you order if you want to know where the build queue actually stands.

Ryzen AI servers, asked and answered

The questions engineers actually send us before ordering one.

Which of the two should I take?

Memory decides it more often than throughput does. The HX 370 at $249 carries 128 GB, twice what the faster Max+ 395 has, so it is the one that will hold a model the 64 GB machine refuses. Take the Max+ 395 at $399 when what you are running fits in 64 GB and you want the 126 TOPS and the better Radeon behind it.

Is this a whole machine or a share of one?

A whole machine. One tenant, full root, nothing virtualised between you and the silicon. The NPU, the integrated Radeon and the memory are yours for the month and are not carved up for anybody else.

How much storage can I have?

These boards carry a single NVMe slot, and 1 TB is included in the price. You can take 2 TB or 3.84 TB instead at the configurator. There is no second slot, so if you need more spindles than that, or RAID, this is the wrong line and we will happily build you the right one.

Does it come with an inference stack installed?

No, and that is deliberate. It arrives as a clean Fedora 43 and gives you full root access to the hardware. Install vLLM, Ollama, llama.cpp, PyTorch or whatever else you have standardised on, pinned to the versions your code was tested against rather than the ones we happened to bake in.

How long does it take to get one?

We build each one for its own order rather than holding stock, and the configurator quotes a 4 to 24 hour window. If your timing is tight, ask us on chat before you order and we will tell you where the queue stands.

Where can I have it racked?

New York, Miami or San Francisco in the US, Bucharest or Amsterdam in the EU. The price is the same in all five, and you choose the hall at the configurator rather than being routed to whichever one has space.