Atlas R2 · now shipping · Horizon platform live in 31 regions

Enterprise compute, however you want to buy it.

Amazendev builds the servers that large companies run their heaviest workloads on — and operates the same platform as a service, in our cloud, metered by the hour. Rack it yourself, or consume it. One architecture, one support line, either way.

Explore Atlas servers See the Horizon platform

AMAZENDEV

ATLAS R2 · 2U

PWR

HLTH

UID

ALM

PWR

UID

USB

OCP

Hover a bay for live drive telemetry, or tab in and use the arrow keys · click to pin 2 × 64-core · 1.5 TB DDR5 · 24 × NVMe Gen5 · 2U

Two ways to buy the same thing

Own the iron. Or consume the platform.

Rack it yourself.

Atlas compute and Vault storage — dual-socket servers, dense compute, GPU nodes and petabyte arrays, built to a published bill of materials. Capex you depreciate, in your own racks, with a serial number you can look up.

— 5 server families, from a 1U front-end to an 8-GPU trainer — 4-hour part replacement in 41 countries — Two lines: Penang and Austin

Or let us run it.

Horizon is the same compute, data and security stack, operated as a service in 31 regions. Metered by the hour, backed by a 99.99% SLA, with capacity you can turn up in minutes and off just as fast.

— Provision through the console, the API or Terraform — 99.99% SLA, with the credits written into the contract — Your data stays in the region you pick — no exceptions

Same architecture, same images, same support desk — whether the machine is yours or ours.

Servers

The lineup, with the numbers that decide it.

Compare every model

Model	Class	Processors	Memory (max)	Storage / accelerators	Availability	Spec sheet
Atlas R22U · dual PSU	General compute	2 socket · up to 128 cores	6 TB · 24 DIMM	24 × NVMe Gen5	In stock	Spec sheet
Atlas C42U · 4 nodes	Dense compute	8 socket · up to 512 cores	4 × 3 TB	6 × NVMe per node	In stock	Spec sheet
Atlas G85U · liquid-ready	GPU / training	2 socket · up to 112 cores	4 TB · 32 DIMM	8 × GPU · NVLink	6 weeks	Spec sheet
Vault S14U · dual PSU	Storage server	1 socket · up to 48 cores	1 TB · 8 DIMM	90 × 3.5″ · 1.8 PB raw	In stock	Spec sheet
Atlas E11U · single socket	Front-end / edge	1 socket · up to 64 cores	2 TB · 16 DIMM	10 × NVMe Gen5	In stock	Spec sheet

Core counts are maximum supported per chassis and depend on the processor you configure. Memory maximums assume the largest qualified DIMM. Full options in each spec sheet.

Horizon platform · software as a service

The whole stack, delivered as a service.

Four services that run on our infrastructure, in the region you choose. No cluster to stand up, no version to chase — you provision what you need and pay for what you use.

Read the API reference

Compute

Horizon Compute

Virtual machines, bare metal and containers on the same Atlas hardware we sell — spun up through the console or the API, billed by the second.

99.99% SLA · 31 regionsPricing

Data

Horizon Data

Managed Postgres, object storage and a petabyte-scale warehouse. Backups, replicas and point-in-time restore are on by default.

99.99% SLA · 28 regionsPricing

Observability

Horizon Observe

Metrics, logs and traces from your fleet and ours, at one-second resolution, with the dashboards and alerts already written.

99.95% SLA · 31 regionsPricing

Security

Horizon Guard

Identity, secrets and key management, plus hardware-rooted attestation for every Atlas node you enrol — cloud or on-prem.

99.99% SLA · 31 regionsPricing

Services

Or we run it with you.

Buying the servers is the easy part. Our engineers size the fleet, move your workloads onto it, and stay on call once it's live.

Capacity planning and deployment

We size the fleet against your real workload, rack and cable it, image it, and hand you a cluster that is already running — whether the racks are yours or a slice of Horizon.

8–16 weeks
Fixed scope
Rack-ready handover

Managed operations

We run the fleet you bought. Firmware, patching, capacity and 24/7 on-call, using exactly the telemetry our own operators watch in Horizon Observe.

Rolling, 12-month
24/7 on-call
Named engineer

Cloud and data-center migration

Move off legacy servers, or shift the right workloads to the platform, without a maintenance window you cannot afford. Dual-run, cut over one rack at a time, decommission.

6–20 weeks
Zero-downtime plan
Rollback at every step

Solutions

Built for the places that cannot go dark.

All industries

ATLAS R2 · HORIZON

Financial services

Low-latency compute for risk and trading, and a region that keeps the data in-country.

VAULT S1 · GUARD

Healthcare

Petabyte imaging archives on-prem, and an accredited cloud for everything else.

ATLAS · SOVEREIGN CLOUD

Public sector

Air-gapped racks you attest, or an accredited region — same images, same tooling.

ATLAS G8 · HORIZON COMPUTE

AI and research

Eight-GPU trainers you own for the baseline, and burst capacity you rent by the hour.

HORIZON · ATLAS E1

Retail and e-commerce

Own the everyday load, scale into the platform for the peak you can't rack for.

ATLAS G8 · HORIZON DATA

Media and streaming

GPU encode farms and a warehouse that keeps up with the catalogue.

Proof

What the fleet actually does.

38,400

Servers under
management

99.98%

Horizon platform uptime,
2025

Cloud regions,
six continents

Countries with 4-hour
part replacement

We own the racks that run every day, and rent the platform for quarter-end. It's one vendor, one bill of materials and one support desk — so the baseline and the burst never argue about whose fault it is.

Priya Raghavan · VP Infrastructure, Kestrel Financial · 9,200 cores

Newsroom