Jobs keep running when a node fails.
Aleria Cloud runs GPU and CPU clusters as one pool. When a node fails, its jobs move to another.
Where every request is computed.
Aleria builds and runs AI data centers on NVIDIA GPUs, and Aleria Cloud walls each organization off from the rest. You get the compute without running a data center. Or we install the same stack in your own building, offline if you need it.
01The campus
An AI factory, from the grid to the chip.
Power comes in through our own substation, with battery storage and standby generators behind it. Heat leaves through fields of dry coolers, without cooling towers. Two fibre paths enter from different sides.
02The hall
Thousands of GPUs, run as one cloud.
Liquid-cooled racks in sealed pods, with coolant and fibre overhead and a clean floor below. Aleria Cloud schedules every job across them and serves every model through the Inference Gateway.
03One rack
Your tenant, walled off from the rest.
Compute trays, the NVLink switch spine and power shelves, cooled by liquid with no fans. Each organization gets its own network, keys and logs; the same software runs on your hardware.
The software that turns GPUs into one cloud.
Aleria Cloud runs on top of the data center. It schedules GPUs and CPUs, isolates every tenant, and deploys the rest of the stack on top. The same software installs on your hardware.
Aleria OSapplications
Aleria Fusionyour data
Inference Gatewaymodels
Aleria Cloudorchestration · isolation
Our data centers or your hardwareNVIDIA GPUs
Each organization runs on its own.
Each organization gets its own network, its own keys and its own logs, and shares no compute with another.
It runs where you choose.
In our data centers, on your own hardware, or air-gapped, with signed updates brought in offline.
The rest of the stack runs on it.
The Inference Gateway, Aleria Fusion and Aleria OS deploy on Aleria Cloud, in that order.
NVIDIA makes the chips. Partners bring the power. We run the rest.
As an NVIDIA Cloud Partner, we choose and deploy the chips; power comes from partners on each site. Everything above them, from the data center to the chat, we build and run ourselves. Each country's stack runs separately.
We run the NVIDIA systems ourselves.
NVIDIA DGX systems with DDN storage, in AI data centers designed after NVIDIA's reference designs. Aleria is an NVIDIA Cloud Partner (NCP).
Each site has its own power.
Grid power comes in through our own substation, with battery storage and standby generators on site. On request, we report the energy each workload uses.
Every country needs AI inside its own border.
We build the whole stack once and run it inside each border: the site, the capacity, the models in the national languages, and the apps every agency uses. Start on our capacity now, then bring it home.