AI inference near your data.Models next to the data they score.
Run AI inference for recommendation, vision, language and scoring services on single-tenant clusters near your users. Compute nodes are sized to your models, with GPU nodes where needed and fast storage for model files.
AI inference where the requests start.Score data without shipping it far.
Training happens once. Inference happens every time a user clicks, a camera frame arrives or a transaction needs a score. Those requests are sensitive to distance, so we place inference clusters close to the users and data they serve.
Your models, their files and the data they read share one private switching fabric. Requests stay on single-tenant hardware instead of crossing a provider's network.
What is inside the rack.Sized to your models.
- Compute
- Compute nodes sized to the models you run.
- Acceleration
- GPU nodes where your models need them, alongside CPU nodes for everything else.
- Model storage
- Fast storage for model files, pooled by Ceph with three copies of every block.
- Network
- 25/100 GbE links to every node on your own private switching.
- Management
- Out-of-band access for remote console and power control, staffed by U.S.-based engineers.
AI inference at the edge.Metro edge or your colocation space.
Edge clusters put hyperconverged compute and storage in a metro data center near users and devices. We run single-tenant racks in Chicago, Ashburn and Los Angeles. We can also deploy in the colocation facility you already use, then manage it around the clock through out-of-band access.
Every site gets the same cluster design, out-of-band management and Ceph self-healing storage, so a model that runs in one metro runs the same way in the next. The same U.S.-based engineers support every location, 24/7/365.
Inference services we build for.Common model types.
Recommendation engines
Score products and content for users from a cluster in their region.
Vision
Process camera and sensor streams near the source and send results upstream instead of raw data.
Language
Serve language models on single-tenant nodes with GPU capacity where needed.
Scoring services
Run risk and fraud scores next to the transactions they evaluate.
Questions, answered.
Do I need GPUs for inference?
Not always. Some models run well on CPU compute nodes. We add GPU nodes where your models need them and size the rest during the design session.
What does AI inference at the edge mean here?
It means running models in a metro data center near your users and devices, or in your colocation facility. We do not deploy on customer premises.
How is inference priced?
Each rack build has one fixed monthly price, quoted after a free design session. There are no egress fees or per-request storage charges.
Can I run inference in more than one metro?
Yes. We can build the same cluster design in Chicago, Ashburn and Los Angeles, so each region scores requests from a nearby location.
Keep exploring.
Talk to the engineers who will build it.
Tell us what you run today and where you’re headed. We’ll come back with a design and a fixed monthly price.