Anchor builds and operates micro datacenters in the US and the Caribbean. We started with customer-premises equipment, and later built our own sites. We offer bare-metal high performance servers and clusters rental, and related services like batch compute.
We developed edge infrastructure in partnership with a renewable energy utility, to monetize low-cost and green power opportunities. We have successfully remotely operated sites for several years.
We are now exploring new monetization strategies like AI inference. Workloads like this open further opportunities for hardware specialization and location optimization.
Our micro datacenters are simple shelves that can be installed at any site. They only require low-power air conditionning. We do not use regular racks due to the format of our liquid cooling solution.
Power is distributed through a high voltage DC bus directly connected to standard solar battery racks. Power is converted to DC using high-power solar inverters. Our sites usually include solar arrays.
We can customize over the following dimentions:
Nodes are usually single CPU, and between 4 and 16 GPU. The cooling system and high voltage DC bus allow for up to 10kW per node.
All nodes are compute nodes, but carry entreprise NVMe drives. IO requests are sharded over NVMe-oF, using ConnectX offload when available to avoid jitter. That configuration allows line rate throughput and optimal hardware utilisation.
Our micro datacenters are single tenant, so we only offer basic networking services and monitoring like VPN and load balancers. The internal network is usually Infiniband, and services configured over rdma.
We can customize both the PCIe and Infiniband topologies to achieve the best result for a given workload. We developed a custom PCIe switch board to build arbitrary cabled trees for each node. Infiniband nativelly supports various network shapes between nodes.
We serve an OpenAI compatible API at api.anchor.tech
Please contact us to negociate models and options, and get an API key.
We can serve any open source model. We currently optimize for the main token markets.
Use a standard AI key, e.g.
curl https://<endpoint>/v1/models -H "x-API-key: xxxxxxxxxxxxxxxxxxxxxxxxxxx"
We plan to expand our inference solution. Running custom models for our customers. Explore models where our platform has a particularly competitive edge in performace or price.
We are looking for long-term customer relationships around high value services and hard chalenges like training and fine tuning, large data management, and applied use case like industry settings.
contact@anchor.tech
2093 Philadelphia Pike 5756 Claymont, DE 19703-2424 United States