Atlas in a hardened 10′ x 8′ x 8.5′ ISO container: flush roof and side radiators joined by a manifold, dual equipment doors with a hydrogen vent, ISO corner castings, and an integrated private 5G mast.
A ruggedized, immersion-cooled AI compute platform integrated inside a hardened 10-foot ISO container. Data center-class GPU compute, enterprise storage, a full Cisco fabric, and private 5G, powered by hydrogen and operational within 72 hours. Anywhere the cloud is too far, too slow, or too risky, Atlas brings the data center to the workload.
Atlas in a hardened 10′ x 8′ x 8.5′ ISO container: flush roof and side radiators joined by a manifold, dual equipment doors with a hydrogen vent, ISO corner castings, and an integrated private 5G mast.
| Form Factor | Hardened 10-foot ISO shipping container, approximately 3 m x 3 m x 2.4 m |
| Transport | C-130 aircraft, CH-47 helicopter, standard flatbed truck, ISO container ship |
| Time to Operational | 72 hours from arrival, not weeks |
| Operating Environment | -40°C to +50°C ambient, 0 to 100 percent humidity, up to 15,000 ft altitude |
| Primary Power | Hydrogen PEM fuel cell, 10 kW scalable to 50 kW, NATO NSPA qualified |
| Off-Grid Runtime | Multi-day continuous operation on internal hydrogen storage, extendable with refueling or an external H2 skid |
| Alternate Power | Diesel generator, shore power, utility tie-in, with automatic source transfer |
| Cooling | Single-phase dielectric immersion; full submersion of compute and production networking |
| Connectivity | Integrated private 5G (CBRS band, 1 to 5 km range, 1,000+ devices), Wi-Fi 6E, wired WAN, optional satellite uplink |
| Acoustic Signature | Near-silent (no server fans); reduced visual and thermal signature in tactical use |
Every Atlas configuration shares a common platform: Cisco networking, VAST Data storage, private 5G, hydrogen fuel cell, and full immersion cooling. Choose the GPU tier that fits the mission.
All configurations: 300 TB–2.4 PB VAST Data AI OS storage • Cisco enterprise networking • Private 5G AI-RAN • Hydrogen fuel cell • Full immersion cooling • 10’ ISO container. Off-grid runtime on internal H2 storage at sustained load; extendable with refueling or an external skid.
NVIDIA Blackwell-class accelerators with a documented upgrade path to B300 and Rubin generations. DGX Spark nodes handle AI-RAN baseband and edge inference on enterprise server platforms.
VAST Data AI OS with simultaneous NFS, SMB, and S3 presentation. Inline AES-256 encryption, dedup, and compression. 300 TB to 2.4 PB usable per unit, scaling independently of GPU tier.
Cisco Nexus data center fabric with VXLAN/EVPN multi-tenant segmentation, MACsec on every internal link, Firepower NGFW, and a dry out-of-band management path for remote serviceability.
O-RAN compliant radios in CBRS band with GPU-accelerated NVIDIA Aerial baseband. SIM-based authentication, per-mission network slicing, and 1,000+ concurrent devices across a 1 to 5 km radius.
Single-phase dielectric immersion with all compute and production networking fully submerged. No fans, no particulate intrusion, continuous duty from -40°C to +50°C ambient.
Hydrogen PEM fuel cell primary with independent management-plane backup, plus diesel, utility, and shore power alternates behind automatic source transfer and UPS ride-through.
Workloads built for the cloud run on Atlas without rewrite.
| Layer | Components |
|---|---|
| Orchestration | Kubernetes (RKE2 or OpenShift), NVIDIA GPU Operator, Istio service mesh |
| AI & ML | NVIDIA AI Enterprise, PyTorch, TensorRT, Triton Inference Server, RAPIDS, NVIDIA NIM microservices |
| Data | Apache Kafka, Redis, PostgreSQL, MinIO (S3-compatible) |
| Storage Services | VAST Data AI OS, snapshot and replication |
| Management | Cisco Intersight (local appliance for disconnected mode), Prometheus, Grafana, NVIDIA DCGM |
| Security | MACsec on fabric, IPsec or TLS on WAN, LUKS volume encryption, HSM-backed key management, STIG-hardened OS |
| Standard | Applicability |
|---|---|
| MIL-STD-810H | Environmental qualification: temperature, humidity, shock, vibration, altitude, dust, rain |
| MIL-STD-461G | EMI/EMC and EMP shielding |
| DoD STIG & RMF | Security hardening and Authority to Operate pathway |
| FIPS 140-3 | Cryptographic module validation for storage and key management |
| NIST 800-53 / 800-171 | Federal control mapping and CUI handling |
| HIPAA Security Rule | Safeguards for ePHI in clinical deployments |
| NERC CIP | Critical infrastructure protection for electric utilities |
| IEC 62443 / API 1164 | Industrial control system and pipeline SCADA security |
| O-RAN / IEEE 1588 | 5G radio access compliance and precision fronthaul timing |
| NATO NSPA | Fuel cell qualification for allied logistics interoperability |
Self-contained deployable units in three GPU tiers, from multi-tenant inference and digital twin to foundation model training.
Fixed-site variant on Cisco UCS, VAST Data, and IGEL with two-site data mobility and a Blackwell AI expansion path, for operators with existing data center space.
Two to N units federated over MACsec-encrypted dark fiber or microwave, presenting as one logical fabric with per-site data residency.
Turnkey delivery with annual support: monitoring, updates, and AI model lifecycle management for customers without on-site infrastructure teams.