EarthFrame offers sovereign, energy‑efficient compute hardware ranging from portable workstations to modular 5‑node clusters, engineered for repairability and optimized performance‑per‑watt. Its open‑source software stack—including the WARPT CLI for real‑time power profiling and the Mar archive format for high‑ratio compression and selective redaction—enables research and cultural institutions to run AI workloads on locally owned hardware while maintaining full data sovereignty.
Funding
Funding not disclosed
Founders
Product
Problem
Organizations that steward large-scale scientific, cultural, linguistic, or environmental datasets face three intertwined challenges: excessive energy consumption of AI‑grade compute, limited control over where and how data is stored, and hardware that is difficult to repair or upgrade, leading to high total cost of ownership and environmental impact.
Solution
EarthFrame delivers a portfolio of sovereign, energy‑efficient computing solutions that let data custodians own and govern their own infrastructure. The hardware line spans backpack‑sized workstations (Ember) to modular 5‑node clusters (Mesa), all engineered for repairability and optimized for performance‑per‑watt. Complementary software includes the open‑source WARPT CLI, which provides fine‑grained monitoring of AI hardware configuration, power usage, and stress‑test results, while the Mar archive format offers high‑ratio ZSTD compression, random‑access reads, built‑in checksums, and selective redaction. By manufacturing in the United States or partnering with local assemblers, EarthFrame ensures a supply chain that supports on‑site maintenance and reduces transport emissions. The combined stack enables organizations to run AI workloads on sustainable hardware without sacrificing compute capability or data sovereignty.
Target Audience
Primary customers are research institutions, cultural heritage archives, environmental NGOs, and any organization that requires sustainable, sovereign compute for large‑scale AI or data‑intensive workloads.
Features
- Ember portable workstation (7 L chassis) with hot‑swappable components and low‑latency interconnects for field deployments
- 1 single‑node foundation and Mesa 5‑node 24U cluster, both featuring modular power supplies and serviceable boards
- WARPT CLI (Python) for real‑time power profiling, hardware inventory, stress‑test automation, and opt‑in local telemetry
- Mar archive format supporting ZSTD/Deflate compression, byte‑level random access, immutable checksums, and on‑the‑fly redaction
- Repair‑first design philosophy: standardized fasteners, documented service manuals, and a parts marketplace for rapid replacement
- Performance‑per‑watt optimization through custom firmware, dynamic voltage/frequency scaling, and workload‑aware scheduling
- Open‑source software stack with API hooks for integration into existing data pipelines and orchestration tools
- End‑to‑end data sovereignty guarantees: all data resides on customer‑owned hardware, with no mandatory cloud dependencies