GigaIO, a specialist in composable infrastructure, has announced the launch of Gryf, a portable AI supercomputer designed to deliver data center-level performance directly to the edge.

Co-developed with SourceCode, Gryf is described as “the world’s first suitcase-sized AI supercomputer,” enabling organizations to perform real-time analytics without depending on remote processing. Built entirely in the United States, the system has already attracted “significant orders” from the U.S. Department of Defense and the intelligence community, according to the company.

GigaIO is also targeting sectors such as healthcare, scientific research, manufacturing, oil and gas exploration, and sports analytics.

Gryf leverages GigaIO’s proprietary FabreX AI memory fabric, offering up to 30 teraflops of FP64 performance and supporting Nvidia’s H100 NVL and H200 NVL GPUs. Bandwidth capabilities reach 3.9 Tbps and 4.8 Tbps, respectively.

The system integrates compute, storage, and networking sleds in a ruggedized, field-ready chassis, allowing critical data to be processed on-site and minimizing latency from off-site transfers. Up to five Gryf units can be stacked and interconnected via FabreX, enabling shared resource access across servers as if operating on a single node.

“Gryf represents a fundamental shift in how organizations access and utilize high-performance computing at the edge,” said Alan Benjamin, Chief Executive Officer of GigaIO. “By bringing supercomputing capabilities to field operations in a portable form factor, we’re enabling real-time intelligence and analytics that were previously impossible without massive infrastructure.”

The strong early demand across defense, energy, sports, and media markets reflects growing interest in edge-based AI deployments, Benjamin said.

The rise of portable AI supercomputers reflects a broader shift toward decentralizing compute power, allowing organizations to process sensitive data closer to the source while reducing dependency on centralized infrastructure.