It is a matte black cube, about the size of a bar of soap, and it appears almost comically small as it sits on a desk. Two inches deep, two inches tall, and two inches wide. Not a fan vent. There are no rotating drives. Absolutely no sound. The CuBox is a cube-shaped device that weighs only 91 grams and has dimensions of about 2 by 2 by 2 inches. It was created by the Israeli company SolidRun. When you pick it up, it feels like a prop that was abandoned after a product pitch by an industrial designer. The prop suddenly doesn’t seem so tiny when someone tells you it’s managing real-time video analytics inside a Boeing facility or operating the edge node for a 5G tower in Ohio.
Based in the ancient Mediterranean port city of Acre, Israel, SolidRun manufactures embedded systems components such as computer-on-module devices, single-board computers, and mini computers. The place seems almost purposefully hidden. Tel Aviv’s startup corridor is not Acre. It’s not a location that appears in breathless profiles of the newest competitors in Silicon Valley. Nevertheless, SolidRun has quietly and unobtrusively created something that a startling number of significant American institutions rely on. Boeing, Google, Ericsson, Dell, and Sony are among its clients. That is not a business that operates on the periphery of its sector.
Based on the Marvell Armada 500-series system-on-chip, the first-generation CuBox was reportedly the world’s smallest desktop computer at the time, according to SolidRun. That assertion, made in 2011, sounds like the arrogance of a startup, the kind of slogan that every small hardware company uses in its initial press releases. However, the timing proved to be crucial, and the device did support it. At the time, edge computing hardly had a name. Conceptually, the notion that data processing might have to take place closer to the source—at the factory floor, the traffic intersection, or the hospital bed—was still developing. SolidRun was producing hardware for an undeveloped market.
As edge nodes for automation systems, real-time video analytics, IoT gateways, and AI inference workloads, fanless PCs and industrial PCs are being used more and more. In environments that are vibration-prone, dust-rich, and thermally constrained, where active cooling is either impractical or prone to failure, these systems must maintain deterministic performance. Although that description sounds clinical, there are important practical ramifications. A spinning fan is an inevitable mechanical breakdown. You can replace it in a data center. You cannot be bolted to a wall inside an industrial freezer or inside a sealed kiosk at a transit hub. It’s the whole point of the design to have no moving parts.
System-on-modules, single-board computers, and industrial mini-PCs are just a few of the platforms that SolidRun’s compact embedded solutions, which are based on ARM and x86 architecture, offer. The company also provides a full range of services, including hardware customization, software support, product branding, and enclosure design. It may seem insignificant, but that final section is crucial. Many businesses in this industry sell boards or chips and let others handle integration. One reason SolidRun keeps showing up in deployments where reliability is crucial is because of its methodology, which handles the entire stack, from silicon to enclosure. Instead of putting something together yourself, you will receive a completed, tested product that is about the size of a Rubik’s cube.
This is part of a larger story about Israel’s technology ecosystem and how it continues to produce software and hardware that is integrated, frequently undetectably, into American infrastructure. According to Dr. Atai Ziv, CEO of SolidRun, edge AI is leaving controlled settings and entering the real world, including mobile platforms, challenging industrial sites, and mission-critical deployments where longevity and serviceability are just as important as raw compute. Something genuine is captured by that framing. Proof-of-concept deployments in friendly environments dominated the early years of edge computing. The current situation is much less forgiving, messier, and more demanding.
With the same 50x50x50mm form factor, the CuBox-M integrates NXP’s iMX8M Plus processor and adds 2.3 teraops of neural network acceleration. It can be deployed with a single cable thanks to its integrated Power over Ethernet support and high-speed wireless connectivity. The final detail, one cable, may seem insignificant until you’re in charge of thousands of scattered nodes and each one needs to be physically maintained. At scale, significant operational savings result from hardware simplicity.
It’s difficult to ignore the fact that the most important technology in American infrastructure is frequently the least obvious. Not the headlines about billion-dollar data centers or AI supercomputers, but the quiet little boxes mounted beneath server racks and hidden behind panels that process data before it ever makes it to the cloud. In many respects, SolidRun’s CuBox is the ideal representation of that invisible layer—something truly necessary that the majority of those who rely on it will never see, much less acknowledge. The most intriguing question surrounding it at the moment is probably whether the company stays in that quiet position as edge computing becomes more prominent and contentious. The cube survives for the time being.
