Embedded vendors demo edge hardware

Most artificial intelligence still runs in giant data centers, but a growing slice is moving to the edge, which means the chip sits inside the camera, factory box, car, or local server that sees the data first. That cuts the round trip to the cloud the way doing math on a calculator beats mailing the problem to someone else. (st.com) That shift showed up all over embedded world 2026 in Nuremberg, where the trade show said about 36,000 people attended from March 10 to March 12, 2026. The floor was packed with companies trying to prove they can run sensing, security, and machine learning on smaller, closer, cheaper hardware. (embedded-world.de) STMicroelectronics came in from the tiny-device end of that market. Its own recap said it showed more than 50 live demonstrations across edge artificial intelligence, cybersecurity, radio connectivity, automotive systems, sensing, power, and the STM32 ecosystem. (st.com) STM32 is STMicroelectronics’ family of microcontrollers, which are the simple computers hidden inside thermostats, motor drives, and battery tools. STMicroelectronics said visitors could see complete chains from sensing to decision-making to actuation, which means the company was pitching not one chip but a full parts-and-software path from input to action. (st.com) Intel showed the same edge idea from the opposite direction: not a coin-sized controller, but a server chip for boxes that sit in telecom racks and factory rooms. Intel says its Xeon 6 system on chip is built for network and edge use, with built-in accelerators for virtualized radio access networks, media, artificial intelligence, and network security. (intel.com) A system on chip is a bundle of parts that used to be separate, like putting the engine, transmission, and electronics into one tighter package. Intel’s product brief says the Xeon 6 system on chip is a highly integrated, power-optimized server platform aimed at higher throughput in edge workloads. (intel.com) Intel’s pitch is that some artificial intelligence jobs cannot wait for a faraway cloud reply. In a March 2, 2026 fact sheet, Intel said the Xeon 6 system on chip enables immediate, real-time inference in live networks, which is telecom language for making split-second model decisions where the data is generated. (intel.com) Tenstorrent pushed the idea one step closer to an ordinary desk. Its QuietBox 2 workstation is listed at an expected retail price of US $9,999, is slated for the second quarter of 2026, and is sold as a local machine for running and testing advanced artificial intelligence models without relying on a cloud service. (tenstorrent.com) Tenstorrent says the QuietBox 2 has 128 gigabytes of graphics memory and can run models up to 120 billion parameters locally. In plain terms, that is the company betting some developers would rather keep experiments in-house than rent remote computing by the hour. (tenstorrent.com)