The future of AI is at the edge
The large models released by DeepSeek in early 2025 have remained hugely popular, reigniting global enthusiasm for artificial intelligence (AI). Deep learning models are expanding exponentially with constant breakthroughs in performance at increasingly lower costs, thus disrupting all sectors. How to successfully migrate these powerful AI capabilities from the cloud to actual applications, especially in resource-constrained edge environments, has become a top priority for technology leaders worldwide.
Edge Computing: The Golden Track for AI Applications
Edge computing is a disruptive paradigm that directly deploys AI algorithms on edge nodes, such as IoT devices, sensors and servers, instead of relying on the cloud. This "proximity" enables real-time responses, greatly reduces latency, strengthens data security, and significantly reduces bandwidth consumption. From millisecond-level decision-making in autonomous driving to the detailed management of smart cities, and even intelligent inspections in industrial automation, the application scenarios of edge computing are expanding at an unprecedented rate.
Allied Market Research recently reported that the global AI edge computing market reached a value of US$9.09 billion in 2020, and is expected to soar to US$59.63 billion by 2030, with a compound annual growth rate of 21.2%. These statistics confirm that the era of edge AI will be arriving sooner rather than later.
Against the backdrop of the rapidly developing edge computing market, hardware platforms have also taken a leap toward AI. Avnet's MaaXBoard OSM93 is renowned for its excellent hardware settings, making it an ideal choice for developing high-performance edge AI solutions.
MaaXBoard OSM93: Powering Edge AI Applications
MaaXBoard OSM93 is powered by NXP’s i.MX 93 SoC, which integrates the dual-core Arm® Cortex®-A55 (up to 1.7 GHz) and Arm® Ethos-U65 NPU. This dynamic duo is not only able to smoothly perform general computing tasks, but also raises the inference speed of machine learning (ML) and artificial intelligence (AI) models to new heights through dedicated AI accelerators. The 1 GHz Ethos-U65 NPU, with its large-scale parallel computing capabilities, gives edge devices the "superpower" of real-time AI processing.
MaaXBoard OSM93
In addition, the board has ample memory, including 2 GB LPDDR4 SDRAM, 16 GB eMMC 5.1 flash memory, and 16 MB QSPI NOR flash memory. The OSM is soldered onto the main SBC carrier board and provides additional interfaces for three USB 2.0 ports, dual 1 Gbps Ethernet, MIPI-DSI, MIPI-CSI, and a standard 40-pin header with UART, SPI, I2C, I2S, and GPIO signals. The carrier board also includes interfaces for JTAG, debugging, ADC, CAN-FD, SAI digital audio, two PDM microphones, and an M.2 connector to support connection to a Wi-Fi/BT.802.15.4 wireless module.
Aligned with the growing emphasis on data privacy and device security, the EdgeLock security enclave of MaaXBoard OSM93 builds an indestructible line of defense for AI tasks. The trusted hardware quarantine effectively prevents AI data leakage and malicious attacks, ensuring the safe and reliable operation of equipment in complex environments.
The Path to Energy-efficient Edge AI with a Lower Carbon Footprint
Efficiency and performance are equally important in real-world edge AI applications. A successful application requires the ability to keep devices "cool" and "durable" while performing high-intensity computing tasks.
Gartner's "Top 10 Strategic Technology Trends for 2025" pointed out that carbon footprint has become a critical issue for IT organizations. Computing-intensive applications, such as AI training, simulation, optimization and media rendering, are becoming significant sources of corporate carbon emissions. Gartner predicts that we will see explosive growth in low-power computing technologies, such as optical computing, neuromorphic computing, and new accelerators, by the end of the 2020s.
For example, the MaaXBoard OSM93 is equipped with NXP's PCA9451 power management chip and adopts the Energy Flex architecture. This enables detailed control of power consumption in different processing domains and achieves the perfect balance between performance and energy consumption. The frequency and voltage can be dynamically adjusted based on task load of the application domain (Cortex-A55), real-time domain (Cortex-M33), and Flex domain (Ethos-U65 NPU), eliminating resource waste and maximizing energy efficiency. This detailed power management not only extends battery life and reduces the risk of device overheating, but also significantly reduces overall power consumption.
Conclusion
As AI technology gains traction and the concept of green computing is embraced, edge devices will become an increasingly powerful force driving the development of a more intelligent world. They will provide continuous AI services for a wide range of industries through efficient power management and AI algorithms, thus freeing corporations from a reliance on traditional cloud computing while reducing their carbon footprints. The future of AI is at the edge; and the edge is at the forefront of AI.
