The battle for the definition of an AI PC is heating up in the PC market, with major players like Microsoft and OEMs like Intel and AMD all vying for dominance. It has become clear that running AI workloads locally is essential for sustainable growth, and the NPU has emerged as a popular choice for next-generation AI tasks due to its power efficiency. At its core, an NPU is a specialized accelerator designed specifically for AI workloads, offering low power consumption and high efficiency in performing AI tasks.
Companies like Qualcomm have been investing in improving their NPUs over the years, with Qualcomm’s Hexagon NPU offering 45 TOPS of AI performance. Intel and AMD have also made strides in NPU development, with the Intel Core Ultra model and the XDNA NPU in AMD’s Ryzen 8040 offering competitive AI performance. As more applications shift from the cloud to a hybrid model, there is a growing dependency on local NPUs to offload AI functions from the cloud, especially in areas like security.
The focus on NPUs in the AI PC landscape has intensified, with applications needing to take advantage of the NPU to justify its existence. The trend towards more on-device AI is expected to continue, with applications becoming more optimized and demanding higher TOPS from AI chips. Qualcomm’s Snapdragon X Elite is expected to lead the NPU TOPS race, with AMD and Intel also planning to deliver high levels of NPU performance in their processors. Developers can target a single NPU per vendor, with the help of low-level APIs and frameworks to optimize for NPU performance.
The move towards on-device AI will likely result in a shift from hybrid models to mostly on-device computing, making NPUs a crucial component of AI PCs. While some platforms may initially have less powerful NPUs, the importance of NPUs is expected to grow as companies strive to deliver the best possible AI PC experiences. The industry is still grappling with unknowns about the complete AI PC vision, but clarity is expected to emerge at Microsoft’s Build conference in late May. The battle for dominance in the AI PC market is likely to continue well into 2025 as chip vendors and OEMs push for faster and more capable NPUs.
