Shipments of AI PCs—PCs equipped with the necessary hardware and software to run generative AI tasks locally—are expected to reach nearly 50 million units this year, but will swell to more than triple that number in just a few years.

Until recently, running artificial intelligence tasks on a local machine required the use of a CPU, GPU, or a combination of both. This approach is not ideal for several reasons: CPUs and GPUs are not optimized for AI tasks, so using them affects the overall performance of the system and drains the battery faster.

Neural processing units (NPUs) are now starting to make their way into consumer devices, specifically designed to process AI workloads as efficiently as possible. IDC outlines the following three categories of NPU-enabled AI PCs:

  • Hardware-enabled AI PCs include an NPU with performance of less than 40 tera operations per second (TOPS), typically enabling specific AI functions within an application to run locally. Currently, Qualcomm, Apple, AMD and Intel are all launching such chips.
  • The next-generation AI PC includes an NPU with 40-60TOPS performance and an AI-first operating system (OS) that enables persistent and pervasive AI capabilities in the operating system and applications. Qualcomm, AMD and Intel are all expected to launch future chips for this category, with deliveries likely to begin in 2024. Microsoft is also expected to roll out a major update to Windows 11 (and updated system specs) to take advantage of these high TOPSNPUs.
  • Advanced AIPC refers to PCs with NPU performance greater than 60TOPS. While no chip vendors have released such products yet, IDC expects them to appear within the next few years. This forecast from IDC does not include advanced AIPC, but they will be included in future updates.

    • IDC expects shipments of hardware-enabled AI PCs to grow significantly over the next two years, before the next generation of AI PCs takes over. The research firm believes that by 2027, next-generation AI PCs will ship twice as many as hardware-enabled AI PCs. Many of these systems will be sold to business customers, but consumers should have a lot to look forward to, too, including new digital content creation tools and AI-enhanced PC gaming.

    There are three main technical reasons for bringing AI workloads from the cloud to the client: improving performance by eliminating the round trip that current AI workloads must make over the network to the cloud; improving privacy and security by keeping data on the device rather than moving; and reducing costs by limiting access to expensive cloud resources.