The personal computer (PC) industry is heating up and ushering in a new sweet spot: PCs with artificial intelligence (AI) capabilities. The main beneficiaries of this wave of PC revolution, including Intel and Microsoft, are actively promoting PCs equipped with AI CPUs and AI software assistants, and gradually moving AI applications from the cloud to the PC field.
In other words, AI PCs embedded with specialized chips can execute AI models locally without relying on the cloud. Intel CEO Pat Gelsinger believes that this will make AI services cheaper, faster, and more secure than using cloud-centric data center services. He said at CES 2024 in Las Vegas, USA, at the beginning of the year: "In the future, you will unleash this power for everyone, every use case, and every location. â€
While competing with AI giant Nvidia in the server field, Intel clearly sees an opportunity to catch up in its own strength, the "PC processor" field. Intel is currently actively integrating its Neural Processor Units (NPUs) into PC processors; NPUs are specialized semiconductors specifically designed to handle AI tasks.
Intel's Meteor Lake laptop CPUs have an NPU built into their CPUs to support third-party AI software features. Its old rival in the PC hardware field, AMD, is also launching AI PC processors. Immediately afterwards, Nvidia announced three new products at its online event ahead of CES 2024, including GPUs – RTX 4060 Super, RTX 4070 Ti Super, and RTX 4080 Super for laptops with AI capabilities.
Figure 1: Meteor Lake CPUs integrate an NPU to support AI applications. (Source: Intel)
In addition to AI-enabled processors, memory chip manufacturers such as Micron, Samsung, and SK Hynix are also focusing on AI PCs, hoping to enable AI accelerators to perform powerful auxiliary functions on PCs. Currently, new laptops come with up to 8MB of RAM, while Windows-based AI PCs are likely to double the capacity. In fact, large language models (LLMs) that execute AI assistants may require more than 16MB of storage.
Taking the Llama 2 series AI model created by Meta as an example, its general variant requires nearly 30GB of storage. In addition, with the advent of more powerful AI accelerators and processors, the storage capacity of AI PCs may have to be increased.
Figure 2: Copilot AI assistant is built on OpenAI's GPT-4 model (Source: Microsoft)
At present, consumers' enthusiasm for AI PCs can be said to be just in the "warm-up" stage, and it will take some time to be able to provide more native applications for AI PCs. However, as time progresses, AI PCs will become more powerful in both hardware and software, which is great news for semiconductor components such as CPUs, GPUs, and DRAM.