We are seeing the most standardized specifications of AI hardware with the launch of AI PCs, and the operating system and the apps will be key to defining the "AI" on the PCs as a genie or a white elephant, said DIGITIMES Research analyst Joyce Chen, who just recently released a report regarding AI PC.
AI PCs and AI phones are the edge AI solutions to provide computing performance off the cloud to address consumers' concerns of information security and off-line limitations of generative AI.
"Now, with all specifications extremely standardized, squeezing the highest possible computing power into limited physical space, there is not much room left for differentiation," said Joyce Chen. "Hardware's work is largely done, and now it is for the software to define the jobs that consumers want to be done, to optimize the algorithm and power consumption efficiency."
Perhaps that is why after Intel and AMD announced their AI PC CPUs, all eyes are on how Microsoft's Windows 11 will integrate Copilot AI assistant.
"About 70% of the laptop and desktop PCs are using Microsoft Windows, that is why Windows 11 and Windows 12 are receiving so much attention," said Chen.
Tasks to be done
Since OpenAI launched the ChatGPT service, generative AI has been seen as an inevitable trend to start a new industrial revolution. Various tech companies have begun to operate their large language models (LLMs) on the cloud. As consumers access the internet to get solutions from the cloud, the computing power needed for the tasks is all concentrated in the data centers.
Although Qualcomm coined the term 'AI on device' to offload AI computing to devices such as PCs or smartphones, such devices have already been doing AI tasks before the concept is further materialized by having NPU, CPU, GPU, and memory together to process data on edge.
"The devices, after throwing the instructions to the cloud for AI tasks, they still have to get the answers back and store the data on the edge," said Chen. "However, corporate customers, who are the largest users of cloud AI services, are worried about trade secret and privacy leaks and information security issues."
Another problem is internet connectivity. Users cannot access cloud AI services in the basement, in offices that deliberately cut off internet access for security reasons, or in rooms where internet connections are poor. That is why companies start to see business opportunities in launching AI PCs.
"Having generative AI function performed on edge without having to be connected to the internet at all times is a goal of the AI PC, but inevitably will continue to collaborate with the cloud, and that hybrid AI mode is expected to be the mainstream," said Chen.
Tech companies are still trying to devise AI applications that can be applied to edge devices to optimize user experiences. But there are plenty of viable examples. "Take translation, for example, many people would have ChatGPT to translate some contents, but when the answers came back from the cloud, you would have to wait for it to spill out the words," said Chen. "When it is done on edge, the speed may quicken and also more precisely trained by machine learning to meet the need of this one specific master," said Chen.
Some of the AI functions may be performed in a not-so-obvious way. AI can help optimize performance. For example, when doing image processing, AI can boost the performance of GPU while lowering the activity of other irrelevant components. "AI can also manage power consumption efficiency to optimize the best allocation of energy in the device," said Chen.
With PC inventory correction ending, original design manufacturers (ODMs) and semiconductor vendors, such as Intel and AMD, are all hinging high hopes for AI PCs to stimulate a new wave of PC replacement consumption. But since hardware specifications are now very much determined, the software and applications will become the determining factors of AI PCs becoming long-lasting products that generate endless business opportunities or just a one-time fad.
About the analyst:
Joyce Chen is a DIGITIMES Research analyst and project manager of the semiconductor team. She specializes in artificial intelligence (AI) cloud platform applications and IoT applications.