Challenges for AI PCs are just about to start

Chloe Liao, Taipei; Jack Wu, DIGITIMES Asia 0


Barring any unexpected developments, AI PCs are expected to be the central focus of COMPUTEX 2024. It will also set off the first wave of competition in AI end-user products in 2024.

On May 16, Asus Co-CEO Samson Hu, Pegatron Chairman T.H. Tung, and several industry leaders participated in the "Pre-COMPUTEX Event Series - AI PC Industry Exploration Forum." During the forum, they shared numerous insights on the future development of AI PCs.

Both Tung and Hu expressed during the event that AI PCs will become so ubiquitous to the point where it's a standard feature just like Wi-Fi. It will be so commonplace that people will gradually stop deliberately mentioning it as something notable in the future.3

Regarding the AI PC trend, Hu stated that the development of AI PCs is still in its early stages, with 2024 being the inaugural year. Right now, it is merely at the starting point. There is still a lot of room for development and many issues that need the joint attention of various sectors.

Tung agreed with Hu's comments, predicting that 2024 will be a lively year. AI has offered the "old" PC industry an opportunity to rejuvenate but it also brings challenges. Precisely because AI PCs are going to be so commonplace, companies can't just be talking about big words and concepts; the key is to have tangible implementations.

Despite AI PCs still being in the early stages of development, Quanta SVP and Quanta Cloud Technology (QCT) GM Mike Yang believes "the development speed will be very fast." This is mainly because software and hardware integration is in place, with things like large models for training AI already in a mature development state.

Secondly, essential technologies such as edge computing and 5G communication are already in place. Regarding the data, there is a wealth of data generated by public clouds and IoT databases to feed AI. In terms of key hardware support, semiconductor advancements are accelerating as well.

Yang also emphasized that the AI PC ecosystem is built on integrating software and hardware. So far, it appears that everything is well-prepared. He highlighted that while Taiwan is in a leading position in key AI PC infrastructure, future development will hinge on applications, especially various vertical industry applications.

Hu pointed out that shipment volume is neither the key focus nor the biggest hurdle in AI PC development. The challenges posed by different environments and usage scenarios are where the true test begins.

Hu also summarized five key issues for the future development of AI PCs. Firstly, hardware architecture needs to continue to evolve. As application scenarios expand, performing inference with AI at the edge will require even more computing power. Even Asus's upcoming AI PC, which has more than 40 TOPS of computing power, will not be sufficient. Improving and upgrading hardware will be an industry-wide effort from both upstream and downstream.

Secondly, Large Language Models (LLMs) are developing rapidly. However, given the trend toward lighter and thinner end-user devices, refining all kinds of LLMs to be lightweight models so they can occupy a smaller space in end devices while maintaining their intelligence will require ongoing effort.

Thirdly, collaboration with the cloud is crucial. Hu admitted that while there is great enthusiasm for AI development at the edge right now, as AI becomes more widespread, the applications will also become more powerful and complex. Once the complexity reaches a certain threshold, running solely on the edge will be difficult. Ultimately, it will have to return to the cloud. Therefore, figuring out how the edge and cloud will collaborate and how relevant algorithms will be executed is important.

Fourthly, establish an AI application ecosystem. Hu stated that over the past 40 years of development, the PC industry has standardized many software and hardware. In the upcoming AI era, a new ecosystem must be built. A development ecosystem across different platforms will attract more Independent Software Vendors (ISVs) to accelerate industry development.

Finally, user education and promotion are vital. He mentioned that as AI becomes ubiquitous, it will reach a broader user base. Previously, consumer electronics were likely used most by younger people. In the future, even the elderly will be part of the user base. Thus, AI education for the general public should not be overlooked.

In addition, Tung mentioned that beyond discussing the ecosystem and end-user experience, it is important to consider societal, national, and legal aspects as well. He stressed that issues regarding AI privacy and ethics will require new discussions and ideas to avoid challenges to social security systems.