How Much Credit Does AI PC Deserve for the PC Revival in 2024?
-
At this year's 'Tech Spring Festival Gala' CES, AI PC undoubtedly took center stage. Both chip manufacturers and PC brands have set their sights on this field, making it one of the hottest concepts in the tech world overnight.
In fact, in the second half of last year, several PC brands released or previewed their so-called AI PC products. Microsoft even decided to modify the keyboard layout of new PCs by introducing a physical key for the AI assistant Copilot on Windows 11-equipped computers, with the first batch of devices set to hit the market this month. It's worth noting that the last major change to the Windows PC keyboard layout was 30 years ago.
Additionally, many institutions have made optimistic predictions. Canalys' latest report indicates that AI-compatible PCs will account for 19% of the global market share in 2024, growing rapidly to 37% by 2025. According to Sigmaintell's latest forecast, global shipments of AI PC devices are expected to reach approximately 13 million units in 2024, achieving large-scale distribution. Both the recognized technological advancements and the positive attitude of the industry seem to suggest that the PC industry in 2024 will see a turnaround. However, this iteration, primarily driven by the supply chain, raises thought-provoking questions: How exactly will AI revolutionize PCs? And what are the chances of it becoming a new engine for PC manufacturers' shipments?
AI large models have brought about the so-called "iPhone moment" for many tech product sectors, yet the PC industry seems somewhat late to the party.
In September last year, Intel was the first to propose the concept of AI PCs and announced the launch of the industry's first AI PC Acceleration Program. Three months later, it officially released the new Core Ultra mobile processors, featuring Intel's first client-side on-chip AI accelerator—the Neural Processing Unit (NPU). The rapid iterations in upstream supply chains are swiftly transmitted downstream. Leading brands such as Lenovo, Acer, and Asus have officially unveiled new products equipped with Intel's latest processors. Faced with this market opportunity that has been anticipated for forty years, major PC manufacturers clearly couldn't remain reserved, jointly pushing AI PCs into their first wave of prominence.
From a theoretical perspective, the current product forms and user habits make the narrative of AI PCs entirely plausible.
Like mobile phones, smart speakers, and smart glasses, PCs possess high-frequency interactive attributes. As all-scenario productivity tools, the efficiency enhancements brought by AI are more perceptible on PCs. Additionally, compared to other terminal devices, PCs' advantages in storage capacity, device space, core hardware, and other aspects enable them to support a more diverse AI ecosystem. Currently, judging from the applications of AI large models, PCs remain the optimal platform for these models. From ChatGPT to Microsoft 365 Copilot, killer applications often debut on PCs first, where the efficiency improvements brought by AI are immediately noticeable.
PCs handle issues like power consumption and computational power more straightforwardly. Even laptops, benefiting from GPU performance iterations and the emergence of the latest NPU architectures, still have room for performance enhancement. The fact that Lenovo's first batch of AI computers focuses on laptops also proves this point.
In short, PCs, especially laptops, meet the hardware requirements for the large-scale computational power needed by big models without overly disrupting users' habits regarding computer-type products. Huachuang Securities believes that low latency and privacy protection will drive demand for localized large models. AI PCs, with their edge computing capabilities, are expected to address industry pain points and become the first widely adopted AI terminals. Implementing some AI applications on PC user ends could stimulate substantial PC replacement demand, potentially ushering the PC industry into a new growth cycle.
IDC forecasts that the Chinese PC market will experience steady growth over the next five years due to the advent of AI PCs. The total market size for desktops, laptops, and tablets is projected to increase from 68 million units in 2023 to over 80 million units by 2027, a growth of nearly 18%. NVIDIA CEO Jensen Huang has also noted that new AI PCs will replace traditional PCs within the next decade, creating a market worth trillions of dollars.
However, predictions are one thing, and reality is another. The evolutionary process of AI PC is divided into two stages: AI Ready and AI On. In the AI Ready stage, AI PC mainly manifests as an upgrade in chip computing architecture, possessing basic local hybrid AI computing power. By the AI On stage, AI PC will have complete core features, offering groundbreaking AI innovation experiences, providing personal AI assistant services in general scenarios based on a richer AI application ecosystem, and enabling personal large model fine-tuning services in edge private environments.
The first wave of enthusiasm has already begun, but the current changes in processors can only be considered as AI Ready PCs, which have hardware capabilities but cannot independently apply AI. Further support from upper-layer model developers and application software vendors is still needed. Therefore, how much impact it can create in the stagnant PC market remains to be seen with the subsequent product rollouts.
Current AI PCs are like the first-generation iPhone—impressive enough, but making every user own and recognize them is not easy. The collaboration between upstream and downstream PC industries is a huge driving force, but more importantly, the cost of AI PCs must be affordable to inspire and benefit every user. Meeting users' demands for both performance and affordability, enabling everyone to enjoy the extended AI capabilities of the large model era, naturally imposes higher requirements on PC hardware and software.
Conventional PC processors, with upgrades from Intel or AMD, may initially see a 10% to 20% price difference at launch, but within six months to a year, prices typically stabilize to match previous-generation products with similar configurations. AI PCs, however, require significantly higher computational power, whether through dedicated GPUs like NVIDIA's, integrated NPUs in CPUs, or specialized compute cards, all of which substantially increase costs. For example, an NVIDIA card with around 75 TOPS of computing power may cost around 7,000 to 8,000 RMB, potentially doubling the overall system cost.
That said, current AI-enabled products are positioned in the mid-to-high-end segment, where profit margins are already substantial, so the price impact is less noticeable. For instance, the first batch of laptops equipped with Intel's Meteor Lake AI processors recently launched with a starting price of 14,999 RMB, virtually unchanged from their predecessors in the same series. It's worth mentioning that in addition to the CPU, memory, storage, and cooling systems are the components most likely to be upgraded in the future. For example, conventional lightweight laptops typically last around 6 hours, while devices equipped with AI computing cards may have an expected battery life of 4 hours, or even just 2-3 hours when running at full power.
However, these are just theoretical parameters obtained under certain hardware testing conditions. The actual upper limits will depend on the specific products and ecosystem implementation. Huo Jinjie, President of IDC China, stated that the birth of AI PCs marks the beginning of a major transformation in the terminal industry represented by PCs. Every player in the PC industry ecosystem will undergo changes due to AI PCs.
Of course, the specific effects will vary depending on the model, which poses a challenge for the compatibility between AI hardware manufacturers and foundational large models. For some time after their launch, AI PCs may still be more inclined towards B2B corporate procurement. The large-scale emergence of B2C demand might require PC manufacturers to present sufficiently attractive cases and results. Another consideration is how manufacturers will share the high costs associated with large model iterations. In the short term, the training, iteration, and operational costs of large models are unlikely to decrease significantly. However, questions remain about how large model products on personal computers will be implemented and iterated—whether through a one-time purchase model or a subscription-based approach.
From the official product listings, Lenovo's newly released AI PCs, such as the Pro 16 IMH and ThinkPad X1 Carbon AI, currently focus their AI concepts on hardware provided by Intel. This includes the Core Ultra processors and the NPU AI engine within them. The only AI feature mentioned for a specific scenario is an AI meeting function, which will be upgraded via OTA. Concepts like "private large models" and other application aspects are not explicitly addressed.
Current large AI models, if evaluated solely based on chip capabilities, have ample room to perform larger-scale computations and develop additional features. However, balancing the computational power required for large model operations with the system resources they consume remains a significant challenge. For performance-oriented laptops, deploying large models locally means requiring larger storage capacity and higher hard drive costs. For ultraportable business laptops, cloud deployment not only incurs higher costs but also creates a paradox with security concerns. Balancing these factors poses a significant challenge for PC manufacturers. After decades of PC development, many market perceptions have formed strong inertia - aspects like size, weight, design, and even price ranges have very limited room for adjustment.
Frankly speaking, while AI PC is undoubtedly becoming the new trend in PC applications, and cost issues may be solvable through iterations, the key question remains: how much consumer demand can manufacturer-driven upgrades actually stimulate?
Although multiple research institutions are optimistic that AI PC will drive a new wave of PC upgrades, this alone doesn't constitute a foolproof driving force. The current PC market is in an overall downward trend. As a new product form, AI PCs enhance localized computing capabilities, but their use as personal assistants has limited impact on driving overall market upgrades. AI products like ChatGPT initially show high registration numbers and optimistic trends, but users eventually recognize their limitations—they cannot fully replace all tasks and applications. Therefore, AI PCs may attract some early adopters to upgrade, but most people won't replace their devices solely for this feature.
Consumers tend to upgrade according to their normal replacement cycles. Due to global economic constraints and limited product breakthroughs in recent years, the replacement cycle has lengthened. By 2024, a growth period for device upgrades is expected, and combined with the AI PC concept, the market may see relatively noticeable growth. However, a portion of this will still come from regular replacement users.
For the industry, a positive sign is that the PC sector, after enduring challenges, is showing signs of emerging from its downturn. According to IDC's latest data, global PC shipments in Q3 2023 reached 68.2 million units, showing an 11% quarter-on-quarter growth and narrowing the year-on-year decline to 8%. A large number of PC users are expected to enter a peak upgrade period in the next two years.
However, it cannot be ignored that whether consumers are willing to pay for additional AI PC features or upgrade their devices for offline large model capabilities depends on how much these AI PC functionalities actually help them. For example, students might need AI PCs to assist in summarizing and organizing arguments and viewpoints, but this need may only arise at specific moments or periods. Similarly, office workers might require smarter PowerPoint and Word functionalities, which would necessitate AI PC assistance.
In short, the higher the relevance of AI PC features to consumers' daily work, study, and life, the more likely they are to upgrade their devices. Therefore, the actual effectiveness of killer applications or on-device large model applications will ultimately determine the market's upgrade momentum. With the integration of AI technology into PCs, software dependency will increase, and ecosystem binding will become tighter.
The differentiation of future AI PCs will mainly be reflected in software capabilities and user experience. Just like the smart speaker market, hardware differences are minimal, but software optimization and user experience vary significantly. The combination of both software and hardware will drive industry transformation, and only the AI On stage possesses the complete core features of an AI PC. Therefore, starting next year, competition among manufacturers will increasingly focus on software optimization and user experience, and as competition intensifies, the market will gradually expand.
Finally, when discussing how capital and the industry view this AI PC revolution, we must not forget that Moore's Law and global supply chain division of labor still dominate the development of the PC industry. Although their essence is to serve consumers, they also introduce many variables and uncertainties to industry development.