(Report Author: Huaxi Securities, Liu Zejing, Meng Lingruqi)
01 The Year of AI Explosion: The iPhone Moment for AI Has Arrived
AIPC Chip Side: Intel Releases Core Ultra Processor
Intel officially launched the AIPC processor in 2023: Following the announcement of the AI PC acceleration plan on October 19, 2023, Intel is set to showcase its AI hardware extravaganza: the official launch of the AI PC processor, codenamed Meteor Lake, at the “AI Everywhere” event on December 14, 2023. According to Intel, the Meteor Lake processor “represents the most significant architectural change for Intel in 40 years, laying the foundation for PC innovation in the next decade,” aiming to “pave the way for the AI PC era”: this chip features a modular architecture composed of four independent modules: computing module, SoC module, graphics module, and IO module, connected via Foveros 3D packaging technology.
The Intel® Core™ Ultra processor is designed for edge AI: It adopts a 3D performance hybrid architecture that supports advanced AI functions, including office, writing, and creative capabilities. The Intel® Core™ Ultra processor can harness AI energy using three dedicated engines (CPU, GPU, and NPU) to create an immersive graphics experience while maintaining high performance and low power consumption. Intel® AI Boost is an integrated AI engine designed for low-power AI acceleration and CPU/GPU offloading, while the Intel® Gaussian & Neural Accelerator can free CPU resources while processing AI voice and audio applications to ensure overall system performance and responsiveness.
AIPC Chip Side: Qualcomm Snapdragon X Series Fully Supports AIPC
At the 2024 COMPUTEX conference, Qualcomm showcased its latest AI PC products, marking a new era for the PC industry: Qualcomm’s Snapdragon X series platform is the core highlight of this exhibition, with PC products equipped with this platform featuring powerful AI capabilities, such as real-time translation, intelligent assistants, and multimodal interaction. These features demonstrate the broad application prospects of AI PCs in office, creative, and everyday life.
Qualcomm Snapdragon X and Snapdragon Xplus are built for artificial intelligence: The Snapdragon X Elite can run generative AI LLM models with over 13 billion parameters at incredible speeds on devices. It also includes updated dual micro-NPUs within the ultra-low-power Qualcomm® Sensing Hub to enhance security, login experience, and privacy, including the ability to wake the device from sleep mode. Sensitive data can remain on the user’s laptop to enhance security through edge intelligence. The Qualcomm AI Engine supports Windows Studio Effects and many other AI-accelerated applications and experiences.
AIPC Chip Side: NVIDIA Plans to Launch AI PC Chips with Blackwell Architecture
NVIDIA continues to ramp up AI, building a foundation of computing power for the world: At CES 2024, NVIDIA released three new GeForce RTX 40 SUPER series graphics cards, including GeForce RTX 4080 SUPER, GeForce RTX 4070 Ti SUPER, and GeForce RTX 4070 SUPER, which are core to AIPC, providing immense power for the latest games and delivering a completely upgraded experience for users.
NVIDIA plans to launch AI PC chips based on the Blackwell architecture: NVIDIA is preparing to launch a chip that combines the next-generation ARM core with its Blackwell GPU architecture. According to C114, NVIDIA is preparing a system-on-chip (SoC) that combines Arm’s Cortex-X5 core design with a GPU based on its recently launched Blackwell architecture.
The competition landscape for chipsets centered on “CPU + GPU” is becoming clearer: Currently, the competition landscape among chip manufacturers is no longer fragmented. In the past, AMD and Intel were the leaders in CPUs, while NVIDIA led in GPUs. Now, chip manufacturers are gradually strengthening their “weaknesses” to maintain their leading positions in application scenarios, such as NVIDIA’s own Grace CPU or Intel’s release of its Core Ultra AIPC processor, which integrates CPU, GPU, NPU, and IO chips within a centralized graphics card.
AIPC Whole Machine Side: Dell, Microsoft, HP, and Others Fully Join the AIPC Competition
Dell has released multiple AIPC products: On December 19, 2023, global PC giant Dell launched the new AI PC Inspiron 13 Pro, equipped with the new Intel AI PC processor Core Ultra, leading the way into the AIPC era. The Inspiron 13 Pro can smoothly run a 20 billion large language model locally, supporting functions like Stable Diffusion for text-to-image and image-to-image generation. Additionally, on May 21, Dell Technologies announced five new enhanced AI computers at Microsoft’s annual conference, all equipped with Qualcomm Snapdragon X series chips. Acer, ASUS, HP, Lenovo, and Samsung have also “fully supported” AI computers: At the beginning of the year, during a new product launch event, Microsoft CEO Satya Nadella announced that Microsoft would fully integrate the AI assistant Copilot into the Windows system, releasing the first batch of AI personal computers (AI PCs). The new AI PCs are referred to as “Copilot+ PCs,” claiming to be “the fastest and most AI-optimized PCs ever.” Major OEM partners include Acer, ASUS, Dell, HP, Lenovo, and Samsung. We believe AIPC is an inevitable trend, and the competition for AIPC has begun, with clearer scenarios for AI application implementation.
The Explosion of Edge AI Drives a Wave of Device Upgrades: The iPhone Moment for AI Has Arrived
2008-2012 iPhone Sales Surge: Apple launched the iPhone 4 and 4S in 2010 and 2011, respectively. Their outstanding design and performance led the smartphone wave, coinciding with the official launch of 4G networks. Compared to 3G networks, which could only support basic web browsing, voice chatting, and 2D online gaming, the social attributes of 4G networks were significantly enhanced, including video chatting and 3D online gaming, which became widely accepted by the public. The iPhone 4 and 4S, with their closed ecosystems, high-quality hardware and software, and simple yet sophisticated design concepts, led the smartphone wave, as evidenced by iPhone sales soaring from 39.99 million units in 2010 to 150 million units in 2013.
The iPhone moment for AI has arrived, and AI demand is expected to drive a wave of device upgrades: Due to hardware computing power constraints, edge AI requires higher computing power. Older PCs or smartphones do not support AI functions. For instance, Apple has stated that Apple Intelligence will be available on the iPhone 15 Pro series, iPads and Macs equipped with M1, and subsequent models. This is because Apple Intelligence requires robust edge processing capabilities, necessitating A17 Pro or the entire M series SoC to function. Therefore, we judge that AI demand is likely to drive a wave of device upgrades, marking the arrival of AI’s iPhone moment, while the application scenarios for edge AI are becoming clearer.
02 The Trend of AI Hardware Upgrades is Becoming More Obvious
AIPC’s Large Models Can Be Divided into Online and Offline Modes
AIPC’s large models can be divided into online and offline modes: The online mode involves simple preprocessing on the mobile end before uploading user data (images, text, voice, video, etc.) to cloud servers, where powerful computing resources generate inference results, which are then sent back to the edge device. The offline mode involves models trained on servers that are quantized and converted through mobile deep learning frameworks (TensorFlow Lite, NCNN, TNN, MNN, CoreML, etc.) for prediction on mobile devices. This method can be implemented without internet connectivity, ensuring user privacy. Taking Apple as an example, Apple Intelligence can be simply divided into these two parts.
Cloud large models focus on powerful functionality, while edge large models emphasize low latency: The inference platforms for online and offline modes differ. The offline mode (edge models) emphasizes low latency and implementation experience, while the online mode (cloud models) focuses on powerful functionality.
Cloud AI and Edge AI Are Indispensable, and Edge-Cloud Collaboration Becomes a Necessity
We judge that edge-cloud collaboration is essential: The computing power requirements for edge AI are relatively smaller than those for cloud AI due to two reasons: 1. Edge AI applications lean more towards inference; 2. Edge AI models are more focused on “compression” compared to cloud AI. Currently popular features like “one-click background removal in smartphone photography” cannot be accomplished solely with edge large model computing capabilities; edge-cloud collaboration is necessary to complete such tasks. Moreover, edge AI serves as a traffic entry point, similar to the difference between smartphones and PCs compared to previous mobile phones and PCs. Additionally, edge AI is also a necessity for terminal manufacturers for two reasons: first, to enhance their product strength and brand power; second, to increase their technological advantages, thereby enhancing the product strength for consumers.
Lenovo Releases the Lenovo Personal Cloud Collaborative Intelligent Agent “Lenovo Xiaotian”: Lenovo Xiaotian combines the high performance and generalization capabilities of cloud large models with the low cost, data privacy, and high reliability of edge large models. During the experience event, Lenovo Xiaotian demonstrated outstanding natural language interaction capabilities, capable of conversing in a humanized voice and tone to solve various user needs, such as daily conversations, information searches, device control, document processing, and more. According to Zhixian News, the high IQ behind Lenovo Xiaotian is based on the abab6.5 model, which is built on the MoE architecture with a parameter count reaching trillions and a context window of 200,000 tokens. The high emotional intelligence comes from Lenovo training it with billions of high-quality dialogue data for emotional feedback.
03 AI Reshapes Operating Systems
Android 15 Teams Up with Gemini Large Model, Leading AI Innovations into the Future
AI technology has made substantial breakthroughs in mobile operating systems: The Google I/O 2024 developer conference revealed innovations in Android 15 and the Gemini 1.5 Pro large model, including the next-generation mobile operating system Android 15, advanced Gemini large model, and several AI-driven innovative features. Google announced that Android 15 will deeply integrate Google’s Gemini large model, named Gemini Nano, bringing users an unprecedented intelligent experience. The new system will support cutting-edge features like AI voice assistant fraud prevention and circular image search, while expanding platform support through satellite communication to ensure connectivity in remote areas.
The Gemini family welcomes an epic update: Gemini 1.5 Pro expands context window support to 2 million tokens, claiming to have “the longest context window among all foundational models to date.” At this conference, Google announced the launch of Gemini Advanced based on Gemini 1.5 Pro. New members include Gemini 1.5 Flash, which excels in summarization, chat applications, image and video subtitles, and data extraction from long documents and tables. In application scenarios, it excels at summarization, chatting, image/video subtitling, and long document data extraction tasks. Gemini 1.5 Flash inherits its core capabilities by learning from the 1.5 Pro model. Additionally, Google showcased the new multimodal AI project Project Astra, representing the latest advancements in Google’s AI assistant.
Microsoft AI PC Fully Unveils, Copilot Integrated into Windows 11
Microsoft AI PC fully unveils, with Copilot integrated into Windows 11: Microsoft introduced Copilot+ PC as the fastest and most intelligent Windows PC to date. With powerful new chips, it can achieve over 40 TOPS AI computing power and battery life of up to a whole day. Each Copilot+ PC is equipped with a powerful AI Agent, allowing for quick interaction with just a tap of the new Copilot key on the keyboard. In the future, users will gain access to the latest models, including OpenAI GPT-4o, for more natural voice conversations. AI capabilities are also fully embedded in the Windows 11 system, such as a Copilot button for modifying settings in the control panel and a one-click photo editing option in the image file menu.
Microsoft AIPC Features Are Powerful: The Copilot+ PC uses strong processors and multiple state-of-the-art AI models, including Microsoft’s top small language models, allowing users to run AI directly on the device, featuring instant recall, image creation and editing, and AI photo editing capabilities.
The next generation of Windows operating systems is expected to integrate AI technology more deeply: According to reports from Zhongguancun Online and 36Kr, the next generation of Windows operating systems is likely to be restructured around AI functionalities, becoming part of the system’s core. At this point, AI will no longer be constrained by system security, allowing direct calls to various system functions and applications.
04 Comprehensive Autonomous Control of Software and Hardware: Huawei Leads the Rise of Domestic Ecosystems
Hardware Side: Huawei HiSilicon is the Light of Domestic Chips
HiSilicon is a global leader in fabless semiconductor and device design. The company originated from Huawei’s Integrated Circuit Design Center, which started integrated circuit design and R&D in 1991. In 2004, Huawei registered HiSilicon as a wholly-owned subsidiary to provide sales and services for HiSilicon chips, gathering industry talent and leveraging industry integration advantages. HiSilicon products cover various fields, including smart vision, smart IoT, smart media, smart travel, display interaction, mobile terminals, data centers, and optical transceivers.
Product services radiate globally, with continuous R&D and a rich patent portfolio. HiSilicon has 12 offices and R&D centers in China, Singapore, South Korea, Japan, Europe, and other regions, with products and services spanning over 100 countries and regions worldwide. HiSilicon has over 20 years of technological accumulation, currently owning over 200 self-owned intellectual property chips and more than 8,000 patents.
Edge PC: Huawei AI PC Has Applied the Pangu Large Model
The Pangu model is powerful: The new AI summary function can quickly and accurately extract key information from audio content, providing customers with three core components: a one-sentence summary, keywords, and a full summary (around 200 words, varying based on the full text length), thus helping customers grasp meeting points more efficiently. The AI summary can not only transcribe in real-time while recording but also support summarizing local audio and video files. When users lack internet access or forget their computers, they can record or film using their phones or cameras and later upload to their computers for AI summary information extraction.
The Edge AI Revolution is Ongoing: Pangu + HarmonyOS + Kirin + Ascend + Kunpeng Build the Strongest Ecosystem
Launching the critical autonomous control defense battle is imperative: For terminals like smartphones, Apple has its own A-series smartphone chips and M-series PC chips, operating systems like iOS, and cloud capabilities equipped with NVIDIA and AMD AI computing cards. On the large model side, Apple Intelligence integrates ChatGPT. For AI PCs, overseas manufacturers have Intel, AMD, Qualcomm, and other edge AI PC chips, with Windows as the operating system, currently integrating AI functions like Copilot. Since 2018, the U.S. has severely restricted the development of high-tech fields in China through various sanctions, making autonomous control in critical high-tech areas imperative.
Pangu + HarmonyOS + Kirin + Ascend + Kunpeng Build the Strongest Domestic Ecosystem: We judge that the explosion of edge AI is an industry trend, and AI applications are moving closer to implementation. Therefore, in the domestic field, Huawei has the most complete ecosystem. On the software side, Pangu can empower thousands of industries, while HarmonyOS can achieve the Internet of Everything, such as smart screens, wearable devices, vehicle systems, speakers, and smartphones. On the hardware side, edge-wise, Kirin has returned to its peak, and Kunpeng + Ascend are expected to build the second pole of global computing power systems. Thus, we judge that in the year of terminal explosion, Huawei is likely to lead the rise of domestic software and hardware ecosystems with Pangu + HarmonyOS + Kirin + Ascend + Kunpeng.
Report Excerpt:
