**: Zhitong Finance App
At the 2024 International Consumer Electronics Show (CES), Intel and AMD, leaders in PC CPU chips, have successively announced new desktop or laptop ** processors that integrate AI inference modules, and Nvidia, the "shovel seller" in the field of AI training, is also seeking to gain market share in the AI PC market, and launched a new GPU GeForce RTX 4080 Super covering both high-quality games and AI application software at CES. Compared with AI training, the AI inference field is extremely closely related to the needs of large-scale consumer electronics and other application terminals, so the development focus of the AI industry is expected to shift from "training" to "inference".
The advent of ChatGPT at the end of 2022 marks the acceleration of human society into the AI era, and 2023 will be accompanied by the growth of GPU demand focusing on AI training. In contrast, since there are no blockbuster products on the AI application side except ChatGPT, the AI inference field with low TFLOPS numerical requirements (i.e., computing power requirements) is bleak.
However, with the imminent emergence of new AI-integrated consumer electronics such as AI PCs and AI smartphones in 2024, the trend of "AI + everything" has become unstoppable. South Korean tech giant Samsung Electronics held a blockbuster launch event at CES 2024, with the theme of "All for AI: Connectivity in the Age of Artificial Intelligence". Samsung's latest products can be said to be AI-related, such as AI TVs, AI refrigerators, AI washing machines, AI vacuum cleaners, AI laptops, etc. Volkswagen announced at CES that its voice assistant will be embedded with ChatGPT artificial intelligence technology and will be used in all vehicle models. The automotive giant showcased its first ChatGPT-equipped car, which enables users to solve their questions with a voice assistant through a chatgpt-like conversational interface. The AI technology is planned to be rolled out in North America and Europe early in the second quarter of this year.
UBS UBS expects that by 2027, AI technology will be widely used in all walks of life in the world's major economies, thereby promoting AI models and AI software applications to become a market segment worth up to $225 billion, compared with only $2.2 billion in 2022, which is an epic leap forward, and the compound annual growth rate during this period is expected to be as high as 152%.
These large-scale application trends of device-side AI models and AI software based on the background of "AI integration of everything" are bound to be presented on a large scale in a wide range of consumer electronics application terminals such as PCs, smartphones, and smart watches, which also means that the importance of AI inference is becoming increasingly prominent, and the core technology behind the efficient operation of end-side large models and AI software based on inference is required. Qualcomm CEO Cristiano Amon recently emphasized that the main market of global chipmakers will soon fully shift to the field of AI inference.
Compared with AI training, the AI inference field is far less demanding than the training field for GPU parallelization in the context of "massive data bombardment" applications, and the inference process involves the application of trained models for decision-making or identification, and the CPU-based processor that is extremely good at complex logic processing tasks and control flow tasks is enough to cope with many inference scenarios efficiently.
Therefore, with the full influx of AI and all things, the first-class processor with CPU as the core is about to usher in a new round of first-class market scale expansion opportunities, and those long-neglected consumer electronics CPU giants are returning to the field of vision of global investors. As the AI inference market becomes more and more large, the inference side "sells shovels" - such as AMD (AMD.US), Intel (INTC.).US) and Qualcomm (QCOMUS), the three veteran chip giants in the field of consumer electronics, are ushering in their own "Nvidia moment", that is, the moment when stock prices and performance start to skyrocket.
The first year of AI PCs and AI smartphones begins!The CPU giants are back in the spotlight.
2024 can be described as the first year of AI PCs and AI smartphones. Well-known PC brands such as HP, Dell, Acer, Asus, MSI, and Gigabyte are all set to launch the first wave of AI PCs based on Intel or AMD processors in 2024. Samsung is leaning towards AI technology as the most critical factor in achieving larger smartphone sales this year, with new AI smartphone products from Chinese smartphone makers such as Xiaomi, Vivo, Honor and OPPO set to be released this year. With PCs and smartphones, the two core consumer electronics carriers, AI applications are bound to emerge in large numbers.
Intel's new Core Ultra processor integrates AI-specific neural processing units (NPUs) with ARC GPUs into the CPU, of which the NPU is dedicated to AI inference task acceleration, and this integrated CPU+NPU+GPU processor is designed to be the company's "most efficient processor", marking the official arrival of the AI PC era. Intel's Lunar Lake processors for laptops, which will be available in the second half of 2024, feature a "new low-power architecture with significant IPC improvements" and three times the AI data processing performance of GPU and NPU modules compared to Meteor Lake. Intel said the chips are currently available to Intel's partners.
AMD launched the Ryzen 8000G series desktop processor at CES 2024, which is AMD's first desktop** processor with integrated AI functions, integrating the NPU for processing AI acceleration tasks, the Ryzen 8000G has a CPU based on the Zen4 architecture, and integrates the NPU and RDNA3 GPU built by the Ryzen AI engine, positioning the desktop-level platform. Qualcomm Snapdragon 8 Gen3 is Qualcomm's first mobile chip built for AI acceleration, supporting multi-model generative AI models, including Meta Llama 2, which can handle 10 billion parameter scale of device-side AI large models and execute up to 20 tokens per second.
According to Zacks Investment Research, 2023 is a crucial year for the AI industry, with the debut of Nvidia and AMD's training GPU products, as well as various investments and strategic acquisitions. Looking forward to 2024, Zacks said that after technology companies have the basic hardware of chips, they will continue to update AI models and build AI applications, so the further development of AI technology is expected to drive consumers' hardware upgrades - such as turning to AI PCs and AI smartphones, as well as new AI-based software services, such as device-side AI large models, software applications embedded in new AI technologies such as chatbots.
Behind the efficient operation of the device-side AI large model and AI software, it is based on the core technical process of AI inference, and the hardware foundation of the AI inference process lies in the first-class processor with CPU as the core. The architecture foundation of the CPU determines that the CPU can not only perform general-purpose computing tasks, but also focus on the scheduling characteristics of control flow and processing complex sequential computing tasks and logical decisions, making the CPU fully shine in the field of AI inference.
In the field of AI inference, for example, the application scenarios of end-side AI large models of consumer electronics such as AI PCs, AI smartphones, and smart watches, as well as the efficient operation of various AI software, with CPUs focusing on complex logical decision-making as the core processor, and integrating NPUs and GPUs as auxiliary computing power support, can realize the efficient operation of streamlined device-side AI large models and multiple AI software. After all, the AI training end is based on massive parallel computing tasks, and the large model training process involves processing a large amount of data and performing complex mathematical calculations, which are suitable for accelerating processing through parallelized computing.
In most AI inference tasks, the CPU is the core, supplemented by NPUs and GPUs, which can effectively perform AI tasks that do not require large-scale parallel processing, such as processing small to medium-sized datasets or performing normalized model inference tasks. In this set, the CPU is regarded as the core part, equivalent to the "human brain", responsible for processing complex logical decision-making and control task flows, it is the main computing unit of the entire system-on-a-chip, responsible for executing program instructions, handling daily computing tasks, and coordinating and managing other parts of the entire system. The NPU (Neural Processing Unit) is optimized for AI inference acceleration and can provide fast AI inference performance at low power consumption, especially for neural network-related tasks. GPUs are extremely good at parallelized computation, are suitable for performing a large number of matrix and vector operations, and can take on heavy responsibilities when dealing with data-intensive AI inference tasks such as image and analysis.
Therefore, Intel, AMD and Qualcomm, the three long-forgotten CPU giants in the market, have recently returned to the spotlight of the capital market, and the upward trend in stock prices since November last year is the best proof that global funds favor these companies. At one point, the three giants were even mistaken by some analysts for missing out on the global AI boom.
Looking to the future, the AI inference field is expected to outperform the training field, and these "shovel sellers" are about to usher in their own "NVIDIA moment".
From the perspective of industry development trends, there is a high probability that the AI computing power load will gradually migrate from training to inference, which means that the threshold for AI chips may be significantly lowered, and chip companies covering wearable devices, electric vehicles, and the Internet of Things are expected to fully penetrate the field of AI inference chips in the future. Intel CEO recently said that AI inference technology will become more important than AI training technology, and he emphasized that Intel will not rely solely on AI training, and Intel will pay more attention to the field of AI inference.
Qualcomm CEO Amon pointed out that the main battlefield of chip manufacturers will soon shift from "training" to "inference", Amon said in a recent interview: "At present, the AI market is mainly focused on the 'training' stage of using big data to train large language models, and Nvidia is the main beneficiary in this field." But as large AI models become more streamlined, able to run on devices, and focused on inference tasks, the chipmakers' primary market will shift to 'inference,' or model applications. It is also expected that data centers will be interested in processors dedicated to the inference tasks of trained models, all of which will help the inference market outpace the training market. ”
Morgan Stanley, a major Wall Street bank, pointed out in the top 10 investment strategy themes for 2024 that with the significant improvement of consumer edge devices in data processing, storage and battery life, there will be more catalysts in 2024 to promote the edge AI segment to catch up, and the development focus of the AI industry will also shift from "training" to "reasoning".
Edge AI refers to the technology that directly processes AI data streams on devices such as PCs, smartphones, IoT devices, and automobiles. Market research firm Gartner predicts that by 2025, 50% of enterprise data will be created at the edge, spanning billions of devices. This means that the inference of AI large models (i.e., the process of applying models to make decisions or recognition) is expected to be performed in batches on device-side devices, rather than on remote servers or in the cloud.
According to InvestorPlace research report, 2024 will be the year of the explosion of AI application software, and hundreds of new AI-based applications are expected to appear, and these AI applications covering all walks of life will be widely spread in society, and InvestorPlace predicts that by 2025, we will be in the wave of AI software covering all walks of life. Therefore, the explosion of AI software means that the AI PCs and AI smartphones that carry these software applications will usher in a new wave of replacement, and the demand for chips for consumer electronics inference tasks is expected to surge.
Counterpoint Research, a well-known research organization, expects the global PC market to return to pre-pandemic levels in 2024, thanks to the Windows 11 replacement, the next wave of ARM PCs, and AI PCsAI PC is expected to grow at a compound growth rate of 50% from 2020 onwards and dominate the PC market after 2026, with a penetration rate of more than 50%.
Another research institute, Canalys, expects global PC shipments to be around 24.9 billion units, a year-on-year decrease of 124%, and PC shipments are expected to be 26.7 billion units, a year-on-year increase of 76%, AI PC shipments in 2024 will be around 20 million, and in 2027, 60% of PCs will have the latest version of AI functions, and PC shipments will exceed 17.5 billion units. According to Qunzhi Consulting, 2024, as the first year of AI PC development, is expected to reach 7% in the PC market, close to 30% in 2025, and more than 50% in 2026.
According to Counterpoint Research, a well-known market research organization, it is expected that by the end of 2027, the shipment of AI smartphones with built-in generative AI capabilities is expected to exceed 500 million. Counterpoint expects 2024 to be a key year for global AI smartphones, but it is estimated that shipments will only reach 100 million units, so the AI smartphone market is expected to grow at a CAGR of 83% between 2023 and 2027.
"Samsung and Qualcomm are the most direct leaders in this segment because their current product systems and production capabilities make them early movers," Counterpoint researchers wrote in a report released in December. "Just like it does with foldable phones, Samsung is likely to hold nearly 50% of the market share over the next two years, followed by major Chinese OEMs like Xiaomi, Vivo, Honor and Oppo." ”
In terms of stock price expectations, the stock price of AMD, a dual-industry giant of CPU and GPU, has continued to soar recently, and Wall Street analysts are increasingly optimistic about AMD's target stock price. Wall Street investment firm Melius Research upgraded AMD's rating to "Hold" from "Hold" and raised its price target to $188 over the next 12 months from $125 (AMD's latest ** price is 146.).$18). The agency believes that generative AI is expected to have a "halo effect" on enterprise IT spending starting in 2024, and believes that chip companies with high performance can continue to maintain the momentum of soaring stock prices from 2023 onwards.
John Vinh, an analyst at Keybanc, maintained an "outperform**" rating on AMD and raised his price target sharply to $170 from $140. John Vinh said that AMD is very likely to have a market share of about 15% in the AI chip market in recent years, which means a large-scale revenue opportunity of about $20 billion.
Ivan Feinseth, an analyst at Tigress Financial, raised Intel's price target to $66 from $46 (Intel's latest price is $48.)$45), maintaining the "** rating." Feinseth's upward target is based on two catalysts: Intel's new PC with AI-accelerated processors, and IFS (Intel Foundry Services) continues to expand, which will re-accelerate revenue and cash flow growth.
Mizuho analyst Vijay Rakesh raised Qualcomm's price target to $155 from $140 (Qualcomm's latest price is $139.)$03) and maintained a "** rating" on the stockBernstein, another well-known investment institution, raised Qualcomm's price target to $160 from $145 and maintained an "outperform**" rating.