DeepSeek unlocks the future of edge AI, with Qualcomm (QCOM.US) SoC chips leading the demand dividend.

By early 2025, the explosive growth of DeepSeek undoubtedly shocked the global AI industry. Through optimizing algorithms and architecture, DeepSeek achieved performance comparable to world-class large models at a lower computational cost. This led many investors to shift their focus to questioning the biggest beneficiary of the global wave of AI spending by large tech companies over the past two years, NVIDIA Corporation (NVDA.US), which even triggered significant volatility in the US stock market. Nevertheless, regardless of whether the underlying logic of NVIDIA Corporation's computational requirements is refuted, this cannot change another significant implication behind this technological breakthrough - the transition of AI from the cloud to the edge. The rise of edge AI: a paradigm shift from the cloud to the edge In fact, DeepSeek compresses billion-parameter models to run on edge devices through distillation technology, significantly reducing the cost of AI inference. With a "computational power equalization" approach, DeepSeek accelerates the explosion of edge AI. Edge AI refers to AI technology that completes data collection, processing, and decision-making on terminal devices locally. Compared to centralized processing relying on the cloud, edge AI has advantages of low latency, high privacy, and low bandwidth dependency, making it particularly suitable for high-real-time scenarios (such as autonomous driving, AR interaction). After the wave of AI caused by ChatGPT, many companies have focused on edge AI for popularization, but progress has been unsatisfactory, with a major reason being the mismatch between computational requirements and model needs. Traditional edge devices (such as phones, IoT devices) have limited computational power in local SoC chips, making it difficult to support the inference demands of large models. For instance, early large models had parameter scales in the billions (like ChatGPT), while edge chip computational power usually only supported models below the billion scale, leading to significantly reduced functionality. This has caused many businesses to rely more on cloud-based AI model deployment. However, cloud-based AI also has significant flaws: latency and timeliness. Depending on internet connectivity, unstable network environments may lead to data transfer delays, affecting the execution of real-time tasks; privacy and security. Transferring sensitive data for cloud processing poses data privacy and security risks, requiring strict protective measures; cost issues. Establishing and maintaining cloud computing resources requires significant capital investment, especially for small businesses, where cloud computing could be costly. The breakthrough of DeepSeek-R1 lies in its "distilled model" technology, compressing large model parameters and optimizing inference efficiency, enabling complex AI tasks previously requiring cloud computational support to run locally on terminal devices. Recently, Qualcomm (QCOM.US) released a white paper outlining how the emergence of DeepSeek would drive AI development at the terminal side. The company pointed out that thanks to various technologies such as distillation, small models are now approaching the quality of cutting-edge large models. The research showed that DeepSeek's distilled versions based on the TanYi Qianwen model and Llama model demonstrated many advantages, especially in the GPQA benchmark test, achieving similar or higher scores compared to advanced models such as GPT-4o, Claude 3.5 Sonnet, and GPT-o1 mini. The white paper states: "Many mainstream model series, including DeepSeek R1, Meta Llama, IBM Granite, and Mistral Ministral, have released small model versions, and their performance and benchmark testing for specific tasks have been outstanding. Reducing large base models to smaller and more efficient versions not only achieves faster inference speeds, less memory usage, and lower power consumption but also maintains high performance levels, making such models suitable for deployment on terminals such as smartphones, PCs, and cars." The research concluded that advanced AI small models have remarkable performance, model parameter scales are rapidly shrinking, increasing the number of models is helping developers create richer applications on the edge, and personalized multimodal AI agents will simplify interactions and efficiently achieve tasks across various applications. These advancements are expected to lower the cost of training new models and make AI more pervasive. Diverse range of device carriers In fact, not just Qualcomm, many analysts and tech giants are also very optimistic about the prospects of edge AI. Renowned analyst Ming-Chi Kuo stated in a recent research report that the boom of DeepSeek will accelerate the rise of edge AI trends. He pointed out that DeepSeek not only optimized on functionality but more importantly, it sparked a trend of deploying large language models (LLMs) locally. Industry giants including Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR (TSM.US) and NVIDIA Corporation expect significant growth in the edge AI market around 2026. According to DataM Intelligence's edge AI market report released in January, the global edge AI market reached $16.8 billion in 2023 and is expected to reach $73.8 billion by 2031, with a compound annual growth rate of 20.6% from 2024 to 2031. On the application deployment level, due to the large user base, mature software and hardware ecosystems, and rich use cases, smartphones and PCs may be the most perfect application carriers for edge AI. In the future, AI glasses, smart cars, and areas like Siasun Robot&Automation may have greater application potential. From the predictions of multiple institutions, the future of edge AI devices appears to be very promising. AI phones: according to Canalys, the global AI phone market penetration is expected to reach 17% by 2024, and in 2025, more mid-to-high-end models will be equipped with stronger edge AI capabilities, driving global AI adoption.Market penetration rate reached 32%, with shipments close to four hundred million units.AI PC: According to IDC's predictions, the global shipment of AI PCs will reach 150 million units in 2027, with a penetration rate of 79%. AI Glasses: According to IDC's estimates, the global AI glasses market is expected to have a shipment volume of 12.8 million pairs in 2025, a 26% year-on-year growth, and is projected to exceed $300 billion by 2030. Wellsenn XR's forecast data shows that by 2029, the annual sales of AI smart glasses will reach 55 million pairs, and by 2035, it is expected to exceed 1.4 billion pairs. Overall, a report released by market research firm Research and Markets predicts that the global edge AI hardware market size will be 2.3 billion units in 2024, and is expected to reach 6 billion units by 2030, with a compound annual growth rate of 17.6% from 2024 to 2030. SoC: Carrying a huge market growth There is a diverse variety of carriers for edge AI, and the technological breakthroughs of DeepSeek are expected to accelerate the development of more edge AI devices. However, as the hardware base of hardware terminals, the growth potential of SoC chips seems more evident. SoC (System on Chip) chips, as the name implies, integrate all the components needed for a system onto a single chip, including CPU, GPU, and NPU, etc. It can be said that SoC chips are the main control units of various types of hardware devices, carrying core functions such as computing and control, serving as the "brain" of the hardware. In edge AI applications, SoC plays a crucial role. With the continuous development of AI technology, edge devices such as smartphones, wearables, headphones, glasses, etc., require more powerful AI processing capabilities. SoC significantly enhances device AI computing power by integrating modules like NPU (Neural Processing Unit). As mentioned earlier, traditional edge devices' SoC chip processing power is limited, making it difficult to support the inference demands of large models. But with the widespread application of AI on the edge, chip manufacturers have begun to focus on NPU, driving the processing power to tens or even hundreds of TOPS, combined with DeepSeek's technological breakthroughs, significantly reducing the difficulty of deploying edge models, thus encouraging hardware manufacturers to increase research and production of AI devices, and the demand for AI SoC chips is expected to rise. Qualcomm pointed out in a white paper that the rise of DeepSeek signals a potential turning point in AI development. "This critical moment is part of a broader trend, highlighting innovation in building high-quality small language models and multimodal inference models, and these innovations are preparing AI for commercial applications and end-side inference. These new models can run on the edge side, accelerate the scaling of powerful edge-side chips, and create demand for such chips." According to Markets and Markets data, the SoC market is expected to grow from $138.46 billion in 2024 to $205.97 billion in 2029, with a projected compound annual growth rate (CAGR) of 8.3%. The institution points out that the demand for energy efficiency and compact design in mobile devices, IoT, and wearable devices has been driving the widespread application of SoC in the market. SoC provides higher processing power in a smaller size, promoting its popularity in wearable and interconnected devices. With the continuous improvement of AI and machine learning algorithms, there are increasing developments of AI-optimized SoC, especially in the area of local processing for AI PCs and smartphones. These AI-optimized SoC chips have edge computing capabilities, allowing them to improve privacy protection and battery life even in limited connectivity areas. Additionally, the demand for SoC in advanced driver assistance systems (ADAS), autonomous driving, and in-car entertainment systems is steadily increasing, further driving significant growth in the system-level chip market. Qualcomm poised to be a potential winner In recent years, as a giant in the SoC chip field, Qualcomm's stock performance has been far below peers like NVIDIA Corporation that provide chips for AI data centers. After the AI boom, many analysts had expected that edge AI would drive smartphone sales growth, boosting Qualcomm's stock price. However, this effect has not materialized, for reasons that may include delays in the replacement cycle following the outbreak of demand during the pandemic, and the slow deployment of edge AI failing to drive demand. But now, as time passes, the continuous digestion of customer inventory, along with the potential brought by DeepSeek for edge AI deployment, the upgrade cycle for smart devices is likely to start. Qualcomm's latest white paper clearly indicates that the company is ready to embrace the era of AI inference innovation, and from the latest financial reports, it can also be seen that the company has started to benefit from the big trend of AI moving towards the edge. According to the results released earlier this month, Qualcomm achieved a revenue of $11.669 billion in the first quarter of its fiscal year 2025, a 17% year-on-year growth. Qualcomm mentioned that due to AI driving demand growth for smartphones, revenue related to mobile phones increased by 13% to $7.57 billion, exceeding market expectations. The rise of AI smartphones has prompted manufacturers to adopt higher-performance SoC to enhance AI computing capabilities, resulting in a faster business growth rate compared to the previous quarter. Qualcomm expects that next quarter, revenue from the mobile business will continue to grow by 10% year-on-year, continuing the trend of recovery. Melius Research analyst Ben Reitzes wrote in a research report, "Similar to Apple Inc., Qualcomm has also seen growth in the high-end market, especially among manufacturers of smartphones that have introduced AI-related features." Reitzes maintains a "hold" rating on Qualcomm but has raised its target price from $180 to $190. It is also worth noting that Qualcomm's growth potential in the field of edge AI may be much more than just AI smartphones. In fact, Qualcomm is cautious about the future growth rate of the smartphone market. The company predicts that the growth rate of the smartphone market in 2025 will only be in the low single digits. However, this does not affect Qualcomm's potential growth in the field of edge AI.The reason for Qualcomm's optimistic outlook lies in its active expansion of business beyond its core market, injecting new power technology into its revenue growth.Qualcomm pointed out in its financial report that the combination of 5G with high-performance, low-power computing, and on-device AI will drive the expansion of smartphone technology into other industries such as PCs, automobiles, and IoT. This has been the core logic behind Qualcomm's expansion into more business areas in recent years. At an investor day last December, Qualcomm CEO Cristiano Amon stated that the company would expand its long-term strategy with the goal of achieving equal revenue from both smartphone and non-smartphone markets by the end of this decade. As a leader in on-device AI, Qualcomm's solutions already cover billions of smartphones, automobiles, XR headsets and glasses, PCs, and industrial IoT terminals. In the first quarter, Qualcomm's IoT business achieved continuous growth, with revenue reaching $1.549 billion, a year-on-year increase of 36.1%. IoT business revenue includes consumer electronics, edge networks, industrial products, and AI PCs. Thanks to the launch of new products featuring industry-leading processors and on-device AI capabilities, the IoT business has experienced continuous growth after a downturn. Looking ahead, the company expects the IoT business to achieve a 15% growth in the next quarter, driven by the recovery in consumer electronics demand and the growth of AI PCs. Summit Insights analyst Kinngai Chan stated, "Expectations for Qualcomm are high, but we believe this is reasonable as Qualcomm has good prospects for expanding market share in the PC client market and gaining market share in the smartphone market, especially in the high-end market." Qualcomm's automotive chip business revenue also saw a significant increase of 60.7% to $961 million, although the growth rate slowed compared to the previous quarter's 68%, it is still in a phase of rapid expansion. Growth is primarily driven by the expansion of the smart cockpit market, increasing demand for intelligent driving and autonomous driving, and the smart upgrade of the electric vehicle market. Qualcomm pointed out that the demand from automotive companies for its Snapdragon digital chassis solution is increasing, with more manufacturers adopting its high-performance computing platform, driving demand for high-end automotive chips. Qualcomm expects its automotive business to continue to maintain high growth in the coming years, aiming to increase non-smartphone business revenue to $22 billion by the 2029 fiscal year. Morgan Stanley analysts led by Samik Chatterjee stated that based on Qualcomm's accelerated diversification bringing long-term opportunities, it is expected that by the end of this decade, revenue from non-smartphones markets will significantly increase, with PC, IoT, and automotive businesses becoming more important for the company. Its leadership position in edge computing is increasingly crucial, especially in the context of the widespread adoption of AI in on-device devices. Currently, the firm has a "buy" rating on Qualcomm with a target price of $195. In summary, the emergence of DeepSeek undoubtedly demonstrates that AI is undergoing a significant transformation, driving the technology industry to focus more on how to effectively deploy models at the edge for practical applications, thereby accelerating the development of on-device AI. As mentioned in Qualcomm's white paper, "Distillation of large-scale base models has led to a plethora of smarter, smaller, more efficient models, enabling industries to integrate AI at scale more quickly, especially in accelerating integration at the edge. As AI innovation explodes at the edge, Qualcomm's investment in scalable hardware and software will further solidify its leadership position."