The rapidly evolving landscape of artificial intelligence (AI) brings with it a surge in demand for reasoning capabilities, driven largely by advancements in model training and inference processesRecent findings indicate that as the duration of reinforcement learning time and inference reasoning extends, the performance of models like OpenAI's o1 increases significantlyThis pattern is evident in the DeepSeek series models, which engage in extensive reflection and validation, sometimes producing reasoning chains reaching tens of thousands of words in lengthAs the length of reasoning increases, so does the reliability and efficiency of these modelsThis transition illustrates a critical shift in the scaling laws of AI that is now emphasizing inference rather than merely pre-training computationsTherefore, optimizing computing infrastructure to accommodate this growing requirement for inference is imperative for future AI systems.
Data from Quest Mobile reveals a compelling trend emerging in user engagement with AI native applications, particularly evident in the increasing average usage time and frequency of interactionsBy December 2024, the average user in China was expected to engage with AI applications for approximately 133 minutes per month, an increase of over 53 minutes from January of the same yearThe number of interactions rose dramatically from 26.1 to nearly 50 per month, underscoring a burgeoning demand for robust reasoning capabilities in AI systemsTaking a closer look at ByteDance's Doubao model, predictions for 2025 suggest a staggering daily token invocation of up to 40 trillion, which will significantly escalate the need for inference processing powerThis anticipated rise highlights a pivotal moment in the realm of AI technologies.
The introduction of namely DeepSeek stands to drastically reduce costs associated with inference, catalyzing rapid iterations in training processes
Advertisements
The impetus provided by lowered costs serves to hasten the real-world applications of these AI solutions, facilitating heightened efficiency in reasoning modelsThis burgeoning demand for training-oriented applications is poised to advance industries focused on optical modules, particularly as the intricate dynamics of global economics force a shift towards domestic control in key components like optical chips.
The rapid developments in cloud computing, big data, and AI technologies have precipitated an increased need for fast, efficient, and low-power data transmission solutions, simultaneously driving the market for optical chipsAccording to industry research, the global optical chip market was valued at approximately $2.78 billion in 2023, marking a notable growth of 14.4% from the previous yearBy 2024, this market is expected to reach $3.17 billionAs China accelerates its domestic production capabilities, the optical chip market within the country is also on an upward trajectory, showcasing impressive growth figuresThe Chinese market's value hit around RMB 13.762 billion in 2023, reflecting a rise of 10.24%; estimates for 2024 predict a jump to RMB 15.156 billion.
Currently, domestic enterprises have only maintained substantial control over the core technology pertaining to 2.5G and 10G optical chips, with a phenomenal 90% localization rate for 2.5G and below chipsIn contrast, the domestic production rate for 10G chips stands around 60%, while the localization for chips rated at 25Gbps and above remains relatively underdeveloped at just 4%. As the propagation of inference computing capabilities increases and post-training processing scales up, the market for lower-speed optical chips is predicted to expand further, directly benefiting domestic enterprises focused on this segment.
Amidst this technological upheaval, AIAgents are heralding what many industry experts believe could be a golden age for AI applications
Advertisements
The acceleration of AIAgents, propelled by DeepSeek, marks a significant turning point in the AI landscapeAmidst a backdrop of unprecedented data growth, the evolution and development of AIAgents are becoming increasingly viableRecent data indicate that global data volume hit 103 ZB in 2022, with China generating a substantial 23.9 ZBProjections for 2027 suggest global volumes may soar to 284.3 ZB, with China anticipated to reach 76.6 ZB, significantly outpacing worldwide growth trends.
The Chinese AIAgent market is teeming with opportunities, poised for remarkable expansion in both the enterprise (B2B) and consumer (B2C) segmentsIn 2023, the AIAgent market in China was valued at RMB 55.4 billion, with expectations to reach RMB 852 billion by 2028, translating to a compound annual growth rate (CAGR) of 72.7%. The formal introduction of AIAgents was recognized in 2023, and their ability to redefine vertical domains is gaining tractionMarket forecasts suggest that the AIAgent sector may generate tenfold the revenue of traditional SaaS applications, creating opportunities for over 300 billion-dollar enterprises in the technology sector.
Within the AIAgent market, significant advancements are anticipated on both the B2B and B2C frontsFor B2B applications, AIAgents aim to completely re-engineer SaaS solutionsThis contrasts sharply with traditional database models that rely on structured information management; AIAgents utilize vector databases to automatically learn from and comprehend documents, leading to a more efficient means of knowledge managementSimilarly, AIAgents are being utilized widely as generative AI applications across various industries including e-commerce, education, hospitality, and customer service, responsible for driving significant transformations in traditional sectors.
Meanwhile, the fusion of open-source initiatives and cost-effective solutions fosters the rapid deployment of end-side AI applications
Advertisements
DeepSeek's commitment to large-scale models without a focus on specific verticals or applications positions it as a key enabler for both domestic and international AI applicationsMoreover, end-side AI—integrated into the Internet of Things (IoT)—emerges as a crucial implementation mechanismIn tandem with the ongoing miniaturization and open-source trends championed by DeepSeek, the acceleration of end-side AI becomes increasingly plausible.
Market analysis firm Counterpoint anticipates that end-side AI will drive growth in cellular module shipments, with forecasts indicating that only 6% of cell modules shipped globally will be self-powered AI by 2023, a figure expected to rise to 25% by 2030. Additionally, Counterpoint estimates that the shipment of AI cellular modules could show a remarkable compound annual growth rate of 73% from 2023 to 2027. Industry leaders in IoT and cellular modules are taking proactive steps to pivot their strategies towards end-side AI, anticipating substantial market shifts.
Smart terminals—devices integrating AI technology aimed at executing complex tasks and delivering intelligent services—are witnessing a significant uptrendThese include smartphones, AI PCs, smart wearable gadgets, home automation devices, and in-car information systemsAs AI technology continues to be refined and processor capabilities imbibe transformative power, the market for AI-enabled devices is flourishingAccording to QYR data, the Chinese market for AI smart terminals witnessed sales revenues of approximately RMB 34.4 billion in 2023, with forecasts suggesting a surge to RMB 1.4812 trillion by 2030, reflecting a remarkable CAGR of 37.33%. Of particular note, AI PCs are projected to command a significant market share of 73.88% by 2030.
When juxtaposed with cloud-based AI alternatives, end-side AI demonstrates distinct advantages in terms of cost, energy consumption, and privacy protections
Advertisements
Advertisements