The rapid development of artificial intelligence technology is driving computing demand growth at an unprecedented pace. From large language models to multimodal AI applications, from autonomous driving to smart healthcare, the widespread use of AI technology has placed higher demands on computing capabilities. This article will conduct an in-depth analysis of the driving factors, transmission logic, and future trends of the AI computing power market, providing investors with demand forecasts and investment strategy recommendations for 2024-2026.
Core Driving Factors of AI Computing Demand
1. Continuous Evolution of Large Language Models (LLMs)
The exponential growth in parameter scale of large language models is the primary factor behind the surge in computing demand. From GPT-3's 175 billion parameters to GPT-4's trillions of parameters, the training and inference requirements of models have increased exponentially. In the coming years, even larger models are expected to emerge, which will further drive demand for high-performance GPUs and specialized AI chips.
2. Popularization of Multimodal AI Applications
The integrated application of multiple modalities including text, images, audio, and video is becoming a new direction in AI development. Multimodal models need to handle more complex data types, placing higher demands on computing power. For example, models like OpenAI's GPT-4V and Google's Gemini require powerful computing support, creating significant demand for computing hardware such as GPUs and ASICs.
3. Acceleration of Enterprise Digital Transformation
Enterprises worldwide are accelerating digital transformation, with AI technology becoming a key tool for improving efficiency and innovation. From intelligent customer service to predictive maintenance, from supply chain optimization to financial risk control, AI applications are being implemented across various industries. According to McKinsey research, by 2030, AI could create up to $13 trillion in value for the global economy, which will directly translate into significant investment in computing infrastructure.
4. Intensifying Global AI Competition
Major economies such as China, the United States, and Europe have elevated AI development to national strategic importance, investing huge amounts in AI research and application. Competition between tech giants and startups is driving rapid growth in computing demand. For example, the US is vigorously supporting the semiconductor industry through the "CHIPS and Science Act," while China has included AI as a key development area in its "14th Five-Year Plan." These policy supports will further stimulate computing demand.
Transmission Logic of Computing Demand
1. Relationship Between Computing Demand and AI Model Complexity
The complexity of AI models has an exponential relationship with required computing power. Model parameters, training data, and computational iterations together determine computing demand. Large models based on the Transformer architecture have computational complexity proportional to the square of their parameter count, meaning that doubling the model size may increase computing demand by more than four times. This non-linear growth relationship will lead to explosive growth in future computing demand.
2. Relationship Between Computing Demand and Data Scale
AI training requires large amounts of high-quality data, and the growth of data volume is closely related to computing demand. With the explosive growth of data sources from the Internet of Things, social media, and more, data available for AI training is increasing exponentially. Processing this data requires powerful storage and computing capabilities, forming a positive "data-computing-algorithm" cycle that further amplifies computing demand.
3. Relationship Between Computing Demand and Application Scenarios
Different AI application scenarios have varying computing requirements. The training phase requires large-scale parallel computing, while the inference phase focuses more on low latency and high energy efficiency. Applications with high real-time requirements such as autonomous driving and medical diagnosis need edge computing and specialized chip support, while scenarios like scientific research and large model training require centralized high-performance computing clusters. Diverse application scenarios will drive computing infrastructure toward specialization and diversification.
Computing Demand Forecast for 2024-2026
1. Global Computing Market Size Forecast
According to data from multiple market research firms, the global AI computing market size was approximately $100 billion in 2023 and is expected to grow to $300-350 billion by 2026, with a compound annual growth rate (CAGR) of about 40-50%. Among these, the training computing market will grow faster than the inference market, but as models are optimized and specialized chips develop, the inference market share will gradually increase.
2. Regional Market Differences
North America will continue to dominate the global AI computing market, accounting for about 50-60%, mainly benefiting from the strong R&D capabilities and abundant capital investment of US tech giants. The Asia-Pacific region will become the fastest-growing area, with a projected CAGR of 50-60%, with China, Japan, and Korea as the main growth engines. Although the European market started later, it will maintain stable growth with policy support.
3. Changes in Sub-sector Demand
In sub-sectors, GPUs will remain the mainstream computing hardware, but the share of specialized chips such as ASICs and FPGAs will gradually increase. From an application perspective, large model training, autonomous driving, and smart manufacturing will become the three fastest-growing areas in computing demand. Data centers will develop toward high density and high energy efficiency, with technologies like liquid cooling and modular design becoming mainstream trends.
Scenario Analysis
1. Optimistic Scenario
In the optimistic scenario, AI technology breakthroughs accelerate, and AGI (Artificial General Intelligence) may achieve significant progress before 2026. This will lead to computing demand exceeding expectations, with the market size potentially reaching over $400 billion. GPU supply bottlenecks are alleviated, specialized AI chips see significant performance improvements, and cloud service providers actively expand data centers, creating a positive cycle. US AI chip leaders like NVIDIA and AMD will receive excess returns.
2. Neutral Scenario
In the neutral scenario, AI technology develops steadily as expected, with computing demand maintaining a 40-50% annual growth rate. The market size reaches $300-350 billion by 2026. GPU supply gradually improves, and specialized chips steadily increase market share. Data center construction proceeds as planned, with computing infrastructure and AI applications developing in coordination. US AI leaders maintain stable growth, with all links in the industrial chain benefiting evenly.
3. Pessimistic Scenario
In the pessimistic scenario, AI technology development encounters bottlenecks, computing demand growth slows, with CAGR dropping to 20-30%. The market size reaches only about $200 billion by 2026. GPU supply remains tight, and specialized chip development falls short of expectations. Data center construction slows due to energy constraints and cost pressures. Some small and medium-sized AI companies face financial pressure, and industry consolidation intensifies. The US AI market may experience periodic adjustments, with leading companies performing better than the industry average.
Key Risk Factors
1. Technical Bottlenecks
Moore's Law is slowing down, and chip process advancement is encountering physical limits, with single-chip computing power growth potentially slowing. Meanwhile, energy consumption issues are becoming increasingly prominent, with data centers facing dual pressures from energy supply and carbon emissions. Although algorithm optimization can improve computing efficiency to some extent, it cannot fully offset hardware growth bottlenecks.
2. Supply Chain Risks
The global semiconductor supply chain remains fragile, and geopolitical tensions may lead to restrictions on the supply of critical equipment and materials. High-end computing hardware like GPUs faces insufficient capacity and extended delivery times. Talent shortages also constrain the development of the computing industry, with competition for talent in high-end chip design and AI algorithms becoming increasingly fierce.
3. Policy and Regulation
Government regulation of AI technology is tightening worldwide, with requirements for data privacy, algorithm transparency, security, and ethics potentially increasing compliance costs. Chip export controls and technology blockades may affect the stability of the global computing industry chain. The rise of trade friction and protectionism may hinder global cooperation and exchange in computing technology.
4. Economic Fluctuations
Global economic uncertainty is increasing, with inflationary pressures and rising interest rates potentially affecting corporate IT budgets, especially investment in computing infrastructure. Tech industry layoffs may affect the stability and innovation vitality of AI talent. Venture capital investment in AI startups is becoming more cautious, which may affect the healthy development of the innovation ecosystem.
Investment Strategy Recommendations
1. Industrial Chain Layout
Investors should focus on the entire AI computing industry chain, including upstream chip design, midstream manufacturing, and downstream applications. In the US market, GPU leaders like NVIDIA, AMD, and Intel, as well as chip design companies like Broadcom and Qualcomm, are worth attention. At the same time, AI server manufacturers like Supermicro and Dell, and cloud service providers like Microsoft and Google will also benefit from computing demand growth.
2. Risk Control
Investors should pay attention to valuation risks, as some AI concept stocks are already at historical high valuations. It is recommended to adopt a diversified investment strategy, focusing on leading companies with different technology routes (such as GPU, ASIC, FPGA) and different application scenarios. At the same time, pay attention to companies' technical barriers, customer structure, and profitability, avoiding blindly chasing hot concepts.
3. Long-term Value Investment
The AI computing market has long-term growth potential, and investors should focus on companies with core technologies and continuous innovation capabilities. NVIDIA has strong long-term competitiveness with its CUDA ecosystem and GPU architecture advantages; AMD has gained advantages in the FPGA field through acquiring Xilinx; while tech giants like Google and Microsoft are developing their own AI chips to reduce dependence on third-party suppliers. In the long run, these companies are expected to occupy dominant positions in the computing market.
Conclusion
AI computing demand will maintain strong growth in the coming years, driven by factors including large model evolution, popularization of multimodal applications, enterprise digital transformation, and global AI competition. Despite challenges such as technical bottlenecks, supply chain risks, policy regulation, and economic fluctuations, the computing market still has huge growth potential in the long term. Investors should focus on the entire industry chain layout, control risks, and adhere to long-term value investment principles to seize investment opportunities in the era of AI computing power.
