Summary of Key Points
Recently, AI has evolved beyond being merely a chatbot to become intelligent agents capable of performing tasks like “digital employees,” sparking a global boom in AI infrastructure development. Chip manufacturers such as TSMC report that demand for chips far exceeds supply, and companies that sell complete servers, like Dell and Lenovo, have seen a surge in AI-related orders. However, the widespread adoption of AI agents across various industries still faces challenges, including high computational costs, difficulty in measuring business value, and poor industry-specific adaptability.
1. AI Agents are Hot, with Dell and Lenovo Seeing a Surge in AI Server Orders
The immediate beneficiaries of the growing demand for AI agents are companies that sell servers:
- Dell: Revenue from AI servers increased by 757% (more than sevenfold) in the latest quarter, with orders totaling $24.4 billion, and unpaid orders amounting to $51.3 billion. This has raised Dell’s annual revenue forecast from $50 billion to $60 billion.
- Lenovo: AI-related revenue increased by 84%, accounting for nearly 40% of total sales, with unpaid AI server orders totaling $21 billion.
Why are they making such profits? Because AI agents not only need to be trained but also must be able to apply that knowledge to solve practical problems and be customized for enterprises. Server manufacturers can integrate high-performance chips, liquid cooling systems, and advanced power supplies, making their solutions more attractive to customers willing to pay a premium.
2. Changing Demands: AI Infrastructure Requires More Than Just Training Power
The focus has shifted from simply training AI models to a comprehensive range of services:
- Not only model training but also inference (using learned knowledge to solve real problems), new cloud services (Neocloud), on-premises deployments, and data center construction are all in demand.
Experts suggest that the benefits from individual chips are diminishing, with entire server racks becoming the new “super chips.” As server power consumption rises, advanced technologies such as liquid cooling and high-speed data transmission are essential for optimizing rack performance. Companies that can effectively integrate these components will gain a competitive advantage.
3. Which Industries Can Benefit from the Agent AI Era?
Leaders in the tech industry, like NVIDIA’s Jensen Huang and Qualcomm, have declared the arrival of the Agent AI era, which is driving growth in several niche areas:
1. Power Supplies: AI servers require high-power, reliable power solutions, especially those that use high-voltage DC technology.
2. Liquid Cooling: As fan-based cooling reaches its limits, companies specializing in liquid cooling (e.g., cold plates and immersion systems) are poised for success.
3. High-Speed Interconnectivity: Companies providing fast data transfer solutions for data centers will benefit from this trend.
4. Chip Specialization: In addition to GPUs, companies developing inference chips and integrated memory-compute chips can also find opportunities in the new hardware ecosystem.
5. Advanced PCBs: AI devices require more sophisticated circuit boards, benefiting manufacturers in these fields.
4. Challenges to Overcoming in the AI Agent Boom
Despite the enthusiasm, there are several practical hurdles to overcome:
1. High Computational Costs: Companies spend millions of dollars per month on AI (e.g., UCloud’s CEO reported spending over $600,000 per week), yet the benefits are not always immediate and significant.
2. Difficulty in Measuring Value: The value of many AI applications is hard to quantify, leading customers to hesitate with spending.
3. Technological Reliability: Industries like healthcare require 100% accuracy from AI systems, but current large models often produce inaccurate results, limiting their widespread use.
4. Lack of Talent and Organizational Skills: Enterprises need professionals with both AI and industry expertise, as well as the ability to adapt business processes and manage data effectively.
IDC notes that most AI applications are not yet at scale, and Gartner predicts that 40% of AI agent projects will be canceled by 2027 due to high costs and unclear benefits. This means the current boom is more of a long-term investment than a quick profit opportunity.
In summary, while AI agents represent a significant trend, their practical adoption requires addressing these challenges: reducing infrastructure costs, establishing viable business models, and ensuring proper adaptation to industry needs. Only then can AI agents transition from being a concept to a truly useful tool in real-world applications.