Summary of Key Points
Over the past two years, the opportunities in the AI industry have focused on GPU chips. Now, as tech giants such as Meta and Amazon increase their AI capital expenditures, there is a need to assemble GPUs into servers for deployment in data centers, leading to a surge in performance for AI server manufacturers. Companies like Lenovo, Dell, Supermicro Computer, and Foxconn Industrial Internet have seen significant growth in revenue and profits, with order volumes reaching new heights. However, they also face challenges such as tight supply chains, rising prices from upstream suppliers, and the sustainability of long-term demand.
I. AI Server Manufacturers Experience a Surge in Orders and Profits
Recent financial reports from several server manufacturers show impressive growth rates that exceeded expectations:
- Lenovo: Revenue from its Infrastructure Services Group (ISG) reached $8.5 billion, a year-on-year increase of 98%, with potential AI server orders amounting to $54 billion (a 150% increase from the previous quarter).
- Dell: AI server revenue soared by 757% year-on-year to $16.1 billion, and order volumes reached $24.4 billion, causing the company's stock price to rise by 40% after the market closed.
- Supermicro Computer: Revenue nearly doubled, with net profit increasing from $195 million to $1.178 billion (a 500% increase), and backlogs of orders reached a record high.
- Foxconn Industrial Internet: Net profit increased by 95.99% year-on-year, with AI server revenue tripling and GPU cabinet shipments increasing by 3.2 times.
These companies, whether they have their own brands (Lenovo, Dell), specialize in AI solutions (Supermicro), or provide manufacturing services (Foxconn Industrial Internet), are all benefiting from the demand for AI servers, indicating that the focus has shifted from purchasing individual chips to buying complete systems.
II. Why the Sudden Surge?
The reason for this surge is the "downstream shift" in AI computing power requirements. Previously, the focus was on GPUs (such as NVIDIA) because training AI models requires a large number of chips. However, now that giants have acquired these chips, they need to install them in servers and place them in data centers for use—similar to how buying a graphics card requires a computer to function properly.
Companies like Meta, Amazon, and Tencent are increasing their AI investments this year, not only for chips but also for servers capable of supporting these chips. Training large models may require thousands of servers, while inference tasks (such as using AI to answer questions or generate content) require even more servers distributed across various systems. As a result, there has been a sudden surge in demand for servers.
III. Short-Term Challenges: Supply Chain Bottlenecks and Rising Prices
AI servers are not simply assembled; they require multiple components such as GPUs, memory, storage, and cooling systems. The current biggest issue is the tight supply of storage components:
- Storage chip prices have already increased, and there is expected to be a shortage next year.
- Manufacturers are forced to "lock the quantity but not the price" to ensure they can obtain the necessary components, which reduces their profit margins.
- Some customers place orders in advance due to FOMO (fear of missing out), although they may not actually need the products immediately; demand could fluctuate later on.
In short, the problem is not a lack of orders but whether the necessary parts can be obtained to fulfill them and whether profits can be made after completion.
IV. Can the Growth Sustain in the Long Term?
The key to long-term growth lies in two factors:
1. Inference demand has yet to peak: Lenovo's chairman, Yang Yuanqing, noted that current AI usage is 50% for training and 50% for inference, with this ratio expected to shift to 80% for inference and 20% for training in the future. For example, using ChatGPT or generating images with AI involves inference tasks, and these demands have not yet started on a large scale; therefore, the purchase of inference servers by corporate customers is just beginning.
2. Sustainability of capital expenditures: If AI investments generate profits (for instance, by helping companies reduce costs and increase efficiency), giants will continue to invest, sustaining server demand. However, if AI returns are poor, they may cut budgets, leading to a decline in server orders.
According to iMedia Research, the market is concerned about whether the significant investment in AI will be profitable. If not, the rapid growth of the server industry could come to an end.
Conclusion
AI servers are currently in a position of great opportunity, but whether this growth can continue depends on whether the supply chain can keep up, whether rising prices from upstream suppliers can be absorbed, and whether the commercial value of AI can be effectively realized. For the general public, this means that the AI industry has moved from a "chip competition" to a phase of "actual application deployment." The performance growth of related companies is real, but the risks cannot be ignored.