Summary of Key Points
Meta's plan to rent out its idle AI computing power has sparked a heated debate in the global capital markets about the future of AI infrastructure. On one hand, Meta's stock price soared (with a daily market value increase of $127 billion), while on the other hand, companies in the computing power supply chain (such as those producing storage chips and optical modules) experienced significant declines in their stock prices. There are concerns that there may be an oversupply of AI computing power, but industry experts argue that this is due to a "structural mismatch" – with a shortage of high-end computing power and a localized surplus of low-end power – and that the long-term demand for such power remains strong. AI investment is shifting from a focus on "who invests the most" to "who invests most efficiently." The ultimate question is whether it will be possible to create a business model where applications generate revenue to cover the costs of infrastructure.
1. Why Does Meta Suddenly Want to Rent Out Its Computing Power? – A Shift in Business Model from "Burn Money" to "Generate Revenue"
Meta has invested heavily in purchasing AI chips (such as GPUs) and building data centers over the past few years, mainly relying on its advertising business to offset these costs (e.g., using AI to optimize ad recommendations). However, it is under increasing pressure from competitors in the field of large-scale models (such as Google's Gemini and OpenAI), and since its advertising business accounts for more than 90% of its revenue, the return on AI investments has been slow.
Therefore, Meta has decided to rent out its idle computing power to generate direct income. These GPUs, which were previously a pure cost burden, have now become assets that can generate revenue. Renting them out also helps to spread costs such as electricity and depreciation. In fact, Elon Musk's xAI project has already been doing this (renting out supercomputing power to companies like Anthropic and Google), and Meta is simply following suit – essentially seeking new ways to monetize its substantial AI investments.
2. Why Is the Market Panicking? – Concerns About an "Oversupply of Computing Power" That Could Hurt Hardware Manufacturers
Meta is one of the largest purchasers of GPUs globally. If even Meta has idle computing power to rent out, investors immediately wonder whether too much AI hardware has been purchased in the past and whether future demand for hardware will decline.
As a result, stocks in companies along the computing power supply chain have seen significant drops: CoreWeave (an American company that rents out computing power) fell 13.9%, Micron (a memory chip manufacturer) fell more than 10%; South Korean companies Samsung and SK Hynix fell 9% and 14.5%, respectively; in the Chinese stock market, companies like GigaDevice Technology and NeoPhotonics, which produce optical modules, also experienced significant declines. There is concern that the logic of continuous hardware expansion may be disrupted, potentially ending the good times for hardware manufacturers.
3. Is There Really an Oversupply of Computing Power? – Industry Experts Say It's a "Structural Mismatch," with a Shortage of High-End Power
Industry experts generally disagree with claims of an overall oversupply, pointing out that the issue is not a lack of computing power in general, but rather a mismatch in distribution:
- Localized surplus of low-end power: Some general-purpose computing powers and smart computing centers without specific use cases have utilization rates of less than 20%.
- Severe shortage of high-end power: There is a gap of about 40% in high-end computing power needed for training large-scale models, and the demand for such power still exceeds supply.
For example, Lenovo's AI server orders amount to 150 billion yuan, and customers have to wait in line due to tight GPU chip supplies; Jiuzhang Cloud Technology has stated that there is a significant shortage of high-end computing power capable of supporting large-scale models. In other words, it's not an overall oversupply, but rather a lack of high-quality power and an excess of lower-quality power.
4. Will AI Capital Expenditure Slow Down? – Not a Peak, but a Shift from "Extensive Expansion" to "Efficiency Improvement"
In the past two years, the AI industry has focused on who invested the most. Now, the focus is shifting to who invests most efficiently:
- Growth will slow down, but it will still occur: Qizhi Consulting predicts that global AI infrastructure investment will maintain double-digit growth from 2024 to 2028, with a potential increase of 51% in 2026 (although this is lower than the 104% forecast for 2025).
- Giants are still increasing their investments: Samsung and SK Hynix plan to invest 4755 trillion won (about 26 trillion yuan) in semiconductors and data centers; Alibaba and Tencent also intend to continue expanding their AI investments.
- Meta's decision to rent out computing power is not a sign of reduced investment: Everbright Securities believes that this move is aimed at optimizing returns, as the revenue from renting out power can help Meta afford new GPU purchases, further strengthening its financial stability. For instance, xAI has reported that it can recoup its costs within two years from renting out its computing power, indicating a high return on investment.
5. The Ultimate Question for the AI Industry: When Can It Become Self-Sustaining?
Whether it's renting out computing power or adjusting capital expenditure, the core question is when AI infrastructure investments will be able to generate revenue through applications.
This is similar to the early days of the internet, where broadband and servers were built first before profitable applications like e-commerce and social media emerged. The same is true for AI:
- There are still few mature consumer-facing AI products (other than popular ones like ChatGPT).
- Business-grade AI solutions (e.g., using AI to optimize manufacturing processes) are just beginning to take off.
- Industry experts argue that a business model where application revenue covers infrastructure costs is necessary for the AI industry to move away from relying on capital-intensive growth and achieve healthy development.
The current debate is not about whether to continue investing in AI, but when the investments will pay off. This is the real turning point for the AI industry.
In conclusion: Meta's decision to rent out its computing power signals a shift in the AI industry from excessive spending to a focus on profitability. There may be short-term market fluctuations, but the long-term demand for high-end computing power remains strong. As long as profitable use cases are found, investment in AI infrastructure will continue. The essence of this debate is whether AI can transform from a concept into a profitable business model.
*(End of translation)*