Summary of Key Points
The AI industry is transforming the "computing power demands that have not yet occurred into the future" into "assets that can be financed today," a trend known as "computing power financialization." This is evident in various actions: NVIDIA has partnered with global financial institutions to create a $500 billion financing platform to support AI infrastructure; Volta received a $10 billion long-term computing power contract from Anthropic before even completing its data center; CoreWeave raised $8.5 billion by using GPUs and customer contracts as collateral; Google used the demand for Anthropic's TPU to attract external capital for data center construction; and SoftBank even borrowed $10 billion by mortgaging its unlisted shares in OpenAI. While this approach allows the AI industry to expand rapidly, it also carries significant risks. If future AI demand does not meet expectations, the billions invested today could be lost, potentially impacting the entire financial system.
Detailed Explanation
1. What is "Computing Power Financialization?"
In simple terms, it's about using money that will be earned in the future to build the necessary infrastructure now. Traditionally, infrastructure (such as power plants or highways) is built only after a stable demand has been confirmed. However, AI companies need massive amounts of computing power (for training large models), and since building these centers is costly, they turn the expectation that someone will buy their computing power into an asset that can be used to secure funding from banks or investors. For example, CoreWeave purchased GPUs to build a data center and then signed a long-term contract with Microsoft, using these assets as collateral to borrow $8.5 billion from a bank. It's like saying, "Look, I have the hardware (GPUs) and a customer commitment for future computing power; I'm sure I can make money, so lend me money!"
2. Who is involved in Computing Power Financialization?
- Chip Giants: NVIDIA not only sells GPUs but also acts as a financing intermediary, partnering with companies like Blackstone and Goldman Sachs to create platforms that attract financial investment for AI infrastructure. They profit from GPU sales and gain customer loyalty.
- Startup Companies: Ones like CoreWeave and Volta use GPUs and contracts to raise funds, becoming "owners" of computing power resources.
- Tech Giants: Google uses the demand for Anthropic's TPU to leverage external capital for data center construction, reducing their own financial burden while promoting the TPU ecosystem. Microsoft and Amazon also use long-term contracts to share the costs of building computing infrastructure.
- Investment Institutions: SoftBank has borrowed $10 billion by mortgaging its shares in OpenAI, demonstrating how unlisted equity can be used as a financing tool.
3. Why is Computing Power Financialization Necessary?
The cost of AI infrastructure is extremely high. Companies like Microsoft and Google previously funded these projects out of their own pockets, but now the annual investment in AI has reached tens of billions of dollars, putting significant strain on their cash flows. For example, Microsoft's AI-related expenses increased by 50% in 2024, affecting its free cash flow. Therefore, they are collaborating with others to spread the costs by transforming future computing power demands into assets that can attract investment from banks and private equity firms.
4. What are the Risks of Computing Power Financialization?
The core risk is that future demand may not meet expectations. If this happens:
- Overcapacity: If too many data centers are built but not fully utilized, it could lead to equipment idleness and loan defaults. UBS predicts that cloud companies' capital expenditure will grow by 76% in 2026, then slow down to 25% in 2027, indicating a slowdown in market growth.
- Bond Market Cautiousness: Investors were once eager to buy AI-related bonds (with subscription multiples of up to 5 times), but now the interest is only 2 times, suggesting concerns about the return on investment.
- Differentiated Risks for Various Players: Tech giants like Microsoft and Google have other businesses that support them, so they can withstand losses. However, companies focused solely on computing power (like CoreWeave) are more vulnerable; if demand declines, they could face bankruptcy. SoftBank's loan is also at risk if OpenAI's valuation drops.
- Potential Impact on the Financial System: If many AI infrastructure projects default, it could trigger a financial crisis, similar to the real estate bubble burst.
5. Conclusion
Computing power financialization is both an accelerator and a double-edged sword. It enables rapid expansion of the AI industry by leveraging future demand, but if predictions prove incorrect, it can lead to massive losses. It's like buying concert tickets in advance, hoping they will all be sold, only to find out the artist is not popular and having to repay the loan with no return.