Summary of Key Points
The current AI competition among large corporations has shifted from a "technological race" to a "capital battle." AI infrastructure (chips, data centers) requires substantial and long-term investments, but the payback period is lengthy, putting significant strain on these companies' free cash flows. To match the duration of their funding needs and alleviate this pressure, large corporations are adjusting their capital structures (raising capital, delaying share repurchases), selling non-core assets, and signing long-term contracts to secure resources. At the same time, there is a divergence in financing costs—companies with better credit scores can borrow money at lower rates. Furthermore, computing power is now being transformed into financial products for early monetization. Capital is no longer an unlimited resource; it has become a limiting factor that determines which companies will be able to withstand until the AI investment begins to generate profits.
1. AI's High Cost Demands Strain on Cash Flows: Changing How Companies Manage Their Money
AI is not a minor endeavor. Building data centers and purchasing GPUs costs millions of dollars, and these investments will only pay off over several years. For example, Alibaba's capital expenditures increased by 75% in the second quarter (reaching 67.6 billion yuan), but its operating cash flow was only 22.9 billion yuan, resulting in a net outflow of 44.6 billion yuan in free cash flow (the money earned minus the money spent). Google's capital expenditures exceeded its operating cash flow in the same quarter, leading to its first negative free cash flow, which prompted it to halt share repurchases. In the past, excess funds were distributed to shareholders; now, the money must be retained for AI investments.
Large corporations don't actually lack funds (Alibaba has 470 billion yuan in cash on its balance sheet), but the problem lies in the "mismatch in the timing of funds." Using short-term cash (such as readily available deposits) to fund long-term AI projects that will take more than a decade to repay reduces the company's available liquidity. As a result, Alibaba has chosen to issue new shares to raise capital, while Google has issued long-term bonds to cover its AI investments with "equity capital without a maturity date" or "long-term debt," thereby preventing the depletion of short-term cash.
2. Long-Term Contracts Lock In Future Expenses
Building AI data centers requires the signing of long-term contracts. Renting space typically spans 20 years, and power purchases take more than a decade; even chip purchases require upfront payments. Once these contracts are signed, companies must make payments regardless of the future performance of their AI businesses. For instance, Meta has signed an unexecuted contract worth 278.9 billion yuan (mainly for data center leases) with a maximum term of 30 years, and Microsoft has contractual obligations totaling 743.8 billion yuan, of which 329.1 billion yuan is for leases that have not yet begun. Even if AI applications prove ineffective, companies cannot cancel these contracts; they must continue to pay for rent and electricity costs. Building an additional data center today could mean paying for it for the next 20 years, resulting in high exit costs.
3. Different Financing Rates for Companies with Different Credit Scores
The need for significant borrowing to expand AI operations means that not all companies can access funds at the same rates. Companies with high credit ratings (such as Google and Amazon) have bond yields of only 2.5%-3.5%. However, Oracle, due to negative free cash flows and high capital expenditures, has a credit rating on the brink of being classified as "junk," and its bond yields are as high as 7%-8%. Similarly, when purchasing GPUs, companies with lower borrowing costs can afford longer payback periods (for example, five years) and can offer lower prices to attract customers, while those with higher borrowing costs may struggle to even cover the costs of building data centers. CoreWeave, for example, faced a difference of more than 2 percentage points in financing rates due to different customer contracts, directly affecting the cost of its computing power expansion.
4. Selling Non-Core Assets to Free Up Funds: Prioritizing Strategic Investments
In the past, large corporations had enough funds to pursue multiple businesses simultaneously (for example, ByteDance invested in gaming, and Alibaba entered the retail sector). Now, with AI requiring concentrated funding, they must sell non-core assets to free up capital. Alibaba sold Lingxi Huyu (which produced a hit game like "Romance of the Three Kingdoms: Strategy Edition"), and ByteDance sold Mutong (whose valuation has even increased). These profitable businesses were cut because their strategic value is lower than that of AI projects.
This is known as "capital allocation"—limited funds mean that for every non-core business retained, less money is available for AI investments. The strategic value of a business alone is not enough; the decision must be based on whether investing in that business is more profitable than in AI.
5. Turning Computing Power into Financial Products: Using Future Revenue Now
Large corporations have devised new ways to utilize their resources. For instance, Nvidia collaborated with financial institutions to create a 500 billion yuan computing power financing platform, where funds are used to purchase GPUs, which are then rented to AI companies (such as xAI). AI companies do not need to pay the full cost upfront but can pay rent instead. Meta's Hyperion project allows funds to hold 80% of the equity while the company leases the equipment itself.
This "financialization of computing power" turns future rental income into immediate construction funds, reducing the company's immediate pressure. However, it also shifts the risk: if future customers do not renew their contracts, the losses may be borne by the investors. Nevertheless, this approach enables AI companies to access more external capital and no longer rely solely on their own cash flows.
6. Who Will Survive in the Long Run?
The technological gap in AI can be closed, but the gap in capital structure is more difficult to bridge. Companies with good credit, substantial assets, and access to cheap long-term funding can afford longer trial and error periods. Those that rely on costly financing have fewer options. The ability to withstand financial pressures essentially boils down to "buying time"—whoever can wait until the AI infrastructure pays off and the models start generating profits will be the winner.
The core message of this article is that the AI competition has shifted from a focus on technology to a focus on capital. Large corporations are using various strategies to address their funding challenges, and these factors will ultimately determine which companies will emerge victorious in the AI era.