In the first three quarters of 2025, Meta’s net profit was $37 billion, but CEO Mark Zuckerberg announced that he will invest $60 billion in AI this year.
Profit is far from enough for investment, where does the money come from?
This is not only a problem for Meta, but also a common secret for all the tech giants involved in the AI arms race.
The ticket to the arms race is computing power, and the entity of computing power is thousands of expensive AI chips and the data centers that carry them. Building all of this requires an enormous amount of capital, often billions of dollars. But directly using the company’s reserve cash or issuing bonds on a large scale will immediately worsen the balance sheet and trigger stock price volatility.
So, a core contradiction emerged: how to get a huge amount of funds while keeping the financial report numbers intact? This has become the top financial game for giants. Their solution is to make the debt invisible.
Morgan Stanley estimates that by 2028, the size of such ‘invisible debt’ will reach as high as $800 billion. This huge sum of money actually exists, but it will not appear in the financial reports of tech giants.
How do giants achieve the existence and non existence of huge sums of money?
Advanced Financing Techniques: Hidden Borrowing
The most popular and compliant method currently is the SPV (Special Purpose Vehicle) segregated debt financing method.
You can understand SPV as a godson company recognized by giants outside. The mission of this godson is singular and pure: to borrow money, purchase chips or data centers for his’ biological father ‘, and then rent them out exclusively to his’ biological father’ for long-term use.
The subtlety of this design lies in the fact that the huge assets and their corresponding debts are completely recorded in the accounts of the godson (SPV), completely isolated from the financial statements of the “biological father” (tech giant). This not only achieves risk isolation, such as SPV bankruptcy not affecting the parent company, but also achieves report optimization – the giant only reflects stable rental expenses on the report, rather than heavy asset depreciation and debt interest.
How does this financial technology work? Only four steps are needed:
1、 Recognize a godson. The giant establishes an independent SPV company and promises to sign a long-term, exclusive lease agreement with it. This lease is the foundation for everything that follows.
2、 Borrowing money with ‘dad’s’ credit. SPV itself does not have credit, but it holds long-term leases with giants (i.e. guarantees of stable cash flow in the future) and seeks credit rating agencies.
3、 Obtain a credit rating as a second-generation wealthy individual. Rating agencies evaluate the creditworthiness of the giants behind it. As long as the giant has a good reputation, the bonds issued by SPVs can receive high investment grade credit ratings.
4、 Successful financing. SPVs are able to raise huge amounts of funds from institutional investors at lower interest rates through high rated bonds, and then purchase computing power assets to be delivered to giants for use.
This year, Meta only added $30 billion in new bonds to its public statements through this strategy, but successfully leveraged a total of $60 billion in computing assets through off balance sheet SPVs.
Similarly, using credit to exchange money makes Google easier to play.
It finds cryptocurrency mining companies that have data center assets but weaker credit, and provides debt repayment guarantees to their lending banks: “If they can’t repay, I’ll pay.” This credit endorsement makes banks willing to lend on favorable terms. In exchange, miners will prioritize leasing the built data center capacity to Google. Here, mining companies play another form of non staff SPV.
Giant Collaboration: Circular Financing and Empty handed White Wolf
For startups like xAI founded by Musk, the SPV model is also applicable due to the difficulty of applying for large loans due to their own qualifications.
XAI’s SPV was led by institutions such as Valor Equity Partners and Apollo Global Management, and ultimately raised $20 billion to purchase Nvidia chips and exclusively lease them to xAI. XAI only needs to commit to a 5-year lease agreement and bears almost no direct debt risk.
In this case, a more advanced form of circular financing was also played – essentially building a closed loop for funds to circulate within related parties, enabling one to do business with oneself.
For example, Nvidia can be one of the investors and participate in the establishment of an SPV; SPV will use the funds raised to purchase chips on a large scale from Nvidia. In the end, Nvidia successfully sold its chips and earned revenue and profits; XAI uses top-notch chips and is debt free; As the leading investor of SPV, the investment institution also earns management fees and spreads.
All parties involved are happy.
With this circular law, giants can even leverage capital that far exceeds their actual investment.
In September, Nvidia announced a $30 million investment in Oracle. This investment is seen as a strong industry signal, and the market interprets it as a deep binding between the two giants. As soon as the news came out, Oracle’s stock price skyrocketed by 36%, and its market value skyrocketed by $251 billion overnight. Oracle, in turn, is making large-scale purchases of chips and systems from Nvidia.
In the end, Nvidia sold more products, while Oracle earned hundreds of billions of dollars and gained computing power in one day with just one piece of news, making the two giants rich together.
Just looking at these two months, there have been similar collaborations between giants: OpenAI and Nvidia AMD、 Oracle, SoftBank and other giants have reached infrastructure and chip agreements worth over $1 trillion; Microsoft and NVIDIA jointly invest in Anthropic, while Claude fully integrates into Azure.
This circular financing and strategic alliance have formed a highly bundled capital community among cloud service providers (Oracle, Microsoft Azure), chip manufacturers (Nvidia), and top model companies (OpenAI, Anthropic, xAI).
The giants hope to achieve mutual prosperity through alliances. Behind this also means that risks are unprecedentedly concentrated, and any loss will also result in losses.
Risk: Will history repeat itself?
SPV and off balance sheet financing are not new things, they have triggered crises twice in history and are notorious.
In 2001, energy company Enron established thousands of SPVs, hiding huge losses and debts in these off balance sheet entities, while inflating profits in its own financial reports. When the truth came out, this energy giant quickly collapsed.
In the 2000s, banks packaged a large number of high-risk subprime mortgages into SPVs and used off balance sheet processing to disguise them as high-quality safe assets for sale to global investors. When the housing market collapses, the assets held by these SPVs become poison, ultimately triggering a systemic financial tsunami.
It must be noted that the current AI financing is fundamentally different from the above case.
The operation process of AI giants is reasonable and legal, and the purpose of transactions is to isolate risks and optimize reports, rather than conceal losses or fraud. Its underlying assets are real, high-value hardware, and its cash flow comes from long-term leases of top credit giants.
However, when the scale of financing using this method reaches the trillion dollar level, the risk cannot be ignored.
The lifecycle of AI chips is extremely short and the iteration speed is astonishing. The GPU purchased at a huge cost today may experience a sharp decline in value 18-24 months later with the emergence of a new generation of chips. The risk of rapid depreciation or even elimination of assets held by SPVs has been cleverly transferred to financial investors who ultimately hold bonds.
The lifeline of SPV depends on whether a few tech giants (such as Meta and xAI) can pay rent on time and in full. Once one of the giants defaults due to technological route failure, competitive failure, or failure to establish a business model, it will trigger a systemic disaster of a large number of SPV defaults.
Circular financing deeply binds giants together. Once companies on the chain (such as xAI) are unable to generate expected income to pay rent, risks will flow back along the supply chain, impacting SPV creditors and ultimately affecting chip suppliers and investors, as well as investment institutions.
Returning to the original source: Scaling Law
The origin of everything begins with the industry’s infinite pursuit of the Scaling Law – the belief that the performance of the model continues to improve with increasing parameter size, computational power investment, and data volume. This has driven the exponential expansion of computing power demand and made Nvidia, which has mastered the core hardware, the biggest winner, with its performance soaring all the way.
But top figures in the industry have issued a calm warning.
Google CEO Sandal Pichai accidentally mentioned when releasing the new model Gemini 3.0: “There are indeed irrational factors in the trillion dollar AI investment boom, and no one will be spared once the foam bursts. ”
Regarding this, Chinese researcher Tian Yuandong, who has just been laid off by Meta, has a concise summary:
Scaling Law is a pessimistic future All resources on Earth will be depleted, and all electricity will be used to train large models. In that situation, can we still change our world with this ability? At some point, people will realize that computing power is not everything. ”
As the biggest beneficiary of the arms race, Huang Renxun, CEO of Nvidia, showed full confidence in the latest investor call, saying: “There are many statements about the AI foam, but from our perspective, the situation is quite different from that of the Internet foam. AI is changing the existing workload, and there is no AI foam. ”
But language is the purpose, behavior is the truth. The actual actions of the capital market often reflect real trends more accurately than the public statements of CEOs.
In the past year, a dense high-level reduction in Nvidia stock has quietly occurred:
Huang Renxun himself sold a total of 8.297 million shares in 88 trading days from early 2025 to November, cashing out approximately 1.548 billion US dollars.
Famous investor Duan Yongping and H&H Investment, which he controls, reduced their holdings of Nvidia by 38.04% in a single quarter.
When Nvidia’s stock price reached a historical high in October 2025, Masayoshi Son/SoftBank cleared all 32.1 million shares and cashed out $5.83 billion, essentially doubling their investment.
It is worth noting that Masayoshi Son’s liquidation of Nvidia is not a bearish view on AI. He quickly shifted his funds to a larger bet – leading OpenAI’s $40 billion financing, aiming to become its largest shareholder. This move clearly indicates the flow of capital: they are not withdrawing from AI, but making a bold gamble – switching from hardware chips that provide shovels to applications and models that may mine gold.
At this point, the giants have temporarily solved the financing problem of the shovel through a series of financial techniques. But this is just the prelude to the AI craze. When the torrent of capital crosses over hardware and rushes towards the broader gold mine of models and applications, a more brutal and unpredictable competition has just begun.
No comment content available at the moment