AI Computing Power Bubble? Or the Biggest "Productivity Leverage Experiment" in Human History? - Dissecting the $30 Trillion Financial Long March

Original Title: Is AI Infrastructure a Bubble, or a “Collective Buy Time”? Analyzing the Financial Structure Behind the $30 Trillion

Original author: Distill AI

Source text:

Reprinted: Daisy, Mars Finance

As tech giants spend over $300 billion this year on AI computing power, with total expected investments exceeding $30 trillion over the next three years, a question arises: Is this a replay of the 2000 internet bubble, or the largest bet on productivity in human history?

This is not a simple binary debate of “bubble vs non-bubble”; the answer may be more complex and nuanced than you think. I do not have a crystal ball to predict the future. But I try to delve into the underlying financial structure of this feast and construct a framework for analysis.

The article is long and detailed, so let's start with the conclusion:

In terms of direction, I don't think this is a big bubble. However, there are high risks in certain segments.

To be more precise, the current AI infrastructure resembles a long march of “banding together + buying time.” Major companies (( Microsoft, Google, Meta, Nvidia, etc.) leverage massive amounts of financial engineering, but outsource the main credit risk to special purpose vehicles (SPV) and capital markets, tightly binding the interests of all participants.

The so-called “buying time” refers to betting on whether their cash flow and external resources can endure until the day when “AI truly enhances productivity” arrives.

If the bet wins, the AI will fulfill its promise, and the big companies will be the biggest winners. If the bet loses (the AI's progress is not as expected or the costs are too high), the first to be hurt will be the external resources that provide financing.

This is not the kind of “bank leverage excess, single point explosion” bubble of 2008. This is a gigantic experiment in direct financing, led by the smartest and most cash-rich entrepreneurs on the planet, using complex “off-balance-sheet financing” strategies to break down risks into many tradable fragments, distributed for different investors to digest.

Even if it's not a bubble, it doesn't mean that all AI infrastructure investments can achieve a good ROI.

01 Understanding the core: The benefit binding mechanism of “grouping together”

The so-called “team-up” refers to the close binding of the interests of five parties in this AI infrastructure.

Tech giants (Meta, Microsoft, Google) and their large model partners (OpenAI, xAI): need computing power but don't want to spend a large amount at once.

Chip supplier (Nvidia): Needs continuous large orders to support its valuation.

Private equity funds (Blackstone, Blue Owl, Apollo): Need new asset classes to expand asset management scale and collect more management fees.

Neocloud (CoreWeave, Nebius) and hybrid cloud service providers (Oracle Cloud Infrastructure): provide infrastructure and computing power, but at the same time require large companies to sign long-term contracts to leverage financing.

Institutional investors (pension funds, sovereign funds, traditional funds like BlackRock): require stable returns higher than government bonds.

And these five parties formed a “community of shared interests”, for example:

Nvidia prioritizes supply to CoreWeave while investing in its equity.

Microsoft provides a long-term agreement to CoreWeave while assisting with its financing.

Blackstone provides debt financing while raising funds from pension funds.

Meta and Blue Owl jointly established an SPV to share risks.

OpenAI, along with other large model vendors, continues to raise the standards for model parameters, inference capabilities, and training scale, which effectively increases the overall industry's computational power demand threshold. Especially under the deep binding with Microsoft, this “technology outsourcing, internalized pressure” cooperation structure allows OpenAI to become the igniter of the global capital expenditure competition without spending money. It is not a funding party, but the actual curator driving the overall leverage increase.

No one can stand alone; this is the essence of “banding together.”

02 Capital Structure — Who is funding? Where is the money flowing?

To understand the overall architecture, we can start with the funding flow chart below.

Tech giants need astronomical computing power, and there are two paths:

Self-built data center: This is the traditional model. The advantage is complete control, while the disadvantage is slow construction, and all capital expenditures and risks are borne on one's own balance sheet.

Seeking External Supply: Giants are not simply renting servers, but have spawned two core “external vendor” models. This is the current trend and also the focus of our analysis.

The first type is SPV (Special Purpose Vehicle) / Special Purpose Entity, which is purely a financial instrument. You can think of it as a special entity established specifically for a “single project, single client.”

Business model: For example, if Meta wants to build a data center but does not want to spend a large sum of money at once, it can partner with an asset management company to form a Special Purpose Vehicle (SPV). The sole task of the SPV is to construct and operate this center exclusively for Meta. Investors receive high-quality debt instruments (a mix of corporate bonds and project financing) backed by rental cash flow.

Customer type: extremely singular, usually only one (e.g. Meta).

Risk level: Life and death completely depend on the credit of a single customer.

The second type is Neocloud ( like CoreWeave, Lambda, Nebius ), which are independent operating companies (Operating Company, OpCo) with their own operational strategies and full decision-making authority.

Business Model: For example, CoreWeave raises funds (equity and debt) to purchase a large number of GPUs and leases them to multiple clients, signing “minimum guarantee/reservation” contracts. It is flexible but the value of equity fluctuates greatly.

Customer type: theoretically diverse, but in reality, it heavily relies on large companies in the early stages (for example, Microsoft’s early support for CoreWeave). Due to its smaller scale, unlike SPV which has a single wealthy backer, Neocloud is more dependent on upstream suppliers (Nvidia).

Risk level: The risk is diversified among multiple clients, but operational capacity, technology, and equity value all affect survival.

Although they are fundamentally different in terms of legal and operational structures, the business essence of both converges: they are both “external suppliers of computing power” for the giants, removing massive GPU purchases and data center constructions from the giants' balance sheets.

So where does the money for these SPVs and Neoclouds come from?

The answer is not traditional banks, but private credit funds. Why?

This is because after 2008, the Basel III Accord imposed strict requirements on banks' capital adequacy ratios. Banks taking on such high-risk, high-concentration, long-term massive loans are required to set aside reserves that are not cost-effective.

The businesses that banks “cannot do” and “dare not do” have created a huge vacuum. Private equity giants like Apollo, Blue Owl, and Blackstone have filled the gap—they are not restricted by bank regulations and can offer more flexible and faster financing, albeit at higher interest rates. Secured by project rents or GPU/equipment with long-term contracts.

For them, this is an extremely attractive pie - many have traditional infrastructure financing experience, and this theme is sufficient to grow the scale of asset management several times, significantly increasing management fees and carried interest (Carried Interest).

So where does the money for these private equity credit funds ultimately come from?

The answer is institutional investors (LPs), such as pension funds, sovereign wealth funds, insurance companies, and even general investors (for example, through the private credit ETF issued by BlackRock - which includes the 144A private debt Beignet Investor LLC 144A 6.581% 05/30/2049 under the Meta project).

The transmission path of risk chain is thus established:

( ultimate risk bearer ) pension funds/ETF investors/sovereign funds → ( intermediary institutions ) private credit funds → ( financing entities ) SPV or Neocloud ( such as CoreWeave ) → (end users) tech giants ( such as Meta )

03 SPV Case Study — Meta's Hyperion

To understand the SPV model, Meta's “Hyperion” project is an excellent case study (with enough public information):

Structure/Equity: Meta and Blue Owl manage the fund group JV (Beignet Investor LLC). Meta holds 20% equity, Blue Owl 80%. Bonds are issued under the SPV 144A structure. The JV covers the assets, and Meta leases them under a long-term agreement. Capital expenditures during the construction period are in the JV, and assets gradually transfer to Meta's balance sheet after financing leases begin.

Scale: Approximately $27.3 billion in debt (144A private placement bonds) + approximately $2.5 billion in equity, making it one of the largest single corporate bond/private credit project financings in U.S. history. The maturity date is in 2049, and this long-term amortization structure essentially locks in the most challenging time risk first.

Interest Rate/Rating: Debt rated A+ by S&P (high rating allows insurers to allocate), with a coupon rate of approximately 6.58%.

Investor structure: PIMCO subscribed 18 billion; BlackRock's ETFs totaled over 3 billion. For this group of investors, this is a highly attractive high-quality stable return.

Cash Flow and Lease: What Blue Owl is interested in is not the potentially depreciating GPUs (I believe some people in the market are misfocusing on the assumption that the depreciation period for GPUs is too long, because GPUs are just the hardware part, while the overall value of AI lies in hardware + models. The price of older hardware decreases due to iterations, but this does not mean that the value of the final AI model application also decreases), but rather the SPV cash flow supported by Meta's long-term lease (starting from 2029). Funding during the construction period is also allocated in U.S. Treasury bonds to reduce risk. This structure integrates the liquidity of corporate bonds with protective clauses for project financing, and is also 144A-for-life (limited to a circle of investors).

So why is the short-term risk of this architecture extremely low?

This is because under this structure, the Hyperion task is simple: collect Meta rent with the left hand and pay Blue Owl interest with the right hand. As long as Meta doesn't collapse (the likelihood in the foreseeable future is extremely low), the cash flow remains as solid as a rock. There's no need to worry about fluctuations in AI demand or GPU price drops.

This 25-year ultra-long maturity, rent-repayment debt structure locks in all recent refinancing risks as long as the rent comes in steadily and interest is paid normally. This is the essence of “buying time” (allowing the value created by AI applications to gradually catch up with the financial structure).

At the same time, Meta uses its own credit and strong cash flow to secure substantial long-term financing that bypasses traditional capital expenditures. Although under modern accounting standards (IFRS 16), long-term leases ultimately still appear on the balance sheet as “lease liabilities,” the advantage is that the pressure of capital expenditures amounting to billions of dollars during the initial construction phase, as well as the associated construction risks and financing operations, are first transferred to the SPV.

Transforming a one-time substantial capital expenditure into lease expenses amortized over the next 25 years greatly optimizes cash flow. Then, bet on whether these AI investments can generate enough economic benefits in 10-20 years to pay back the principal and interest (considering a bond with a coupon rate of 6.58%, the ROI calculated based on EBITDA must be at least 9-10% to provide a decent return for equity holders).

04 Neocloud's Cushion — OpCo's Equity Risk

If the SPV model is “credit transfer”, then CoreWeave, Nebius, and similar Neocloud models represent “further stratification of risk.”

Taking CoreWeave as an example, the capital structure is much more complex than that of an SPV. Multiple rounds of equity and debt financing involve investors such as Nvidia, venture capitalists, growth funds, and private debt funds, creating a clear sequence of risk buffers.

What would happen if AI demand is not as expected, or new competitors emerge, causing CoreWeave's revenue to plummet and unable to pay high interest?

The first step is the evaporation of equity value: CoreWeave's stock price plummets. This is the “equity cushion” — the first to absorb the impact. The company may be forced to finance at a discount, significantly diluting the equity of original shareholders, or even losing all their capital. In contrast, the SPV has a thinner equity cushion, as it cannot directly raise funds in the public market.

The second step is the creditor's loss: Only after the equity is completely “burned out” and CoreWeave is still unable to repay the debt, will it be the turn of private creditors like Blackstone to bear the loss. However, these funds usually require excellent collateral (latest GPUs) and strict repayment priority when lending.

CoreWeave and Nebius both adopt the approach of “first securing long-term contracts, then financing against those contracts,” allowing for rapid expansion through refinancing in the capital markets. The brilliance of this structure lies in the fact that large clients can achieve better capital utilization efficiency, leveraging future procurement contracts to unlock more capital expenditures without direct investment, thereby limiting the risk contagion to the entire financial system.

On the contrary, Neocloud shareholders need to be aware that they occupy the most turbulent yet thrilling position in this gamble. They are betting on rapid growth while also praying that the management's financial operations (debt extension, equity issuance) are nearly flawless. Additionally, they must pay attention to the debt maturity structure, pledge range, contract renewal windows, and customer concentration in order to better assess the risk-reward ratio of equity.

We can also imagine who would be the marginal capacity most easily abandoned if the demand for AI grows slowly. SPV or Neocloud? Why?

05 Oracle Cloud: The Rise of an Atypical Cloud Player

While everyone is focusing on CoreWeave and the three major cloud giants, an unexpected “dark horse” in the cloud is quietly rising: Oracle Cloud.

It does not belong to Neocloud, nor is it part of the first-tier camp of the three major tech giants, but it has secured contracts for a portion of the computing load from Cohere, xAI, and even OpenAI through its highly flexible architecture design and deep collaboration with Nvidia.

Especially when the leverage of Neocloud gradually tightens and traditional cloud space is insufficient, Oracle, positioned as “neutral” and “replaceable,” becomes an important buffer layer in the second wave of the AI computing power supply chain.

Its existence also shows us that this power struggle is not just a showdown among the three giants, but also that non-typical yet strategically significant suppliers like Oracle are quietly vying for position.

But don't forget, the game table is not just in Silicon Valley, but extends to the entire global financial market.

The government's “implicit guarantee” that everyone is coveting.

Finally, in this game dominated by tech giants and private finance, there is a potential “trump card” - the government. Although OpenAI recently publicly stated that it “does not have and does not hope” for the government to provide loan guarantees for data centers, the discussions with the government are about potential guarantees for chip factories rather than data centers. However, I believe that their (or similar participants') original plan must have included the option of “bringing the government in to form a coalition.”

How to say? If the scale of AI infrastructure becomes so large that even private equity cannot bear it, the only way out is to upgrade to a contest of national strength. Once the leadership position of AI is defined as “national security” or “the lunar competition of the 21st century,” government intervention becomes logical.

The most effective way to intervene is not to directly provide money, but to offer “guarantees.” This approach brings a decisive benefit: significantly reducing financing costs.

Investors around my age should still remember Freddie Mac ( and Fannie Mae ). These two “Government Sponsored Enterprises (GSEs)” are not formal departments of the U.S. government, but the market generally believes they have an “implicit government guarantee.”

They purchase mortgages from banks, package them into MBS and guarantee them, then redirect the capital back to the mortgage market after selling them in the open market, increasing the funds available for lending. Their existence is what made the impact of the financial tsunami in 2008 even greater.

Imagine a future where there is a “National AI Computing Company” backed by implicit guarantees from the government. The bonds it issues would be regarded as quasi-sovereign debt, with interest rates approaching those of U.S. Treasury bonds.

This will fundamentally change the previously mentioned “buying time to wait for productivity to rise”:

The cost of financing is extremely low: the lower the borrowing cost, the lower the requirement for the “speed of AI productivity improvement”.

Unlimited extension of time: More importantly, it allows for continuous rollovers at extremely low costs, which is equivalent to buying nearly unlimited time.

In other words, this approach significantly reduces the chances of the gamble “blowing up”. However, once it does blow up, the impact could expand by dozens of times.

$6 trillion bet — the real key to “productivity”

All the aforementioned financial structures - SPV, Neocloud, private placement debt - no matter how sophisticated, are merely answering the question of “how to pay.”

The fundamental question of whether AI infrastructure will become a bubble is: “Can AI really increase productivity?” and “How fast is it?”

All financing arrangements lasting 10 or 15 years essentially “buy time.” Financial engineering gives giants a breathing space, without the need for immediate results. However, buying time comes at a cost: investors in Blue Owl and Blackstone (pension funds, sovereign funds, ETF holders) require stable interest returns, while equity investors in Neocloud are looking for several times the valuation growth.

The “expected return rate” of these financing parties is the threshold that AI productivity must overcome. If the productivity improvements brought by AI cannot cover the high financing costs quickly enough, this intricate structure will begin to collapse from its most vulnerable point (the “equity buffer”).

Therefore, in the coming years, special attention should be paid to the following two aspects:

The speed of launching “application solutions” in various fields: having a powerful model (LLM) is not enough. We need to see real “software” and “services” that can make enterprises spend money. These types of applications need to be widely adopted, generating cash flow that is large enough to repay the principal and interest of the huge infrastructure costs.

External constraints: AI data centers are electricity monsters. Do we have enough power to support the exponentially growing demand for computing power? Is the upgrade speed of the power grid keeping up? Will the supply of Nvidia's GPUs and other hardware face bottlenecks, causing delays compared to the timelines required by financial contracts? Supply-side risks may drain all the “bought time.”

In short, this is a race between finance (funding costs), physics (electricity, hardware), and commerce (application landing).

We can also use a quantitative approach to roughly estimate how much productivity improvement AI needs to bring to avoid a bubble:

According to Morgan Stanley's estimates, this round of AI investment is expected to reach 3 trillion dollars by 2028.

The aforementioned SPV bond issuance cost of Meta is approximately 6-7%, while according to a report by Fortune, CoreWeave's current average debt interest rate is around 9%. Assuming that most private debt in the industry demands a return of 7-8% and an equity-to-debt ratio of 3:7, the ROI of these AI infrastructures (calculated using EBITDA and total capital expenditure) needs to be at 12-13% to achieve an equity return rate of over 20%.

So the required EBITDA = 3 trillion × 12% = 360 billion; if calculated at an EBITDA profit margin of 65%, the corresponding revenue is approximately 550 billion.

With an estimated GDP of about 29 trillion in the name of the United States, an additional output equivalent to approximately 1.9% of GDP needs to be sustainably supported by AI.

This threshold is not low, but it is not a fantasy. In 2025, the global cloud industry is expected to generate approximately 400 billion dollars, in other words, we need to see at least one to two cloud industries transformed by AI. The key lies in whether the speed of application monetization can be synchronized with the physical bottlenecks.

Risk Scenario Stress Testing: When is there not enough “time”?

All the financial structures mentioned above are betting that productivity can outpace financing costs. Let me use two stress tests to simulate the chain reaction when the speed of AI productivity realization is not as expected:

In the first scenario, we assume that AI productivity is realized “slowly” (for example, it takes 15 years to achieve scaling, but much of the funding might be for a 10-year term):

Neocloud was the first to fall: Independent operators like CoreWeave, due to their revenue being unable to cover high interest rates, have burned through their “equity cushion,” leading to debt defaults or distressed restructurings.

SPV faces maturity risk: For SPV debt like Hyperion, Meta must decide whether to refinance at a higher interest rate (the market has already witnessed the failure of Neocloud), which would erode core business profits.

Private credit fund LPs have suffered huge losses, and tech stock valuations have been significantly revised downwards. This is an “expensive failure,” but it will not trigger a systemic collapse.

In the second scenario, we assume that AI productivity has been “falsified” (technological progress stagnates or costs cannot be reduced and scaled):

Tech giants may opt for “strategic default”: this is the worst-case scenario. Giants like Meta may determine that “continuing to pay rent” is a bottomless pit, and subsequently choose to forcibly terminate leases, forcing SPV debt restructuring.

SPV Bond Crash: Bonds like Hyperion, which are considered A+ rated, will instantly decouple from Meta, causing prices to plummet.

It could completely destroy the private equity “infrastructure financing” market and is highly likely to trigger a crisis of confidence in the financial markets through the aforementioned interconnectedness.

The purpose of these tests is to transform the ambiguous question of “whether it is a bubble” into specific situational analyses.

07 Risk Thermometer: A Practical Observation Checklist for Investors

As for the changes in market confidence, I will continue to focus on five things as a risk thermometer:

The speed of achieving productivity in AI projects: including the acceleration or deceleration of expected revenue from major model vendors ( linear growth or exponential growth), and the application status of different AI products and projects.

Neocloud company's stock price, bond yield, announcement: including large orders, defaults/amendments, debt refinancing (some private bonds are set to mature around 2030 and require special attention), and capital increase pace.

SPV bond secondary price/spread: whether 144A private placements like Hyperion maintain a price above par, whether trading is active, and whether ETF holdings are increasing.

Changes in the quality of long-term contracts: take-or-pay ratio, minimum retention period, customer concentration, price adjustment mechanisms (adjustments to electricity prices/interest rates/pricing in relation to inflation).

Power progress and potential technological innovations: As the most likely external factor to become a bottleneck, attention needs to be paid to policy signals regarding substations, transmission and distribution, and electricity pricing mechanisms. Additionally, it should be considered whether there are new technologies that can significantly reduce electricity consumption.

Why is this not a repeat of 2008?

Some people might draw parallels to the bubble of 2008. I think this approach could lead to misjudgment:

The first point lies in the fundamentally different nature of core assets: AI vs. housing.

The core asset of the 2008 subprime mortgage crisis was “housing.” Housing itself does not contribute to productivity (rental income growth is very slow). When housing prices deviate from the fundamentals of residents' income, and are bundled into complex financial derivatives, it is only a matter of time before the bubble bursts.

The core asset of AI is “computing power.” Computing power is the “tool of production” in the digital age. As long as you believe that AI has a high probability of significantly increasing overall productivity in society (software development, drug research and development, customer service, content creation) at some point in the future, there is no need to worry too much. This is a “prepayment” for future productivity. It has real fundamentals as an anchor point, but it has not yet been fully realized.

The second point is that the key nodes of the financial structure are different: direct financing vs. banks.

The 2008 bubble spread significantly through key nodes (banks). Risks were transmitted through “interbank indirect financing.” The bankruptcy of one bank (like Lehman) triggered a crisis of confidence in all banks, leading to a freeze in the interbank market, ultimately igniting a systemic financial crisis that affected everyone (including a liquidity crisis).

Currently, the financing structure of AI infrastructure is mainly based on “direct financing”. If AI productivity is discredited, CoreWeave goes bankrupt, and Blackstone defaults on its $7.5 billion debt, it will result in significant losses for Blackstone investors (pension funds).

The banking system has indeed become stronger after 2008, but we cannot oversimplify and think that risks can be completely “contained” in the private equity market. For example, private credit funds themselves may also amplify returns using bank leverage. If AI investments generally fail, these funds could still incur huge losses that might spill over through two pathways:

Leverage default: The fund's default on leveraged financing from the bank will transmit risk back to the banking system.

LPs Impact: Pension funds and insurance companies have seen their balance sheets deteriorate due to significant investment losses, causing them to sell off other assets in the public market, triggering a chain reaction.

Therefore, a more accurate statement would be: “This is not the kind of interbank liquidity crisis that caused a single-point detonation and a complete freeze in 2008.” The worst-case scenario would be “expensive failures,” with lower contagion and slower speed. However, given the opacity of the private placement market, we must remain highly vigilant about this new type of slow contagion risk.

Insights for Investors: Which layer are you on in this system?

Let's return to the initial question: Is AI infrastructure a bubble?

The formation and bursting of bubbles stem from the huge gap between expected benefits and actual results. I believe that, on a larger scale, it is not a bubble, but rather a precise high-leverage financial layout. However, from a risk perspective, aside from certain aspects that require particular attention, we should also not take lightly the “negative wealth effect” that small-scale bubbles may bring.

For investors, in this multi-trillion dollar AI infrastructure race, you must understand what you are betting on when holding different assets:

Tech giant stocks: You are betting that AI productivity can outpace financing costs.

Private credit: You earn stable interest, but bear the risk of “time may not be enough.”

Neocloud Equity: You are the first buffer of highest risk and highest reward.

In this game, position determines everything. Understanding this series of financial structures is the first step to finding your own position. Knowing who is “curating” this show is the key to determining when this game will end.