Structured financing, which uses chips as collateral, is becoming the core funding engine for the global AI expansion and is rapidly developing into a brand-new asset class that could exceed one trillion dollars in size.

The embryonic form of this market began in 2023 with a $2.3 billion private credit facility provided by Blackstone to CoreWeave, marking the first significant transaction involving chip-backed debt.

Today, Broadcom is in negotiations with Wall Street institutions to arrange $60 billion in chip financing for Anthropic, while SpaceX is reportedly seeking $40 billion in financing for NVIDIA chips. NVIDIA itself has also announced a cooperation framework worth $500 billion with six major lending institutions.

JPMorgan analysts predict that total chip spending could exceed $3 trillion by 2030, with a large portion of it relying on debt financing.

However, this debt expansion also brings risks that cannot be ignored. Since most of these transactions are private placements, investors lack convenient exit channels; the cross-exposure among chip manufacturers, cloud computing giants, and emerging cloud service providers is highly concentrated.

CoreWeave Corporate-level junk bonds have recently tumbled significantly, with some falling to 80 cents face value, and yields climbing to around 15%, reflecting the market's deep-seated doubts about the commercial returns of AI.

A financing deal that creates a new paradigm

The starting point of this revolution was the cautious assessment of a unfamiliar concept by Blackstone's Tactical Opportunities team. According to the recollection of CoreWeave, co-founder of the time, who was involved in the due diligence process along with Brannin McBee, Blackstone executives directly asked questions at that time: "What is GPU?" "Does AI really exist?"

After obtaining a satisfactory answer, Blackstone led the completion of a $2.3 billion private credit deal for CoreWeave in 2023, setting a precedent for debt financing secured by chip collateral.

The core logic of this structure is as follows: A special purpose vehicle ( SPV ) is established to purchase and hold chips, and the cash flow generated from target customer contracts is used to repay debts. The debt term is matched with the five to six-year depreciation cycle of the chips, thereby keeping the risk of hardware depreciation within a predictable range.

The model of CoreWeave is derived from experience in the commodity market – similar to the development model in the liquefied natural gas industry, which involves ‘signing a contract first, then financing to build a plant’. The first transaction was secured by a Microsoft contract, and at that time, the financing cost was as high as double-digit interest rates. McBee frankly described this as ‘the most challenging due diligence process of my career’.

As the customer base of CoreWeave continues to expand, its financing costs have decreased accordingly. In March this year, the company completed $8.5 billion in debt financing supported by a Meta contract, with the borrowing interest rate dropping to around 6% – Wall Street priced the risk based on Meta's high rating rather than CoreWeave's own junk rating.

Chip costs far exceed those of data centers.

To understand the scale of this financing revolution, one must first comprehend the magnitude of chip expenditures. The cost of chips is two to four times that of the data centers that house them, yet these costs are fully depreciated within five to six years, creating a huge financing gap that traditional capital markets find difficult to absorb on their own.

Taking the flagship project of cloud service provider AI in West Virginia as an example, the supporting power plant is expected to cost $6.5 billion, while the construction cost for data centers and equipment amounts to $18 billion. Furthermore, the procurement costs for GPU are as high as $47 billion.

Google, Microsoft, Amazon, and other investment-grade cloud computing giants under Meta can rely on ample cash or cheap bond financing to cover the aforementioned hardware costs, but OpenAI, Anthropic, and a host of emerging cloud service providers do not possess such financial strength in their balance sheets.

OpenAI Business Finance Officer Mike Liberatore frankly stated: "The scale of infrastructure construction is unprecedented. There is no script that tells you how to move from scratch towards general artificial intelligence using traditional methods in an environment that is extremely capital-intensive."

For this reason, the industry has adopted mechanisms from project financing and structured financing: by using SPV, the risk of each transaction is isolated. Borrowing is based on the quality of the transaction itself and the creditworthiness of the customer, rather than the issuer's own rating, and such borrowing is often not included in the company's balance sheet.

Chip manufacturers directly get involved and provide endorsement.

The deep involvement of Broadcom and NVIDIA has pushed this market to a new level. Both companies have provided 'guarantee' commitments for the related debt transactions:

In the event of a borrower's default, the chip will be re-leased or sold to repay the remaining debt, and the chip manufacturer is responsible for covering the shortfall. This mechanism allows the priority bonds issued by SPV to be linked to the chip manufacturer's investment-grade rating, significantly reducing financing costs.

Apollo and Blackstone recently completed a $35 billion “Big Sky” transaction to support Anthropic. This is also the first transaction on the “AI XPV” platform jointly established by Broadcom, Apollo and Blackstone. According to estimates by strategists at Bank of America, if this platform achieves a financing capacity of 20 gigawatts, it may require $235 billion in endorsed debt by mid-2029.

In this Anthropic transaction, the two priority bonds for which Broadcom provided backing received an investment-grade rating, with the highest interest rate being 5.75%; the riskier $4.5 billion subordinated tranche was priced at 8.5% – which is still significantly lower than the independent financing costs of Anthropic, a unrated private company.

In August this year, NVIDIA brought together Apollo, Blackstone, Bridgewater, Brookfield, KKR, and Goldman Sachs to jointly provide financing for the infrastructure of AI. NVIDIA itself committed to covering up to 25% of that share.

From private placement to syndicated lending market

To achieve true scalability, chip financing must break through the boundaries of the private equity market and enter the syndicated loan and even investment-grade public bond markets – the latter being one of the largest and lowest-cost capital pools in the world.

CoreWeave took the lead in completing syndicated leveraged loans supported by chips this year. Both transactions were steadily subscribed by institutional investors, and the trading prices in the market were close to par value, which stands in sharp contrast to the corporate-level bonds of the same company.

Morgan Stanley's Global Head of Structured Products and Private Capital Markets, Rich Myers, stated: "These financings are directly linked to the most valuable assets in the market, namely the computing power generated by chips. There is substantial evidence indicating that the demand for computing power will continue to grow and exceed the supply."

Apollo, the person in charge of global asset allocation, stated that the traditional public debt market alone cannot meet the financing needs of digital infrastructure. It requires the participation of various types of capital such as project financing, private placements, insurance funds, and commercial real estate. Obtaining an investment-grade rating is the key to unlocking the largest pool of capital, including insurance funds.

The accumulation of risks cannot be ignored.

However, as capital flows in at an accelerated pace, risks are also accumulating in tandem.

Fitch Ratings Senior Director Jason Pompeii pointed out that the current financing scale of manufacturers is unprecedented, and the risks are highly concentrated. "This only increases one possibility: once AI ultimately disappoints, the entire supply chain will be impacted."

The private nature of the transaction structure raises doubts about liquidity. Since most chip financing transactions do not circulate in the public market, it is difficult for investors to exit conveniently. Moreover, these transactions deeply bind a few chip manufacturers, cloud computing giants, and emerging cloud service providers together, resulting in a highly concentrated cross-exposure. Magnetar, the managing director replacing Dave Snyderman in credit and fixed-income business, warns: “The giants will only become bigger. Only they have access to financing channels.”

The legal validity of customer contracts is also in question. Carlyle, the person in charge of asset-backed financing, Akhil Bansal, stated bluntly: 'Everyone assumes that these contracts are unbreakable, but what I'm concerned about is the practical aspect – if only seven institutions provide all the credit support for GPU, then whose hands do the chips really lie in?'

Nuveen, the person in charge of private fixed-income investments, stated that she is currently putting off participating in large-scale investment-grade chip transactions and will re-evaluate the situation once the market structure becomes more mature. 'This is a completely new structure, and we are not yet clear about how the outcome will evolve in the event of any kind of shock.'

In addition, there are variables at the macro level as well: local opposition forces in Texas have led to a suspension order for new projects, Oracle's projects in New Mexico have been delayed due to regulatory obstacles, and OpenAI and Anthropic are also discussing slowing down the deployment pace due to security concerns.

Moving towards the "computing power asset category"

Despite the challenges, market participants remain optimistic about the long-term potential of this paradigm.

John Watson, the Senior Managing Director of Blackstone's Tactical Opportunities Team, compares the current moment to the early days of the automotive industry: just as those who provided financing for Ford's Model T faced the question "will people really learn how to drive?", today's automotive finance has developed into a massive and mature market.

In terms of the evolution of the chip market, early research and actual usage data indicate that the actual lifespan of chips is exceeding the years reflected in current accounting practices, which will improve their financing viability. Karen Fang, the global head of infrastructure and sustainable financing at Bank of America, expects that semiconductor valuation tools similar to used car price guides will gradually emerge, thereby fostering a more active chip leasing and hedging market.

The ultimate vision depicted by Watson is to transform a unit of computing power into an asset class that can be traded and financed on a global scale – “We understand the economic value of computing power, and therefore finance it.”

To achieve this goal, it still depends on the AI commercialization process truly meeting market expectations.