Concerns over AI chip financing are shifting from a "demand-boosting" narrative to one focused on "credit risk," placing significant selling pressure on Broadcom, a key rival to NVIDIA.
On Friday, August 14, shares of Broadcom (AVGO) opened lower and continued to decline, hitting an intraday low in the afternoon with a drop of approximately 7%, before closing down over 5.9%. Media reports indicate that investors are reassessing the credit risks associated with the rapid expansion of AI infrastructure financing, bringing Broadcom's recently launched AI XPV financing platform into the spotlight.
This sell-off is not triggered by a sudden deterioration in Broadcom's latest earnings, but rather reflects a market repricing of its AI growth model. The core concern is that Broadcom, through partnerships with institutions like Apollo and Blackstone, is combining expensive AI chips and computing infrastructure with private credit to accelerate customer purchases via financing. As the platform scales, chip residual values, customer debt-servicing ability, and Broadcom's potential guarantee obligations could all become new sources of credit risk.
Notably, while Broadcom faced pressure, NVIDIA is also pushing forward with a larger AI financing platform. Reports emerged this week that NVIDIA is collaborating with six Wall Street giants, including Apollo, Blackstone, BlackRock, and Goldman Sachs, aiming to mobilize over $500 billion in third-party capital through a compute financing platform.
Commentators suggest this model highlights the rapidly increasing dependence of AI infrastructure buildout on external financing, prompting the market to question: if AI computing demand requires increasingly more debt financing to be sustained, how much of this AI capital expenditure boom is driven by genuine cash flow versus financial leverage?
Broadcom's AI Financing Platform: From a $35 Billion Deal to 20GW of Compute
Broadcom's AI XPV Platform was officially launched in June, established in partnership with Apollo and Blackstone, with an initial capital solution sized at $350 billion, aiming to support over 20GW of AI compute capacity by 2028.
The first transaction primarily serves Anthropic's plan to expand compute capacity by over 1GW, utilizing Broadcom's custom XPU chips and networking solutions. Apollo stated at the time that this was one of the largest private financings it had participated in.
The key to this model is transforming the substantial capital expenditures typically borne directly by AI companies or cloud providers into infrastructure financing funded by the private credit market.
For Broadcom, the benefits are clear: customers can reduce upfront capital expenditure pressure, while Broadcom can leverage financial institutions to expand XPU deployments, thereby amplifying AI chip revenue.
However, the problems arise precisely from this structure.
If AI chips themselves become a critical collateral for financing transactions, their future residual value directly impacts the safety of the loans. Unlike mature assets like servers or aircraft, custom AI XPUs lack a developed secondary market, creating significant uncertainty about whether chips can be quickly sold to other buyers and at what price in the event of a customer default.
Bank of America analyst Tom Curcuruto noted that the XPV platform currently has high lessee concentration. The initial transaction relies heavily on Anthropic, and while OpenAI could become a future client, other lessees are not yet confirmed.
Bank of America Downgrades Broadcom: The Real Concern Isn't Now, But Platform Expansion
Market concerns about the XPV platform began to crystallize this week.
On Monday, Bank of America downgraded Broadcom's issuer and bond ratings from Overweight to Marketweight, specifically citing the credit risk uncertainty from the XPV platform's expansion.
BofA pointed out that since early June, the spread on Broadcom's bonds relative to similarly-rated semiconductor companies has widened by approximately 20 to 30 basis points. Currently, the spread on Broadcom's 4.95% bonds due 2036 is about 105 basis points, and the spread on its 5.7% bonds due 2056 is about 118 basis points, roughly 30 to 45 basis points higher than non-AI semiconductor companies like Texas Instruments or Qualcomm.
Crucially, Bank of America is not turning bearish on Broadcom's fundamentals. On the contrary, the same note raised its forecasts for Broadcom's fiscal 2026 revenue and EBITDA by 10% and 13%, respectively.
This means the market is currently trading not on "Broadcom's AI business slowing down," but on a different layer of risk: the stronger the AI business and the faster the financing platform expands, the larger the potential credit exposure may become.
BofA believes the risks of the XPV platform stem from two main sources: first, the lack of a mature secondary market for XPUs makes future residual values difficult to assess; second, there is high customer concentration. If investors begin to price some XPV risk into Broadcom's parent company credit risk, possibly by purchasing Broadcom CDS for hedging, it could further increase its financing costs.
What Does $370 Billion Mean? Guarantee Exposure in Extreme Scenarios
What truly alarmed the market was Bank of America's stress test for the XPV platform's future scale.
Broadcom does not provide all financing directly to customers. Instead, it supports related debt through arrangements like Residual Value Guarantees (RVG). For the initial $350 billion in XPU assets, Broadcom previously disclosed that under the extreme assumption of 100% defaults with zero collateral recovery value, the maximum loss exposure on the senior debt would be $290 billion.
In a stricter stress test, Bank of America assumed chip prices decline by 20% annually, with an additional 25% price shock upon default. The results showed a peak RVG exposure of approximately $260 billion for the initial transaction, but the corresponding maximum actual loss was around $2.9 billion.
What truly sparked concern was the expansion scenario.
If the XPV platform ultimately expands to 20GW by adding 2GW per quarter, BofA estimates that by mid-2029, Broadcom's maximum RVG exposure could reach $370 billion. In the extreme case of 100% defaults, this would correspond to a maximum loss of approximately $42 billion. If the default rate were 25%, the potential loss would be around $10.5 billion.
It is important to emphasize that $370 billion does not mean Broadcom has incurred $370 billion in debt, nor is it BofA's forecasted actual loss. It is the maximum guarantee exposure calculated under the extreme stress scenario of the XPV platform expanding to 20GW. BofA explicitly states that 100% defaults is an extreme and unrealistic assumption.
Furthermore, even under this extreme scenario, BofA estimates Broadcom's free cash flow after dividends could still reach approximately $85 billion in 2027, providing a substantial loss-absorption capacity. The bank believes the XPV platform alone is not sufficient to undermine Broadcom's credit foundation.
Therefore, the current market fear is not that "Broadcom will face a debt crisis immediately," but rather that as the XPV platform expands, some risks originally belonging to customers and financial institutions might gradually transmit to Broadcom itself through guarantees, credit spreads, and investor risk premiums.
NVIDIA's "$500 Billion Financing Plan" Further Exposes the Issue
Broadcom's stock decline on Friday also occurred against the backdrop of the entire AI financing model being re-evaluated by the market.
According to a report on Monday, NVIDIA is working with Wall Street institutions like Apollo, Blackstone, Goldman Sachs, and KKR to launch a "Compute Financing Platform," aiming to mobilize over $500 billion in third-party capital for AI infrastructure. NVIDIA CEO Jensen Huang stated the next day that the company could choose to support up to $125 billion of the transactions, equivalent to 25% of the potential deal size.
Media reports point out an increasingly apparent reality behind this financing model: many AI clients need vast amounts of computing power but may lack the balance sheets to independently bear the costs of data centers and chip purchases.
Consequently, Wall Street is becoming a crucial source of capital connecting AI chip suppliers with AI infrastructure demand.
From an industry logic perspective, this could further unlock AI chip demand. However, from a financial market perspective, it also means the AI boom is increasingly reliant on private credit, asset securitization, and guarantee structures.
Reuters, citing Bank of America data on Friday, reported that by mid-2029, Broadcom's chip financing program could generate approximately $370 billion in senior debt to fund 20GW of compute, with roughly $150 billion in new debt potentially added in 2027 alone.
This explains why the market has begun to scrutinize Broadcom's and NVIDIA's financing plans together: the AI infrastructure story is evolving from a "chip sales narrative" into a "chip + data center + private credit" financialization narrative.
The Next Test for AI CapEx: Who Bears the Risk?
Over the past two years, the core question for AI industry investors has been "Is demand strong enough?" Now, with individual AI infrastructure projects often involving hundreds of billions of dollars in investment, the market is asking another question: Who ultimately pays for these investments?
If hyperscale cloud providers like Microsoft, Google, Meta, and Amazon bear the cost directly, the risk manifests primarily as CapEx, depreciation, and free cash flow pressure. If AI startups bear it, the issue becomes one of business models and financing ability. But when an increasing number of projects are funded by the private credit market and supported by chip manufacturers through guarantee mechanisms, the risk begins to enter the credit market.
Broadcom's recent stock price decline is a microcosm of this shift.
In the short term, Broadcom's AI chip business remains on a high-growth trajectory, and Bank of America has not negated its operational fundamentals. However, the capital market is beginning to realize that if future AI computing demand requires ever-larger financing platforms to be sustained, the valuation logic for the entire AI industry chain must also account for asset residual values, customer default rates, debt costs, and supplier guarantee obligations.
For AI chip stocks already trading at high valuations, this narrative shift from "growth risk" to "credit risk" can often trigger significant stock price volatility.
Broadcom's sharp decline on Friday may reflect precisely this rapid repricing of risk premiums.
Comments