NextFin News - Broadcom Inc.'s cost of insuring against default has jumped to a record as the chipmaker negotiates one of the largest off-balance-sheet financing structures in technology history: a special-purpose-vehicle debt package that could reach roughly $100 billion to fund artificial-intelligence chip deployments for customers including Anthropic PBC. The company's five-year credit-default-swap spread climbed to 122 basis points, the widest on record, after reports that Broadcom is arranging more than $60 billion of senior secured debt plus a junior tranche of about $30 billion through SPVs it partially guarantees.
The move marks a sharp escalation of a strategy Broadcom began in June, when it partnered with Apollo Global Management Inc. and Blackstone Inc. on an initial $35 billion financing vehicle aimed at expanding Anthropic's computing capacity. Credit markets are now repricing what that strategy implies: a chip supplier increasingly acting as a leveraged financier for the AI buildout, with tens of billions in contingent obligations that sit outside its balance sheet but could land on it if the AI spending cycle turns.
The central question is whether the market is overreacting to a financing innovation or correctly identifying a business-model shift. The answer matters because Broadcom is not an outlier. The same off-balance-sheet architecture is spreading across the AI supply chain, and the bond market's verdict on Broadcom will set the template for how investors price every chipmaker that starts underwriting its customers' debt.
The Deal Architecture: How $100 Billion Stays Off the Balance Sheet
The proposed financing follows the template of Broadcom's June arrangement. A special-purpose vehicle raises debt, uses the proceeds to purchase custom AI accelerators and networking gear designed by Broadcom, and leases the equipment to a large AI customer. Lease payments service the debt, and the chips' residual value provides collateral. If the customer stops paying, the SPV re-leases or sells the hardware; only if a shortfall remains does Broadcom's backstop kick in.
The structure is tranched to separate risk appetites. In the June transaction, the $35 billion of debt was split into a $6 billion senior A1 tranche and a $24 billion A2 tranche, both backed by Broadcom's credit endorsement and residual-value guarantees, with Apollo's Atlas SP Partners contributing roughly $800 million of equity. The expanded package under discussion follows the same logic at a much larger scale: lenders are targeting a senior-secured tranche of $60 billion to $70 billion, with Broadcom providing credit endorsement and residual-value guarantees, plus a junior tranche of roughly $30 billion held inside the SPVs with less direct backing. The structure is designed to grow with the platform: analysts at Bank of America have modeled that if Broadcom's AI XPV platform reaches its stated target of more than 20 gigawatts of compute by 2029, maximum residual-value-guarantee exposure could climb to approximately $370 billion by mid-decade's end.
The accounting treatment is the crux of the debate. In its most recent quarterly filing, Broadcom disclosed that its maximum loss on the platform's first transaction is capped at $29 billion. The company describes the backstop as growing as racks are deployed and shrinking as the customer pays down five-year leases; if a customer defaults, Broadcom can assume the lease or sell the assets. That framing lets management argue the exposure is bounded and amortizing. Fixed-income investors hear something different: a $29 billion cap on a single transaction, with the platform still in its first inning.
The distinction between "maximum exposure" and "maximum loss" is where the two sides of the debate actually live. Maximum exposure is the notional ceiling — the full $370 billion of senior debt that could be outstanding by 2029. Maximum loss is what Broadcom would actually pay after repossessing and redeploying or selling the hardware. Bank of America's modeling puts the worst case at about $42 billion if every customer defaults at once, and about $10.5 billion under a 25% default scenario. Bulls see the gap between $370 billion and $42 billion as proof the panic is overdone. Bears see the $42 billion as a loss that would arrive in the same quarter that AI revenue disappoints — a correlated hit, not a diversified one.
Why Bond Investors Are Repricing the Risk
The 122-basis-point CDS reading is not an isolated twitch. Bank of America analysts note Broadcom's bond spreads have widened 20 to 45 basis points relative to industry peers, and the bank downgraded Broadcom's debt rating, citing the guarantees tied to the AI financing platform. Rating agencies have also begun to adjust their posture. Moody's Ratings warned that a substantial increase in Broadcom's contingent obligations could limit the company's financial flexibility and create an overhang on its credit profile, even while its reported leverage remains low. S&P Global Ratings said it treats Broadcom's residual-value support as a "contingent debt-like obligation" that it adds to adjusted debt when assessing the company.
The repricing reflects a recognition that the risk is not Broadcom's existing balance sheet. Total debt to EBITDA is expected to fall below 1.5x by fiscal 2026, down from about 3x at fiscal 2024, assuming modest paydown. The concern is the contingent balance sheet: obligations that are invisible in headline leverage ratios but become real precisely when the industry weakens. That is the pro-cyclical nature of residual-value guarantees. They are cheap to carry in a boom, when chip demand outstrips supply and no lessee defaults. They are expensive in a bust, when used-chip prices fall and customers renegotiate or fail at the same time.
Broadcom's own numbers show how much depends on the cycle holding. The company reported record fiscal second-quarter revenue of $22.2 billion, up 48% year on year, with AI semiconductor revenue of $10.8 billion, up 143%. For full-year fiscal 2026, management expects AI semiconductor revenue of $56 billion, up approximately 180% from fiscal 2025, and has reiterated guidance for fiscal 2027 AI semiconductor revenue in excess of $100 billion. Chief Executive Hock Tan has described the 2027 figure as "significantly in excess of $100 billion." That projection exceeds Broadcom's entire prior annual revenue and represents a bet that enterprise AI adoption will not merely continue but accelerate.
There is a second dependency embedded in those numbers: the VMware integration. Moody's Ratings, in upgrading Broadcom's senior unsecured rating to A3, noted the company has made significant progress converting VMware's largest enterprise customers from perpetual software licenses to recurring subscription services, with infrastructure-software operating margins on track to rise about 12 percentage points year over year in fiscal 2026. If the subscription conversion slows, the company loses the cash-flow cushion that makes the guarantees feel safe. Credit investors are therefore underwriting two cycles at once: the AI capex cycle and the enterprise-software transition cycle. Both have to work.
The Second-Order Risk: When a Chip Supplier Becomes a Bank
The first-order story is straightforward: Broadcom is using financing to remove a capital bottleneck for its customers, which should boost chip sales. The second-order problem is that the company is importing a credit cycle into a business that has historically been insulated from one.
Semiconductor companies earn revenue when chips ship. Their earnings fluctuate with order cycles, but they do not typically carry five-year lease-performance risk on equipment they no longer control. Broadcom's structure changes that. The guarantees mean the company's economics now depend not only on whether customers order chips today, but on whether those customers remain solvent and whether the leased hardware retains value years from now. In effect, Broadcom is selling chips twice: once for cash at delivery, and again through a long-dated promise to stand behind the financing that made the delivery possible.
The mechanism matters because it concentrates risk at the worst possible moment. If AI demand cools, three things happen simultaneously: chip orders slow, reducing revenue; used-chip prices fall, increasing the shortfall on any re-lease or sale; and customer stress rises, raising the probability that backstops are triggered. The guarantees do not diversify Broadcom's exposure; they multiply it along correlated dimensions.
This is the same structural flaw that made off-balance-sheet vehicles dangerous in previous credit cycles. The lesson of past securitization episodes is not that guarantees are inherently bad; it is that they migrate risk to the party least able to see it building. When obligations live in footnotes rather than on the face of the balance sheet, they escape the discipline of leverage ratios and interest-coverage tests until the moment they can no longer be ignored. Broadcom's disclosure discipline is better than that — the $29 billion cap is explicit — but the market is discounting the structure anyway, which is itself a signal.
The equity market has begun to price the same concern. Broadcom's shares are trading roughly 22% below the 52-week high set in June 2026, near the $368 to $373 range, as the debt-downgrade report and broader unease about AI funding structures weighed on the stock. The decline has been company-specific rather than sector-wide, pointing to a reassessment of Broadcom's particular risk profile rather than a broad technology selloff. That is a more damaging signal than a beta-driven drop: it says investors are re-underwriting the business, not just trimming exposure.
The Counter-Thesis: Why the Panic May Be Premature
The strongest argument against the credit bears is that the structure is deliberately designed to amortize. The debt is secured by long-term customer contracts, the backstop shrinks as lease payments are made, and Broadcom retains the option to repossess and redeploy the hardware. In a market where compute demand is expected to outstrip supply for years, the probability of a lessee walking away from working AI infrastructure is low. The company's disclosure that its maximum loss on the first transaction is capped at $29 billion is not a forecast of loss; it is a ceiling that would require a customer to stop paying entirely while the racks became nearly worthless.
There is also precedent for the arrangement working cleanly. Meta Platforms Inc., which pioneered residual-value guarantees for its own data centers, addressed the accounting directly in its regulatory filings:
RVG payments are not probable, and therefore no liability has been recorded to date.
Broadcom's supporters argue the same logic applies: these are tail-risk backstops, not expected losses, and the financing vehicle simply shifts capital formation to investors willing to hold the debt.
That argument is coherent — but it rests on a single assumption: that the AI capex cycle does not turn before the leases amortize. The falsifying signal is specific. If Broadcom's AI semiconductor revenue fails to approach the $100 billion run-rate it has guided for fiscal 2027, or if the company's CDS spread remains above 120 basis points while used AI-accelerator resale prices decline materially, the "remote tail risk" thesis is wrong. A sustained CDS level near current records would indicate that protection buyers are pricing more than a tail; they are pricing a structural reassessment of the business model.
Cyclical Wave or Structural Shift: The Call
The verdict requires separating two forces that are operating at different speeds. The near-term repricing is cyclical: CDS spreads mean-revert, financing terms get renegotiated, and the market's appetite for AI credit risk will ebb and flow with the earnings cycle. If Broadcom hits its $100 billion AI revenue guide and no lessee defaults, today's 122-basis-point spread will look like a transient overshoot.
The deeper shift, however, is structural and it will not revert on its own. Broadcom has crossed from a pure semiconductor supplier into a business that underwrites long-duration credit risk on assets it no longer controls. That change is durable: once investors learn to price the contingent balance sheet, they will not unlearn it. The company can shrink the exposure over time, but it cannot return to the pre-guarantee valuation multiple, because the earnings stream is now genuinely different. The cyclical leg says the panic is overdone; the structural leg says the multiple compression is permanent. The correct read is both: spreads will tighten if execution holds, but the discount to a pure-play chip peer is here to stay.
What Comes Next
In the short term, the focus shifts to Broadcom's September 2 earnings release, where investors will scrutinize any updated disclosure on the financing platform's exposure and any change in the company's maximum-loss cap. The consensus expects earnings per share of $3.22 and revenue of $29.47 billion for the quarter; the credit conversation will center less on the print than on management's commentary about the guarantees. A widening of the maximum-loss disclosure, or any indication that the $35 billion June vehicle has required support, would be the clearest near-term negative.
Over the medium term, the key variable is execution on the 20-gigawatt platform. If Broadcom can deploy the capacity and the leases perform, the current CDS widening will look like an overreaction to a financing innovation. If deployment slows or a lessee restructures, the contingent obligations move from footnotes to the income statement. The base case is that the platform ramps and the guarantees remain dormant; the downside case is that a single large lessee renegotiates, forcing Broadcom to recognize what a "contingent" obligation costs in practice.
Longer term, the episode is a test of whether the market will tolerate chip suppliers becoming quasi-lenders. Broadcom is not alone: Nvidia Corp. has announced a $500 billion financing partnership with six large asset managers and said it may provide residual-value support for up to 25% of an opportunity, assessed case by case. It has also committed up to $105 billion to finance an OpenAI data center in Ohio. Across Big Tech, corporate bond issuance has reached roughly $200 billion this year, nearly double the full-year 2025 total, and technology-linked CDS trading reached almost $650 million per day during the second quarter, up 20% sequentially and nearly 600% from a year earlier, according to DTCC data. The repricing is broad: Oracle Corp.'s five-year CDS has surged to around 200 basis points after touching a high of 219 in late July; Meta Platforms' is up 39 basis points to about 93; Nvidia's has gained 32 basis points year to date to roughly 82.
The verdict on Broadcom's credit risk will not come from a single earnings call. It will come from whether the AI spending wave lasts long enough for the leases to amortize — and whether a chip company can underwrite a credit cycle without becoming a casualty of it. For now, the bond market has rendered its preliminary judgment: Broadcom's guarantees are no longer free, and the bill for underwriting the AI boom has come due in the credit-default-swap market.
Explore more exclusive insights at nextfin.ai.
