NextFin News - Broadcom is seeking more than $60 billion in a new debt financing tied to artificial-intelligence infrastructure, the largest single round yet in a capital machine that now runs alongside the chipmaker's core semiconductor business. The deal, pitched to lenders this week, extends the AI XPV Platform the company launched in June with Apollo Global Management and Blackstone - a structure that lets frontier AI labs lease compute without putting the debt on their own balance sheets, while Broadcom stands behind the lease payments with guarantees.
The financing push lands as Broadcom's shares have already surrendered a meaningful chunk of their AI-era gains, as Bank of America's credit team downgraded the company's bonds over the very guarantees that make these deals possible, and as a modeled exposure ceiling of roughly $370 billion by mid-2029 began circulating among fixed-income investors. The central question is no longer whether Broadcom can sell AI chips. It is whether the chipmaker is quietly becoming the banker for the customers who buy them - and what that does to a credit profile that, on the face of the balance sheet, still looks conservative.
The Deal and the Machine Behind It
The new round is being assembled on top of the AI XPV Platform, announced June 9, 2026, by Broadcom, Apollo, and Blackstone's Credit & Insurance business. That platform was designed to enable more than 20 gigawatts of compute capacity through 2028 using Broadcom's custom XPUs and networking gear, with Anthropic and OpenAI named as anchor customers. It opened with a $35 billion tranche led by Apollo to fund more than 1 gigawatt of Anthropic's expansion, deploying at Fluidstack-operated sites from mid-2026.
The mechanics are borrowed from asset-backed finance rather than corporate lending. A special-purpose vehicle raises the debt, buys the AI racks - in the first deal, Google's custom tensor processing units co-developed with Broadcom - and leases them to the AI lab. The debt is sliced into tranches: roughly $6 billion of A1 notes, $25 billion of A2 notes, and $4.5 billion of B notes in the initial structure. Broadcom then provides a residual-value guarantee on the senior tranches, meaning that if the customer misses lease payments and the resale value of the chips does not cover what senior lenders are owed, Broadcom makes up the shortfall.
The structure keeps the borrowing off both balance sheets that matter most. Anthropic does not record the debt as its own. Broadcom does not book the full financing as a liability either; instead, its Form 10-Q filed June 9, 2026, caps the maximum loss on the platform's first transaction at $29 billion - a contingent exposure that grows as racks are deployed and shrinks as the customer pays down the lease. Total debt on the balance sheet stood at $64.9 billion against cash and cash equivalents of $19.6 billion, leaving net debt of about $45.3 billion, or well under one times annualized adjusted EBITDA at the current earnings pace.
"We are at a historic inflection point where the demand for AI compute is fundamentally reshaping the global economic landscape," Hock Tan, Broadcom's president and chief executive, said when the platform launched. "This strategic Platform with Apollo and Blackstone synchronizes the world's most sophisticated capital with Broadcom's advanced technological roadmap."
For Broadcom, the commercial logic is immediate: it can book large chip orders from AI labs that would otherwise need hundreds of billions of dollars of upfront capital they do not have. The chipmaker is, in effect, helping finance the demand for its own products - a virtuous circle as long as the AI build-out stays funded, and the source of the current credit anxiety when it does not.
What the $370 Billion Number Actually Means
The figure that moved markets is not the more-than-$60 billion being sought now, nor the $35 billion that launched the platform. It is a modeled ceiling produced by Bank of America's credit analysts: if the AI XPV Platform scales to its full 20-gigawatt design, the cumulative financing stacked across those deals could reach approximately $370 billion by mid-2029, including roughly $150 billion of new issuance in 2027 alone. Much of that would carry Broadcom's guarantees.
That number is a ceiling on hypothetical future commitments, not debt on Broadcom's balance sheet today. Even Bank of America's own modeled worst case - every customer defaulting at once - puts Broadcom's losses at about $42 billion, with roughly $10.5 billion at a 25% default rate, because the analysis assumes Broadcom recovers some value on the racks. For perspective on the gap between the two numbers: Broadcom earned $9.3 billion in net income in its fiscal second quarter, ended May 3, 2026, on revenue of $22.2 billion, up 48% year over year. A total wipeout on the first transaction would equal roughly nine months of profit at the current pace - painful, but an order of magnitude removed from $370 billion.
The distinction matters because it frames the entire debate. Bulls see a capital-light way to lock in multi-year chip demand. Bears see a contingent liability that compounds with every rack deployed and that credit markets are only now beginning to price. The truth sits in the accounting: these guarantees are not booked as liabilities because the company judges a draw to be improbable, the same treatment Meta Platforms applied to its own data-center residual-value guarantees in its filings.
The Second-Order Risk: Underwriting Your Own Demand
The first-order effect of these deals is obvious: Broadcom sells more chips. The second-order effect is what bond investors are starting to price. By backstopping the leases, Broadcom is underwriting the creditworthiness of private AI labs whose ability to pay depends on the AI build-out staying funded - the same build-out that determines whether Broadcom's own revenue guidance holds.
This creates a form of circular exposure that does not appear in the headline debt figure. If AI capital spending slows, three things happen at once: chip orders fall, the resale value of deployed racks falls, and the customers' lease payments come under pressure - precisely when the guarantee is most likely to be drawn. The guarantee is designed as counter-cyclical protection for lenders; in a downturn it becomes pro-cyclical risk for Broadcom. That is the transmission channel that a static reading of the $29 billion cap misses.
The mechanism also has a pricing consequence. The guarantees effectively convert the senior tranches of these deals to Broadcom's own investment-grade credit quality rather than the AI lab's. That is the feature that makes the debt cheap and the deals bankable - and the reason Bank of America downgraded Broadcom's bonds to market weight. The bank is treating a growing stack of off-balance-sheet commitments as a real credit consideration, even though accounting rules do not require them to be booked as liabilities yet. In credit markets, a guarantee that is not on the balance sheet can still be in the price of the bond.
Broadcom is not alone in this. Nvidia has said it may provide residual-value support for up to 25% of an opportunity, assessed on a case-by-case basis. Meta Platforms pioneered the structure for its own data centers and described it plainly in filings: residual-value guarantee payments were "not probable, and therefore no liability has been recorded to date." The model has become a standard tool for containing AI infrastructure costs without loading balance sheets - which is exactly why regulators and credit analysts are beginning to look through it. The comparison that lenders are drawing is less to traditional corporate debt and more to real-estate and aircraft financing, where the asset is the collateral and the sponsor's credit is the backstop.
The Counter-Thesis: Why the Bond Market Is Not Overreacting
The strongest case against Broadcom's structure is not that $370 billion is owed today. It is that the platform's economics depend on a single assumption: that AI compute retains residual value and that frontier labs keep paying through the cycle. History offers few analogs for a financing vehicle backed by hardware that depreciates as fast as AI accelerators and whose end-users are private companies burning cash at unprecedented rates. Aircraft and property have decades-long useful lives and observable resale markets; a generation of AI chips can be rendered economically obsolete by the next one.
There is also customer concentration. Anthropic and OpenAI are the named anchor customers, and Anthropic's lease payments depend on its run-rate revenue continuing to climb - the company said in April that run-rate revenue had surpassed $30 billion, up from about $9 billion at the end of 2025. That is rapid growth, but it is also the kind of trajectory that leaves little room for a funding winter. Anthropic has filed confidentially for an initial public offering, with advisers including Morgan Stanley, Goldman Sachs, and JPMorgan Chase working on a process that could reach public markets as early as October. A successful IPO would strengthen the credit story; a delayed or downsized offering would do the opposite, and would be the first observable crack in the financing thesis.
The falsifying signal for the bull case is specific and observable: if Broadcom's guarantee exposure grows faster than its free cash flow for two consecutive quarters, or if the company discloses any draw on the backstop, the argument that these are harmless off-balance-sheet commitments breaks down. Free cash flow was $10.3 billion, or 46% of revenue, in the fiscal second quarter. That is the buffer. The ratio to watch is committed guarantee cap divided by trailing free cash flow - if it rises rather than falls as the platform scales, the credit overhang is real rather than theoretical.
Market Reaction and the Competitive Overlay
Broadcom's shares fell about 5.4% on Wednesday to trade near $360, a fourth straight daily decline, before recovering part of the loss on Thursday, when the stock moved in a range of $357.61 to $381.16 and closed around $365.80 on volume of 35.09 million shares, well above the roughly 19 million average. The selloff compounded pressure from two separate developments: a reported security vulnerability in VMware, and a custom-chip agreement between Marvell Technology and Alphabet that stoked fears Broadcom could be supplanted in some customer designs. The shares sit roughly 26% below their 52-week high of $495, reached in early June, though analyst ratings remain predominantly positive.
The competitive angle matters because it intersects with the financing angle. Broadcom's ability to place its XPUs into the AI XPV Platform depends on winning the underlying chip designs. If hyperscalers shift more workloads to in-house silicon or to rival suppliers, the platform's collateral - racks built on Broadcom's chips - becomes harder to redeploy, and the residual-value guarantees become more likely to be tested. The financing model and the product cycle are not independent risks; they are two views of the same dependency on Broadcom's technology leadership holding.
What Comes Next: Catalysts, Scenarios, and What to Watch
The medium-term catalyst is the fiscal third-quarter earnings report, scheduled for September 2, 2026, after the market close. Consensus expects earnings per share of about $3.16, following a second-quarter beat where non-GAAP diluted EPS of $2.44 topped the $2.32 estimate. Management has guided third-quarter consolidated revenue to about $29.4 billion, an 84% year-over-year increase, with AI semiconductor revenue expected to exceed $16 billion and non-GAAP operating margin stable at 67%. The numbers themselves are likely to be strong; what will move the stock is management's commentary on the platform's rollout, the pace of new guarantee commitments, and any update on the more-than-$60 billion round.
Three scenarios frame the path ahead. In the base case, the new deal closes, the platform scales toward its 20-gigawatt design, and Broadcom's guarantees remain undrawn - the credit overhang fades as cash flow compounds and net leverage stays well below one times EBITDA. In the upside case, Anthropic's IPO prices successfully, AI capital expenditure accelerates, and the financing model is validated as the industry standard, lifting Broadcom's valuation back toward its early-June highs. In the downside case, a draw on the backstop is disclosed, or AI funding tightens and guarantee commitments outpace free cash flow - at which point the market re-rates Broadcom not as a chip designer but as a leveraged financier of AI infrastructure, and the multiple contracts toward levels that price credit risk as well as growth.
Short-term, the stock will track the flow of headlines around the financing round and the VMware issue. Medium-term, it will track whether free cash flow keeps pace with the growing guarantee book. Long-term, this is a structural shift in how AI infrastructure is financed, not a cyclical credit event: the residual-value guarantee model is becoming the industry's standard plumbing, which means the exposure will not revert on its own - it compounds with each deal. The question is whether the AI build-out is being financed on the credit of the companies selling the picks and shovels because the miners cannot fund it themselves - and whether that is a feature of a healthy boom or an early warning that the boom is running on credit rather than cash.
Broadcom's next earnings call will tell investors which path the platform is on. Until then, the more-than-$60 billion deal is less a financing event than a stress test of a new question: whether the AI boom can be underwritten on the credit of the companies selling the picks and shovels, rather than the miners digging with them.
Market data as of the August 20, 2026 close. This article is for informational purposes and does not constitute investment advice.
Explore more exclusive insights at nextfin.ai.
