NVIDIA, Broadcom and Other Chip Giants Guarantee About $300 Billion for AI Infrastructure Purchases
The Financial Times reports that large tech companies are increasingly providing off-balance-sheet financing support for AI expenditures through mechanisms such as "residual value guarantees." Chip giants like NVIDIA and Broadcom are using this to provide implicit credit backing for their customers' AI infrastructure purchases, with the overall off-balance-sheet guarantee scale in the industry reaching about $300 billion.
The operation of this structure involves special purpose entities (SPVs) issuing bonds to finance data centers and chip purchases, with chip manufacturers or cloud vendors promising to cover any shortfall if equipment depreciation exceeds expectations. These guarantees are typically only included in financial statement notes and not counted as liabilities on the regular balance sheet, allowing related companies to avoid recognizing all contingent liabilities on their balance sheets.
Specific cases include NVIDIA providing about $105 billion in guarantees to SoftBank subsidiary SB Energy for a data center project in Ohio serving OpenAI, which does not need to be included in the balance sheet liabilities until the OpenAI lease officially takes effect in 2028; Broadcom has taken on about $29 billion in exposure for a special purpose entity jointly built with Google for Anthropic, aimed at delivering 1 GW of computing power chips; Meta has also provided about $28 billion in residual value guarantees for its joint venture with Blue Owl for a 2 GW "Hyperion" data center project in Louisiana.
Morgan Stanley estimates that the total off-balance-sheet commitments and credit support from major cloud and chip manufacturers have exceeded $3.1 trillion, far surpassing the aforementioned $300 billion in direct guarantees. Rating agencies have begun to include the amounts of these guarantees in their adjusted debt calculations when assessing the credit risk of related companies.
For example, Broadcom's custom chip (XPU) shipments for fiscal years 2027 to 2028 are about 71% dependent on orders from OpenAI and Anthropic, creating a concentration risk that compounds with off-balance-sheet guarantee risks. Despite Broadcom's AI-related revenue soaring 221% year-on-year to $16.7 billion in the third quarter, its stock price has fallen about 28% from its 52-week high, reflecting investor concerns about the sustainability of such financing structures.
In market mechanisms, these guarantee arrangements essentially represent chip manufacturers using their own credit to "infuse blood" into customers' capital expenditures. If AI computing power demand falls short of expectations and data center asset depreciation exceeds the guarantor's predictions, the risk will be transmitted back to the guarantor's actual balance sheet, creating concerns similar to the accumulation of risks from off-balance-sheet special purpose entities before the 2008 financial crisis. Bond investors have begun to demand higher risk premiums for such "shadow credit support."
ABAB AI Insight
NVIDIA and Broadcom have repeatedly bound their major clients through circular investments and supplier financing in recent years. Previously, NVIDIA's commitment of up to $100 billion in circular investment arrangements to OpenAI raised market doubts about "self-reinforcing prosperity." This residual value guarantee continues the chip giants' consistent strategy of backing downstream customers' capital expenditures with their own balance sheet credit, essentially tying sales growth to customers' financing capabilities.
The funding chain presents a multi-layered nested structure of "chip manufacturer credit guarantees → special purpose entity bond financing → data center construction → leasing to cloud vendors/AI labs." On the surface, chip manufacturers only bear limited guarantee responsibilities, but if end AI demand falls short of expectations, the risk will reverse along this chain and ultimately still be borne by the guarantor. This arrangement objectively amplifies the leverage level of the entire AI infrastructure construction cycle while avoiding traditional liability disclosure requirements due to accounting treatment.
This structure is highly similar to the path taken by investment banks before the 2008 financial crisis to hide mortgage-related risks through off-balance-sheet special purpose entities (SPVs)—at that time, they also reduced on-paper leverage ratios through legal entity segmentation but failed to genuinely transfer underlying risks. Currently, AI infrastructure construction is in an accelerated expansion cycle characterized by "capital expenditure scales far exceeding operating cash flow and increasingly complex financing structures," reminiscent of the phase when telecom operators massively borrowed to lay fiber networks in the late internet bubble.
Essentially, this represents capital concentration—few manufacturers with core AI chip supply capabilities are using off-balance-sheet guarantees to internalize part of their customers' capital expenditure risks in exchange for order certainty and customer stickiness. This means that the real leverage and risk exposure of AI infrastructure construction are highly concentrated among a few chip manufacturers and large-scale cloud vendors. If AI application monetization falls short of expectations, the risk transmission path will be highly concentrated rather than dispersed, and the intensity of systemic shocks may far exceed the levels suggested by surface disclosure data.
ABAB News · Cognitive Law
Off-balance-sheet liabilities still need to be repaid on the day of liquidation.
When prosperity is built on guarantees, credit becomes the new leverage.
Risk has not disappeared; it has merely been hidden under a less conspicuous item.