Nvidia’s announcement of a financing partnership with Goldman Sachs and other financial institutions, providing up to US$500 billion (HK$3.9 billion) in funding for companies investing in AI infrastructure, may appear to remove one of the biggest constraints on AI investment: access to capital. However, beneath the surface, the market is already beginning to see signs of rising leverage and financing risks among technology companies.
For Nvidia and other AI semiconductor companies, Goldman Sachs’ guaranteed financing structure is clearly a major positive. It allows them to increase sales without taking on additional receivables risk, while the lending exposure is effectively transferred to the financial institutions and the debt burden ultimately falls on Nvidia’s customers.
Meta was among the earliest companies to adopt this financing model for data-center expansion. Its latest financial disclosures also referred to residual value guarantees, stating: “RVG payments are not probable, and therefore no liability has been recorded to date.” However, from an investor’s perspective, “not probable” is hardly a reassuring answer when significant financing commitments are involved.
Broadcom has adopted a similar approach, working with financial institutions to provide chip financing for customers and reportedly extending guarantees to Anthropic. Under the structure, Anthropic purchases chips using financing secured against contracted customer agreements and the associated future cash flows. If customers terminate their contracts or stop using the products, the underlying assets can theoretically be leased or sold to repay the remaining debt.
Should a funding shortfall remain, the guarantor, Broadcom, would be responsible for covering the difference.
The market currently assumes that AI chip demand will remain structurally stronger than supply for the next several years, making these financing risks appear manageable. Yet investors may be overlooking several critical questions. First, will the current enthusiasm for AI remain intact? Second, as companies continue to invest heavily in AI and experience declining cash flow, could they eventually reduce capital expenditure, causing the supply-demand imbalance to reverse? Third, could the useful economic life of AI chips be significantly shorter than the duration of the underlying debt?
This is particularly important because new generations of AI chips are being introduced roughly every two years. If older chips lose value rapidly, their residual value may prove insufficient to support the loans secured against them, potentially creating losses for lenders and guarantors.
Meanwhile, as semiconductor financing becomes increasingly widespread, private credit funds are also facing growing exposure to leverage and losses. SaaS companies, once regarded as some of the most attractive investments, have recently encountered earnings and financial pressures, resulting in significant losses for parts of the private credit market.
Therefore, while the financing arrangements being developed by Nvidia, Broadcom and financial institutions may not necessarily lead to a crisis on the scale of the 2007 subprime mortgage collapse, investors should not dismiss the risks. The greater the amount of debt used to finance the AI boom, the greater the potential for credit risk to spread beyond the technology sector should AI demand, chip valuations or corporate cash flows deteriorate.
Andrew Wong is a veteran independent commentator
