Wall Street is making a bet that defies conventional financial logic: AI chips won't depreciate as quickly as traditional tech hardware. Nvidia has announced a massive $500 billion partnership where technology companies can lease its AI chips, with financing provided by institutions like Apollo, KKR, Brookfield, BlackRock, and Goldman Sachs.
Financial executives involved in the deal have shared that the core assumption behind this massive collaboration is that, given the sustained high demand for AI computing power, chip prices will remain elevated for longer than many analysts expect. Nvidia CEO Jensen Huang has stated that this arrangement will create a new asset class backed by semiconductors, potentially attracting the vast $22 trillion pool of private capital. Major figures like Blackstone's Jon Gray and BlackRock's Larry Fink have also publicly expressed their willingness to invest, aiming to capture the immense demand for data center construction.
The Core Logic and Risks of the New Model
In traditional financing logic, tech hardware depreciates rapidly. While cars or commercial aircraft have a stable secondary market in the event of a customer default, the long-term residual value of chips is highly uncertain due to rapid technological iteration and fluctuating demand. Currently, most lenders require loans backed by chip leases to be fully repaid within three to five years, assuming the residual value is near zero after that period. At the same time, companies leasing these chips must also build data centers, causing actual revenue to lag far behind the enormous upfront expenditures.
Analyst Ben Bajarin has pointed out that the entire model is built on the premise of "continuous investment." He notes there is a risk of "overbuilding, a slowdown in demand, or a reduction in computing power needs as models improve."
Nvidia's Hedge and Guarantees
Jensen Huang offers a different perspective. He notes that even older H100 chips have retained their value longer than expected due to the explosive demand for computing power, and the six-year-old A100 chip is still in use. Nvidia also extends the practical lifespan of its chips by continuously updating its software platform, CUDA. The latest and most powerful GPUs are primarily used for training cutting-edge models, while older hardware can be used for simpler tasks like answering queries.
More importantly, Nvidia has committed that the chips will retain at least 25% of their initial value during the lease term. This means the $5.3 trillion company will be the first to absorb any depreciation losses that exceed expectations. An executive involved in the deal referred to this as a "first-loss buffer," giving private capital more confidence to package the related financing into securities and sell them to long-term investors like insurance companies.
Private Capital's New Playbook
Some institutions plan to structure GPU leases into different risk tiers, similar to collateralized loan obligations (CLOs), allowing investors to choose based on their risk appetite. Others believe that different parties will adopt different structures, preventing a unified model from emerging. For Nvidia, this arrangement has an additional benefit: shifting the financial burden of its customers off its own balance sheet. In the past, Nvidia often provided direct guarantees for its customers through supplier financing, but now the world's largest financial institutions are taking on that burden, significantly reducing its own exposure.
Wall Street is betting with real money that in the AI era, Nvidia's chips will break the old rule of "rapid technological depreciation." The success of this $500 billion wager ultimately depends on whether the demand for computing power can truly remain high and whether the residual value of the chips can truly hold up.
Comments