As the capital required to build the next generation of AI infrastructure surpasses the limitations of traditional Wall Street financing, Nvidia ($NVDA) is quietly assuming a new role: lender of last resort. According to a discussion on the "All-In" podcast, Nvidia CEO Jensen Huang is bypassing constrained credit markets by offering unprecedented "vendor financing" to its largest customers. According to renowned fund manager and founder of Atreides Management, Gavin Baker, this move transforms the company from simply a hardware supplier into the de facto central bank of the AI ​​economy.

Instead of paying upfront for GPUs, large-scale data centers and emerging cloud providers are borrowing, purchasing the systems, and then repaying lenders from lease/expiration revenues. Nvidia's role is to act as an intermediary between computing buyers and Wall Street capital, and also to provide a guarantee (residual value guarantee) of what price the GPUs can still be leased at 3-4 years later. This guarantee allows banks to view GPUs as a financeable, cash-flow-generating asset. This is closer to aircraft financing than a technology capital expenditure pipeline. “Nvidia is kind of becoming the central bank of AI, the Federal Reserve of AI,” Baker said. “It’s like a way of conceptualizing what they’re doing.”

Here, Nvidia sets the terms (now a guarantee, a share of the revenue above the base price) and private equity – Goldman, KKR, Blackstone and others – execute the commitment transactions on the ground, much like commercial banks communicate monetary policy. Venture capitalist and one of the hosts of the “All-In” program, David Sacks, said Nvidia “eliminates the financing constraint” in the infrastructure development process, thus providing a credit pipeline that allows computing demand to scale as quickly as the actual total market size (TAM), rather than being limited by the balance sheets of large-scale data centers.

Baker made it clear that this is not cyclical financing. He described it as asset-based lending against actual expected cash flows; not as vendor financing supporting its own customer. The comparison both he and Sacks make is to mortgage-backed securities: Nvidia standardizing reference designs so that GPU cash flows can be packaged and securitized like mortgages.

According to him, the scenario that would refute this thesis is this: If data centers are built assuming that the $30-50/watt spot prices will become obsolete, there will be a surplus of computing power – like the GPU version of "dark fiber" after the dot-com bubble burst. He argues that political/regulatory friction over data center construction (paradoxically) acts as a natural constraint on supply, mitigating this risk. Baker described Anthropic's (ANTHRO) upcoming S-1 filing as the industry's most important data point, arguing that macro/value investors who view AI as subsidized/cyclical are wrong about the fundamental principles of the token economy, and that this filing will prove this.