Nvidia this week unveiled a partnership with six major asset management firms—BlackRock, Blackstone, Apollo, KKR, Brookfield, and Goldman Sachs—to create a $500 billion financing pipeline for data centers and GPU clusters. The initiative aims to provide capital to companies lacking sufficient credit ratings or cash reserves to purchase high-end semiconductor equipment outright. Nvidia founder Jensen Huang framed the effort as treating AI infrastructure as a long-term investable asset comparable to commercial real estate or toll roads.
The financing model depends on a critical premise: that Nvidia’s graphics processing units will maintain their value over extended periods. Huang argued that the company’s AI factory platform functions as productive, revenue-generating infrastructure with established market adoption across major cloud service providers. However, industry analysts have identified significant depreciation risks. After initial deployment for advanced model training, chips typically shift to lower-margin inference tasks, which reduces their resale value and collateral worth. Some observers estimate investors may demand returns of 11% to 17% to compensate for anticipated GPU depreciation, treating the equipment as high-depreciation assets rather than stable real estate.
China represents a particular concern for the financing plan’s viability. If Chinese manufacturers rapidly expand domestic semiconductor capacity and engage in aggressive pricing, a market price decline could erode collateral values faster than projected debt terms, exposing investors to substantial losses. Current U.S. export restrictions on Chinese chip manufacturers including Huawei have limited immediate competitive pressure, and Nvidia maintains approximately 75% market share for AI chips in the United States.
Nvidia counters depreciation concerns by highlighting its CUDA software layer, which the company contends continuously enhances hardware performance even after deployment, allowing older chips to remain productive longer than traditional accounting models suggest. Rental rates for its H100 chips have climbed from approximately $1.70 per GPU-hour in late 2025 to roughly $2.35 per GPU-hour currently, Huang noted. The resolution of these competing perspectives may prove decisive for the hundreds of billions of dollars now being mobilized for AI infrastructure buildout.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI