Wall Street's decision to team up with Nvidia to arrange $500 billion (about 712 trillion won) in artificial intelligence infrastructure financing rests on a belief that the value of Nvidia's chips will hold up over the long term, according to analysts. In the deal, which takes the form of a loan backed by Nvidia chips, Nvidia maintains that its chips retain value for up to 10 years. But because lenders typically extend loans for less than the value of the collateral, the actual repayment period for the chip-backed loans has been found to run for up to five years.
The Financial Times reported on the 12th, citing a person in the financial industry involved in the agreement, that behind the Nvidia-Wall Street partnership lay a forecast that the AI boom would keep semiconductor prices at elevated levels for longer than expected.
Earlier, Nvidia said on the 10th that it had signed memorandums of understanding with six financial firms — Apollo Global, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR — to create a dedicated $500 billion funding pool for building AI infrastructure.
The financial industry's belief in long-term value runs counter to the conventional notion of depreciation, under which an asset naturally loses value over time. It also rebuts the "AI bubble" theory, which holds that massive AI infrastructure spending could turn out to be overinvestment if AI semiconductors have shorter useful lives than expected.
Still, judging by loan repayment periods, the effective collateral value of the chips is about five years at most. According to the FT, many lenders on loans backed by chip lease agreements require full repayment within three to five years. That reflects an assumption that after five years at most, the semiconductors put up as collateral will retain little residual value. By contrast, companies that lease the semiconductors need more than five years to build data centers and generate returns, which could increase their financial burden.
Ben Bajarin, a technology analyst at the Silicon Valley market research firm Creative Strategies, noted that there was a risk that data centers could be overbuilt and demand could slow, or that improved models could require less computing power.
Six Years On, the A100 Is Still in Service: A '3-Year vs. 10-Year' Debate Over GPU Lifespan
Michael Burry, the real-life figure behind the film "The Big Short," is a leading voice among AI boom skeptics. He argues that the economic life of AI servers and semiconductor equipment may be shorter than the market assumes. Last November, Burry contended that technology companies such as Meta, Oracle and Microsoft were inflating profits by setting depreciation periods for AI semiconductors longer than their actual usage periods and thus recognizing less expense. He added that the actual replacement cycle for server equipment was about two to three years. Some Silicon Valley firms are already projecting that loans of up to 70% of collateral value are available on three-year repayment terms, with interest rates ranging from 9% to 18%.
Indeed, Nvidia has shown a rapid release pace, unveiling one graphics processing unit (GPU) lineup after another — Hopper (2022), Blackwell (2025), Vera Rubin (2026) and Rubin Ultra (2027). Such an aggressive release pace, combined with the aging of equipment and demand for new technology, could accelerate replacement cycles.
Nvidia counters that the economic life of its GPUs is longer than that. Last year, Nvidia pushed back against Burry's view, saying that customers' actual usage patterns showed GPUs depreciated over four to six years. Nvidia Chief Executive Jensen Huang, in disclosing news of the AI financing platform deal with Wall Street on the 10th on X, formerly Twitter, also stressed that six-year-old A100 chips were still in commercial use. Huang said that because customers were adding capacity for multi-year deployments, the economic life of the A100 was being extended to as long as 10 years.
There are other optimistic signs. Nitin Agrawal, chief financial officer of the AI cloud company CoreWeave, recently told investors that the company had signed a contract to lease Nvidia A100 chips through 2029. According to Agrawal, most of CoreWeave's inventory of Nvidia's older-generation chips has been used up. Silicon Data, a firm that tracks GPU prices, also said that the industry was still learning about the economic life of GPUs, and that it did not appear to be the two to three years that some casually assume.
On this, the U.S. business outlet Business Insider assessed that the fact that the A100 was still attracting customers through 2029 was a strong validation supporting the outlook that AI hardware would hold its value for years, and that so far it could be seen as an encouraging sign.
