How do you raise money and what collateral can you use? Physical assets like gold, land, securities and bonds and, now a report highlights Nvidia GPUs like the H100 that are used to train and power generative AI systems like GPT. Yes that is right. There are now tech companies that are not only buying Nvidia graphics and AI chips to power their services, but are also using these chips as collateral to raise more money.

According to a report by Financial Times, some tech companies like CoreWeave, Crusoe and Lambda Labs are increasing the craze for Nvidia chips. Called neocloud companies, these are specialized cloud service providers that focus on offering the high-performance computing power needed for AI and data-intensive tasks. They are investing heavily in advanced hardware like GPUs to support businesses and researchers developing complex AI models.

According to the Financial Times, these companies are now making full use of Nvidia’s powerful GPUs. In turn, they are also doubling these GPUs as collateral for huge loans. This has led to a frenzy to buy GPUs among Wall Street lenders, who are eager to invest in the current AI boom.

For example, the company mentioned above, CoreWeave, started in 2017 with the hope of making a lot of money by mining cryptocurrencies, but it soon shifted its focus to AI. Now, it’s one of the big names in the neocloud landscape, specializing in powering AI tasks with serious hardware. As the FT reports, CoreWave claims it is the largest private hoarder of Nvidia GPUs in all of North America, with a staggering stockpile of more than 45,000 AI chips. Think of the company as the ultimate GPU collector.

But this is where things get tricky. The GPU economy is growing so fast that some worry it could lead to risky lending practices and Nvidia’s stranglehold on AI technology. Nvidia, which is worth $3 trillion, provides GPUs to these neocloud companies, which then encourages lenders to hand over more cash – so these companies can buy, you guessed it, more GPUs.

Interestingly, according to the report, these neocloud companies are deeply tied to Nvidia, which gives them the GPUs they need to thrive. CoreWeave reportedly secured thousands of Nvidia’s H100 GPUs as a “preferred partner,” but the company also knows that staying competitive also depends on acquiring Nvidia’s upcoming Blackwell chips. Meanwhile, Nvidia says it doesn’t play favorites.

While the buzz continues about AI chips like the H100 and H200 made by Nvidia, experts are sounding some alarm bells. Compared to a year ago, Nvidia is facing a renewed challenge from competitors. Many tech giants like Meta are also designing their own AI chips. AMD, a staunch rival of Nvidia and the world’s second-largest maker of discrete graphics cards, is also moving to compete with high-powered GPUs of its own. In the midst of all this, as neocloud companies rely on Nvidia’s chips as loan collateral, the value of this setup is no longer as certain. Critics in the industry are now asking whether these GPUs, which are being treated like valuable assets, will retain value even as new models come out or whether AI spending will begin to decline.