Financing the AI Boom 3

In the rapidly evolving world of artificial intelligence, GPU prices have taken an unexpected turn, defying expectations and defying traditional market trends. In November of last year, it was suggested that GPU prices might decline slowly or even precipitously, but the reality has been quite different. According to Silicon Data, the cost of renting a Nvidia Hopper H100 chip for an hour has surged to $2.71, up from $1.96 at the end of November.

This upward trend is not slowing down anytime soon, with forward rates projected to continue rising well into 2027 and 2028. Even older GPU models are appreciating in value. CoreWeave, a company that rents out GPU infrastructure to AI developers and enterprises, has signed a contract to lease Nvidia Ampere A100 chips, which were introduced in 2020, all the way out to 2029.

Nvidia CEO Jensen Huang has expressed confidence in the durability of these chips, stating that the A100 fleet is mission-capable from 2020 through 2029. Nvidia's computing platform, backed by CUDA, allows for continual upgrades and flexibility, making it highly versatile and fungible. This versatility is driving utilization and extending the lifetime of these GPUs, making them attractive assets for financing.

Nvidia's decision to mobilize over $500 billion of third-party capital to support the development of AI infrastructure further highlights the potential for this asset class to be recognized as a valuable and financeable investment. Industry leaders like Blackstone president Jon Gray and BlackRock CEO Larry Fink have echoed this sentiment, emphasizing the transformative potential of AI compute in the financial world.

As this new asset class continues to grow, it may even rival established markets like mortgage-backed securities in terms of size and financial engineering opportunities.