In the fast-paced world of artificial intelligence, where innovation is measured in the ability to process vast datasets, a unique challenge is emerging—one that financial institutions might soon address. With generative AI startups experiencing an extraordinary funding boom, the need for compute power, specifically in the form of GPUs, has become a critical bottleneck. Startups are grappling with increased fundraising demands to secure these essential resources, and a recent report by Air Street Capital suggests a novel solution on the horizon: ‘GPU debt funds.’
The AI funding boom
Nearly a year after the launch of ChatGPT-3, the AI landscape has transformed significantly. Generative AI startups, in particular, are reaping the benefits, attracting a staggering $18 billion in venture capital funding in 2023, as outlined in Air Street Capital’s State of AI report. This exponential increase, five times the 2022 figures, underscores the industry’s rapid growth and the increasing relevance of AI applications across sectors.
Amidst this boom lies a pressing challenge for AI startups. The surge in demand for AI services, driven by the scaling of datasets and models, has led to a race for compute power. Startups are struggling to secure GPUs, either through direct purchase from chipmakers or by renting them through cloud providers. The competition is fierce, with market leader Nvidia facing a chip shortage, amplifying the difficulties for startups in accessing the critical hardware.
According to Nathan Benaich, co-author of the Air Street Capital report, startups developing large AI models are resorting to raising significant rounds to finance the acquisition of private clusters. Some are even selling equity to VC firms to gain access to GPUs from cloud providers, a practice that Benaich argues is detrimental to both founders and investors.
The rise of GPU debt funds
In response to this funding dilemma, the report predicts a potential shift in the financing landscape. Financial institutions, particularly banks, may introduce “GPU debt funds” within the next 12 months. These funds could serve as an alternative to traditional VC equity, offering a lifeline to startups grappling with compute power challenges. The concept revolves around leveraging GPUs as collateral for loans, providing startups with the necessary funds to secure the essential hardware.
The longevity of GPUs adds to their appeal for such debt funds, as highlighted in the report. Regulators may find this approach attractive, as it aligns with the encouragement of responsible non-dilutive funding, carrying fewer regulatory requirements compared to equity financing.
The dilemma of using AI chips as loan collateral
While the idea of GPU debt funds presents a potential solution, it is not without challenges. Some startups have already ventured into using their chips as collateral for loans, with Coreweave using its H100 chips for a $2.3 billion debt facility—an approach deemed risky by the report’s authors. Benaich acknowledges that the adoption of GPU debt funds may not happen overnight but suggests that, with prevailing high-interest rates, private credit is becoming increasingly appealing.
Air Street Capital’s predictions extend beyond the financing realm. The report anticipates a further integration of AI companies into mainstream consciousness in the coming year. Databricks, an AI-focused analytics firm, is forecasted to file for an IPO, signaling the industry’s maturation. Also, the report speculates on the possibility of an AI-generated song breaking into the Billboard Hot 100 Top 10, highlighting the growing impact of AI on creative domains.
As the AI industry hurtles forward, the race for compute power has become a defining challenge for startups. The potential introduction of GPU debt funds by financial institutions represents a strategic response to this dilemma, offering a new avenue for funding and potentially reshaping the dynamics of AI startup financing in the next 12 months.