Artificial intelligence: can GPU compute become a financeable asset class?

The question reflects the growing importance of graphics processing units (GPUs), which underpin many AI services and are increasingly being monetised through GPU-as-a-service (GPUaaS) business models. Rather than generating value from ownership of traditional physical infrastructure, these businesses derive revenue from providing customers with access to compute capacity.

The issue has come into sharper focus following Nvidia's plans to support up to $500 billion of AI infrastructure investment. The scale of capital being deployed highlights not only the investment required to build AI ecosystems, but also the need for financing structures capable of supporting businesses whose core asset may be access to computing power rather than the hardware itself.

While much of the market's attention has focused on financing the acquisition or leasing of GPUs, a separate consideration is whether revenue generated from GPU compute services supports the kind of lending structures traditionally associated with project finance and infrastructure finance.

Two models emerging

Two models are beginning to take shape. The first involves financing the acquisition or leasing of physical chips, helping technology companies overcome supply and capital constraints. Questions around title and residual value risk are central to this structure. The second, less developed model focuses on financing businesses that generate revenue by providing access to GPU compute capacity – and this is where the more novel and complex questions arise. This structure turns more on the contractual revenue streams and access rights.

A different risk profile

These businesses may not necessarily own the underlying hardware. Instead, they may hold contractual rights to access computing capacity, which they then make available to customers. That creates a different risk profile and raises questions about whether traditional project finance and infrastructure finance principles can readily be applied.

Unlike conventional infrastructure assets, at the heart of many GPUaaS businesses is a long-term contractual arrangement under which the provider obtains access to compute capacity from a cloud provider, data centre operator or infrastructure platform, as well as the revenue generated from customer demand for that capacity. While comparisons have been drawn with natural resources and energy projects, important distinctions remain.

Absence of established frameworks

In sectors such as oil and gas, lenders can assess reserves, production profiles and commodity markets using established methodologies and independent verification. There is currently no widely accepted framework for assessing the long-term value and durability of GPU compute capacity, and that gap is one of the key challenges that needs to be addressed.

Oil and gas fields can remain productive for decades, whereas the economic life of advanced computing hardware may be measured in a handful of years before newer technologies enter the market. This risk has direct implications for lenders: it affects amortisation schedules, assumptions about residual value, and the appropriate tenor for any financing. While compute infrastructure can be upgraded, lenders still need confidence in the durability of the revenue generated by those assets.

The absence of standardised market documentation may also present challenges as the sector develops.

Cashflow and revenue certainty

Traditional energy and infrastructure projects often rely on long-term revenue contracts that support the financing structure. Power projects, for example, are routinely backed by long-term power purchase agreements. Such arrangements provide lenders with a degree of certainty over future cashflows and form a key component of the credit analysis.

In contrast, there is not yet a market-standard equivalent for GPU compute. The availability of predictable, long-term contractual revenue is one of the factors lenders typically consider when assessing whether an infrastructure-style financing can support meaningful leverage.

Equally important will be the contractual arrangements ensuring access to compute capacity. In an infrastructure-style financing, lenders take into account fuel supply agreements and feedstock agreements. Similarly, where a GPUaaS provider relies on third party infrastructure providers, lenders will assess whether capacity arrangements are sufficiently durable and transferable to support long-term financing.

Security, enforcement and insolvency risk

Questions also remain regarding security and enforcement arrangements. Traditional project financings are commonly supported by security over physical assets, contracts, receivables and project bank accounts. In a GPUaaS structure, the most valuable assets may instead consist of contractual rights to access compute capacity, together with customer contracts and associated revenue streams.

For lenders, a key question is whether effective security can be taken over those contractual rights and what happens if the cloud provider, data centre operator or infrastructure platform becomes insolvent or the underlying compute agreement is terminated or otherwise disrupted. Contractual protections become particularly important because, unlike many infrastructure assets, the value proposition may depend heavily on the continuation of a service relationship.

Geopolitical and regulatory risk

At the same time, geopolitical and regulatory considerations are becoming increasingly relevant.

For instance, export controls imposed by the US Bureau of Industry and Security on advanced computing chips have introduced additional uncertainty into the market. Those restrictions may affect deployment, transfer and enforcement scenarios in ways that are not typically encountered in traditional infrastructure financings.

Lenders will want to understand how those restrictions could affect the transferability, availability and continued use of the compute capacity on which the business model depends.

The path forward

Despite these challenges, the foundations of a GPUaaS financing market are beginning to emerge – particularly in Southeast Asia, where significant investment has already flowed into data centres and digital infrastructure.

The next stage will be determining whether the market can develop the contractual, regulatory and financing frameworks needed to support large-scale lending against GPU-backed revenue streams. Over time, we may see the emergence of standardised compute offtake arrangements, clearer security packages and greater contractual certainty around access rights. That process is likely to require cooperation between infrastructure providers, hyperscalers, customers and financiers.

If those building blocks can be established, GPU compute capacity may increasingly be viewed not simply as a technology input, but as a financeable asset class capable of supporting infrastructure-style lending. As the market matures, the focus may shift from financing the chips themselves to financing the businesses and cashflows they enable.