The buildout of artificial intelligence infrastructure is on track to create a multi-trillion-dollar credit market, with over $7 trillion of debt outstanding by 2029. That figure would establish AI debt financing as the second-largest asset-backed debt market globally, trailing only the U.S. mortgage-backed financing market at just over $13 trillion.
Cumulative AI capital expenditure from 2024 to 2029 is projected to reach approximately $11.1 trillion. Annual AI capex—encompassing GPUs, networking, storage, attached CPU compute and the data centers housing this hardware—will exceed $2 trillion in 2028. Credit markets are emerging as the primary funding source for this buildout.
Major hyperscalers including Google, Amazon, Meta, Microsoft and Oracle initially funded AI expansion through cash flow. Over the past year, Oracle, Meta and now Google have shifted toward debt financing for AI initiatives, signaling a broadening reliance on external capital across the industry.
Nvidia's revenue continues to climb. Despite this, general market estimates for Nvidia's shipment volumes and revenue in the second half of this year remain materially lower than supply chain data tracked in the Accelerator Model suggests.
Executing any AI compute buildout requires addressing what practitioners call the "AI Project Trinity": Capital, Offtake and Datacenter. Lenders currently demand an offtake contract or a backstop from an investment-grade hyperscaler before extending debt financing. Securing an offtake, however, first requires equity capital to place deposits for IT equipment—creating a circular dependency.
A Neocloud must either secure a committed offtaker and lending to convince data center operators to rent colocation space, or build the data center itself. Many Neoclouds have worked through this structure, often with private equity sponsorship.
Nvidia actively seeks to enable smaller buyers rather than concentrating sales among a few large entities. These smaller buyers have the financial capacity to purchase GPUs but often lack the credit ratings creditors require. Nvidia is attempting to serve as a financial backstop to bridge that gap.
Some companies are working to reduce dependence on a single vendor. Anthropic uses Google's Tensor Processing Units, and OpenAI is shifting inference workloads toward non-CUDA paths—moves that hedge against Nvidia's pricing power and aim to capture structural cost advantages.
Beyond chip supply, the industry faces harder constraints. Grid capacity and power availability involve regulatory timelines, infrastructure permitting and national policy decisions that no company can accelerate through large purchase orders.
Nvidia CEO Jensen Huang described manufacturing bottlenecks as a "2-3 year problem." Chip supply issues are significant, but the energy infrastructure required to power approximately 800 AI factories being built globally—most not yet publicly announced—represents a separate, longer-term challenge.
Nvidia has already committed $95.2 billion in supply-related obligations. While the company is innovating in areas such as cooling systems, those efforts reduce specific consumption points—water use within the cooling mechanism, for instance—without resolving the broader energy and water footprint tied to electricity generation and hardware manufacturing. The core constraint for the AI buildout extends beyond hardware supply and into fundamental infrastructure capacity.
