Beyond the Chips: Hyperscalers and the Financing of AI

Over the last 18 months, the conversation around artificial intelligence (AI) has focused mostly on models, chips and the huge capital spending plans of companies like Microsoft, Amazon, Alphabet, Meta and Oracle. These firms, often called hyperscalers because of the massive scale of their cloud and data center operations, sit at the center of today’s AI infrastructure buildout. But there’s another part of the story that doesn’t get as much attention, and that’s how all of this is being financed.

As spending on data centers, computing infrastructure, power systems and networking equipment keeps accelerating, the financing behind the buildout is getting harder to ignore. Debt markets are starting to play a bigger role in the next phase of AI development, and that raises an important question about how much supply investors will ultimately be asked to absorb.

The estimates vary, but they all point in the same direction. Morgan Stanley previously estimated roughly US$2.9 trillion of data center investment through 2028,¹ while JPMorgan estimated broader AI-related infrastructure spending could approach US$5.3 trillion through 2030.² Both forecasts lead us to the same conclusion: the AI buildout is going to require an enormous amount of capital in a short period of time.

Historically, hyperscalers funded much of their infrastructure spending through internal cash flow. That’s still true today. But the scale of projected investment has become so large that many companies are increasingly supplementing internal funding with debt issuance, lease commitments, joint ventures and project-based financing structures.

This shift should not be read as a sign of financial weakness. In most cases, it reflects a deliberate capital allocation decision. For context, we estimate cash flow from operations across the big five hyperscalers at US$5.5 trillion through 2030, exceeding their current capital expenditure plans.  Even companies generating substantial cash flow, however, are reluctant to fund a multi-trillion-dollar infrastructure buildout entirely from their own balance sheets. Accessing outside capital preserves flexibility, diversifies funding sources and allows projects to move forward more quickly.

Various estimates point to approximately US$500 billion of hyperscaler-related financing over a multi-year horizon, alongside roughly US$186 billion of high-yield, leveraged loan, and private credit issuance, plus another US$150 billion of asset-backed securities and commercial mortgage-backed securities issuance.³ We also expect a proliferation of non-dollar issuance as hyperscalers look to further diversify their funding markets. Taken together, AI infrastructure is rapidly becoming one of the largest emerging sources of credit supply across global fixed income markets, and it extends beyond technology issuers alone. Utility issuance has stepped up to support incremental power demand and even bank issuance has increased in part due to AI-related loan growth.

A key component of funding that cannot be overlooked is the scale and breadth of financing that is happening off-balance sheet. A growing share of capital is flowing through private credit markets, project finance structures, securitizations, infrastructure vehicles and bespoke financing arrangements tied to specific assets and cash flows. The market has already started to produce transactions that would have been hard to imagine only a few years ago.

Recent examples include Project Quest, a US$4.6 billion refinancing by QTS backed by a Microsoft-leased data center campus in Atlanta, Project River Bend, a US$3.25 billion investment-grade financing tied to a 245-megawatt data center development in Louisiana supported by a Google guarantee, and the largest single tranche ever issued, Project Beignet, a US$27.3 billion financing to build out Meta’s Hyperion facility in Louisiana. Taken together, these transactions show how financing is moving beyond traditional technology issuers and toward asset-backed structures linked directly to AI infrastructure.

Another recent example came in June when Apollo and Blackstone partnered with Broadcom to support a US$35 billion, one-gigawatt expansion plan for Anthropic. The partnership, AI XPV Platform, is eventually expected to enable more than 20 gigawatts of computing capacity by 2028. Just as importantly, it shows how financing is increasingly coming from a mix of private capital, infrastructure investors and specialized funding vehicles rather than traditional corporate bond issuance alone.

Importantly, the majority of financings thus far are tied to the data center buildings, or “powered shells,” but debt markets are increasingly funding the actual chips (GPUs and TPUs), which were previously expected to be funded with equity capital. CoreWeave, for example, recently completed one of the first large-scale public chip financings, raising approximately US$3.1 billion backed by GPU infrastructure and customer contracts.⁴ The chip-related capital needs could be greater than two times the requirements for the powered shells with faster depreciation schedules. These developments suggest that financing markets are adapting quickly as demand for AI infrastructure expands and financing needs become more specialized.

This pattern is familiar from other infrastructure-intensive industries. Early-stage projects often rely on flexible forms of capital willing to absorb development risk. As projects mature and cash flows become more predictable, financing typically migrates toward larger and lower-cost pools of capital.

Who ultimately absorbs this supply is becoming an increasingly important question. Unlike traditional corporate issuance that may be concentrated within a narrower investor base, AI-related financing is attracting demand from investment-grade investors, private credit funds, insurance companies, pension plans, infrastructure investors and buyers of securitized products. That breadth has helped support issuance so far, but the scale of projected financing needs suggests investor demand will continue to be tested as the buildout progresses.

Investors naturally ask whether this amount of issuance will eventually pressure spreads. The answer is probably not in a uniform way. Traditional corporate credit, project financings, private credit transactions and securitized structures all attract different investor bases and compete for capital in different ways. While periods of heavy issuance may create temporary pressure, they may also create opportunities for investors able to move across sectors and structures. The Bank for International Settlements research recently noted that hyperscaler bond issuance exceeded US$100 billion in 2025.⁵ It also observed that spread widening was most acute among lower-rated issuers, suggesting investors are already beginning to differentiate between balance-sheet strength, execution risk and the potential returns from AI-related investments. In that sense, the more interesting question may not be whether supply increases, but where the market ultimately demands greater compensation to absorb it.

It’s important to bear in mind that not all AI-related credits carry the same risk profile. The hyperscaler group is often discussed as a single category, but the financing profiles are becoming increasingly differentiated. Microsoft, Amazon and Alphabet continue to benefit from enormous cash generation, highly diversified businesses and exceptionally strong balance sheets. Other participants have greater exposure to execution risk, customer concentration, funding markets or the pace of AI adoption itself.

The same distinction applies across the broader market. A bond issued by a large diversified technology company is fundamentally different from a financing tied to a single data center project, a cloud provider dependent on a small number of customers, or a development-stage infrastructure asset. As issuance expands, those differences are likely to become increasingly important.

For investors, that may create opportunity. The supply dynamic has widened hyperscaler debt which had previously traded at relatively tight spreads, reflecting strong balance sheets and predictable cash flows. But the inelastic demand for more borrowing requires a dynamic and active approach from bond investors. Meanwhile, off-balance sheet project finance deals may offer additional spread compensation, but require bespoke analysis of key risk vectors including construction risk, tenant quality, contractual structures, asset quality and refinancing risk. As financing activity expands, investors are likely to encounter a much wider range of risk and return profiles than the market has historically associated with large technology companies.

Although AI is often viewed through the lens of the technology sector, the infrastructure supporting future growth extends well beyond technology companies themselves. Data center developers, electrical equipment manufacturers, networking firms, utilities, power generation companies and fiber infrastructure providers all play important roles in the buildout.

A year ago, most discussions centered on access to advanced semiconductors. Today, power has become the more pressing issue. In many markets, securing electricity, transmission capacity, permits and skilled labor is proving just as important as securing chips. In some regions, power availability is increasingly becoming the gating factor for new development. Access to electricity and transmission capacity can no longer be treated as a secondary consideration behind chips and data centers, and local jurisdictions are increasingly weary of the burdens these data centers place on their populations (increased power costs, disruption, unclear direct long term economic benefit for the community). Investors are responding accordingly as utilities, power developers, equipment suppliers and infrastructure providers are receiving increasing attention based on the view that AI’s growth depends on far more than computing hardware alone.

There’s also been a lot of comparison between today’s buildout and the telecom and fiber boom of the late 1990s. The comparison is understandable. Both periods involved transformative technologies, significant capital commitments and optimism about future growth. Yet there are important differences to highlight. Much of today’s infrastructure is being built against identifiable demand rather than speculative demand; computing capacity is often contracted before projects are completed; and many projects benefit from long-term commitments from highly creditworthy counterparties.

Our internal monitoring of supply and demand trends continues to point toward tight conditions across several parts of the market. Demand for compute remains exceptionally strong while constraints persist across power, equipment, labor and broader supply chains. Rather than pointing toward widespread overcapacity today, many indicators continue to suggest portions of the market remain underbuilt.

All this stated, technology has a long history of surprising investors. Improvements in model efficiency could reduce future infrastructure needs; enterprise adoption may not meet current expectations; economic conditions could weaken; and financing conditions could tighten. Forecasting demand several years into the future remains difficult, particularly in an industry evolving at such a rapid pace, and history suggests that transformative technologies rarely follow a straight line with AI unlikely to be an exception.

At the same time, the financing needs associated with AI infrastructure are becoming large enough to influence credit markets in meaningful ways. New issuance is expanding the investable universe across investment-grade credit, high-yield credit, private credit, infrastructure finance and securitized markets. It’s also creating new relative value opportunities as investors evaluate different combinations of credit risk, asset backing, contractual protections and project economics.

The debate surrounding AI often centers on which company will develop the best model, build the most advanced chips, or capture the largest share of future demand. Those questions are important, but they represent only part of the story. With hundreds of billions of dollars of expected issuance across public and private markets, the AI buildout is creating one of the largest financing waves credit markets have ever seen. Long after today’s debates about models and applications evolve, the infrastructure built to support them and the capital raised to finance that infrastructure may prove to be among the more lasting legacies of the current cycle.