The Big AI Hyperscaler Gamble

It feels like every day we hear about the next multi-million, if not multi-billion-dollar, chip or infrastructure deal fuelling the AI race. We've all heard about the circular funding schemes of chipmaker invests in an AI provider, the AI provider buys cloud capacity from the infrastructure provider, the infrastructure provider places a gargantuan chip order, and round we go again. But how is all of this possible?

Hyperscalers like AWS, Microsoft Azure, Google Cloud, and Oracle are engaged in a fierce race to expand their AI computing capacity. They are effectively building a new global computing infrastructure, but the financial commitments are staggering:

• Hyper-Scale Data Centres: Microsoft added approximately 2 GW of data centre capacity in fiscal 2025 and plans to double its footprint in the next two years. Meta, for example, has committed to investing at least $600 billion on U.S. infrastructure over three years.

• AI Hardware: An estimated 90% of hyperscaler capital expenditure (capex) is now dedicated to equipping these sites with powerful AI hardware, such as Nvidia GPUs and custom chips.

• Power and Connectivity: This requires massive investment in advanced cooling (e.g., liquid cooling) and electric power capacity (tens or hundreds of megawatts per facility). In addition, they are funding colossal network infrastructure, including undersea cables and high-capacity fibre routes, such as the partnership between AWS and Verizon.

The Rise of Creative Financing

The scale of these AI infrastructure programmes is so colossal that even the world’s most cash-rich firms cannot fund it solely from internal operating cash. As their capex balloons (Microsoft’s spend alone is around $80 billion for FY2025; Alphabet’s is $75–93 billion), hyperscalers have been forced to look for new external sources of funding.

They are deploying several sophisticated financing mechanisms:

1. The AI Bond Boom

Tech giants have tapped corporate bond markets at a record scale. In just September and October 2025, investment-grade bonds worth $75 billion were issued by “AI-focused” Big Tech firms in the U.S.. Companies like Oracle are locking in long maturities, with some bonds spanning up to 40 years, to match the long-lived nature of the assets they are funding. This debt provides upfront capital for capex and offers a tax shield (interest payments are tax-deductible), making borrowing cheaper than equity.

2. Off-Balance-Sheet Structures

To avoid overburdening the core balance sheet with debt and to preserve credit ratings, companies are employing structured finance vehicles.

• SPVs (Special Purpose Vehicles): Meta completed a $27 billion deal in September 2025 with Blue Owl Capital. An SPV raised the funds (mostly debt) to build a Louisiana AI data centre, with Meta acting as the long-term tenant. By using this structure, Meta avoids carrying the debt directly on its books.

• Project Finance: Oracle is using a similar playbook for the "Stargate" initiative. In November 2025, a consortium of around 20 banks provided an $18 billion project loan. This loan is tied to the cash flows of the data centre facility, rather than Oracle's corporate balance sheet.

These "off-book" financings allow hyperscalers to share risk and capital burden with infrastructure funds and investors.

3. Leasing and Capacity Contracts

Leasing has re-emerged as a vital tool for speed and flexibility. Rather than owning every site, hyperscalers lease capacity or enter capacity contracts with third-party providers.

• Microsoft, in just Q1 FY2026, added $11.1 billion in new data centre lease commitments, with total future lease obligations signed but not yet commenced reaching a staggering $92.7 billion.

• Microsoft also signed multi-billion-dollar capacity contracts with “neo-cloud” firms such as Nebius and CoreWeave ($33 billion in total). This means the smaller providers assume the capital expense and debt, backed by Microsoft’s contract as collateral.

This outsourcing of capex gives the hyperscalers flexibility to scale rapidly, which is crucial because, as Microsoft’s CEO Satya Nadella noted, "Demand is increasing".

What does this all mean if we are in bubble?

The rationale behind these funding strategies is clear: Speed to Market. There is a competitive “land-grab mentality” in AI infrastructure, and companies are willing to leverage up to seize market leadership. However, this unprecedented financial activity comes with substantial financial, market, and systemic risks.

Risk 1: Debt Burden and Credit Strain

The surge in borrowing means higher fixed costs. While giants like Alphabet and Microsoft maintain modest debt-to-equity ratios (11.5% and ~33%, respectively), firms like Oracle are a cautionary tale. Oracle’s long-term debt has ballooned to ~$82 billion, pushing its debt-to-equity ratio to approximately 450%. Rating agencies have grown cautious; Moody’s gave Oracle a negative outlook, warning of the "significant risks" associated with these massive obligations, even if financed through leases or engineered vehicles.

Risk 2: Demand Uncertainty

The rush to build assumes surging, sustained AI demand. But if the AI “hype” doesn't translate into profitable usage, there is a serious risk of overcapacity, in a very similar way to the dot-com boom.

A prime example is the Oracle-OpenAI commitment, rumoured to be $300 billion over five years. Analysts openly question if OpenAI can generate the required revenue to justify such a monumental spend. If OpenAI cannot honour the contract, Oracle risks having built enormous, underutilised capacity; a classic bust scenario.

Risk 3: Opaque Financial Hotspots

A significant portion of this financing is occurring through opaque, illiquid channels, creating what the Bank of England referred to as “pockets of risk”. The growing reliance on private credit markets and securitised products (like Asset-Backed Securities based on data centre lease streams) means that if AI ventures struggle in any way, the ripple effect could hit lightly regulated credit funds. This heavy use of financial engineering, rewarding aggressive spending through debt and off-sheet funding, is a classic hallmark of speculative fever.

What happens next?

The sustainability of this financed expansion depends entirely on whether AI demand justifies the lofty expectations. If AI revenues flow, the strategy will be validated, and debt can be paid down. If not, a "bursting bubble" scenario could see companies abruptly cutting capex and credit spreads widening dramatically.

A shake-out of some description from all of this has to be likely. Ambitious deals will almost certainly be delayed, renegotiated, or reassigned if AI demand growth lags spending. The largest hyperscalers, backed by diversified businesses (e.g., Amazon’s retail cash flow, Google’s ads), have buffers to weather a moderate bubble burst. But for now, they are leveraging every financial tool available, betting that the cost of capital, even when high, is preferable to the cost of missing the AI wave.