The AI Buildout Is Accelerating. Its Financing Is Already Cracking.

The 30-year Treasury yield hit 5.20% on August 3 — a 19-year high. That number, not any disappointment about artificial intelligence, is what killed Leopold Aschenbrenner's Situational Awareness fund. The fund was leveraged four times. When the yield spiked, the positions cratered, and prime brokers at Goldman Sachs, JPMorgan, and Bank of America forced a liquidation. The $45 billion public equity book went to Citadel in a single block trade ^[1]. Jim Cramer had a different name for it.

In the collapse of Situational Awareness, we realize what we had is a clearing event — it wiped out all your fellow shareholders with weak hands. — Jim Cramer

It cleared nothing. The cause was not an AI disappointment but rate-sensitivity — the vulnerability of highly leveraged capital to a move in government bond yields. And that same rate-sensitivity now runs through the financing stack beneath the AI buildout. The concrete is pouring. In the twelve months through June, five U.S. tech giants — Alphabet, Amazon, Meta, Microsoft, and Oracle — spent $313 billion on AI infrastructure, more than double the $146 billion they spent three years earlier ^[2]. McKinsey estimates global companies will invest nearly $7 trillion in data center infrastructure by 2030 ^[3].

In the United States, AI-related capital expenditures account for about 5% of GDP and have been growing at high-single- to low-double-digit pace. — McKinsey & Company

Fifty-one U.S. utilities plan $1.4 trillion in grid upgrades through 2030, a 21% increase over the previous five-year period ^[4]. OpenAI, which had already projected $600 billion in computing infrastructure spending, raised its own estimate to $750 billion through 2030 — even as its CFO Sarah Friar expressed concern that revenue growth may not keep pace ^[5]. Real demand exists beneath these numbers. Global cloud spending hit $102.6 billion in the third quarter of 2025, up 25% year-on-year, with AWS holding a $200 billion backlog ^[6]. Jefferies reports a 12-gigawatt gap between operational data centers and demand ^[7]. The question is not whether the infrastructure is needed. It is whether the money paying for it holds. The financing has shifted. Where the buildout was once funded from retained earnings — the cash hoards of trillion-dollar companies — it now runs increasingly on debt, private credit, and structures that depend on low rates and willing lenders. Oracle carries $91 billion in total debt. Its interest payments now consume more than 20% of its operating income. OpenAI's cloud partners and data center providers have accumulated roughly $100 billion in debt collectively ^[8]. Meta, rather than funding data centers from its own balance sheet, secured $27 billion from private credit firm Blue Owl Capital for data center construction ^[9]. Aware Super CIO Simon Warner identified this shift as the top financial market risk for 2026.

There's been some instances in the last six months or so where that's softened a bit, there's more circular financing, a bit more conduit financing. Nothing that flashes red, but things that certainly flash orange. — Simon Warner

Warner was describing a specific structural change — the migration from retained earnings to private-credit-backed conduit financing — and he was explicit about what happens if it breaks.

I do think there is a dynamic there, but if one of those pillars was to stumble, then we could have a correction. — Simon Warner

Alex Davis, CEO of Disruptive, warned that speculative data-center developers using short-term financing while relying on "hope that users will arrive" face a significant financial crisis by 2027 or 2028, as hyperscalers increasingly build their own facilities ^[10].

We foresee a significant financial crisis — Alex Davis

The cracks are already visible. Blue Owl Capital — the same firm that provided Meta's $27 billion in data center financing — failed to secure third-party financing for its own $4 billion data center in Lancaster, Pennsylvania. The reason: its tenant, CoreWeave, carries a B+ credit rating, below investment grade. Blue Owl was forced to disclose a $500 million bridge financing obligation due through March 2026 ^[11].

the project, which he said is already under construction, is fully funded, on time, and on budget. — Blue Owl Capital

Hyperscale Data Inc. is monetizing approximately 100 bitcoin and drawing on a bitcoin-backed credit facility at variable rates of 4.5% to 5% to fund an AI data center campus in Michigan ^[12]. The infrastructure is being built at a pace that depends on the money holding. The money is stacked in structures that a single yield spike could unravel: variable-rate debt, short-term financing for long-term assets, private credit dependent on below-investment-grade tenants, bitcoin-collateralized loans. A yield spike already killed the most leveraged player. The same mechanism runs through the structures paying for the concrete.