The AI's Hidden $1.65 Trillion Bet Meets $100+ Oil

Two Crises, One Very Bad Day

On the surface, July 23 looked like a straightforward bad day for markets: oil back above $100 a barrel, shipping routes under pressure, and a rough session for equities. Blame the war. Move on.

But the surface story misses something important, and probably dangerous. Two separate slow-motion crises are now colliding in a way that amplifies each other, and neither one is as well understood as it needs to be.

The Strategic Petroleum Reserve Is More Broken Than You Think

The US Strategic Petroleum Reserve has long been the adult in the room when oil prices spike. Politicians love it precisely because it lets them smooth over the pain at the pump before an election. The problem is that they have loved it a little too much.

Both parties have contributed to drawing down 352 million barrels over the last four years alone. The reserve is now at its lowest point since 1983. That is bad enough on its own. What makes it worse is the structural reality: those 60 underground salt caverns were originally designed to handle roughly five full drawdowns. They have been cycled far beyond that, and last December the US Energy Department flagged that it would take approximately $230 million just to address the most pressing infrastructure repairs.

That money has not been spent. The repairs have not happened. And with the US-Iran war showing no signs of producing a durable peace, the administration has no obvious lever to pull if oil continues to climb. The reserve is not just depleted. In some meaningful sense, it is compromised.

The Hidden Debt Propping Up the AI Build-Out

At the same time, the companies most responsible for keeping equity markets from a sharper decline are quietly carrying a level of financial exposure that most investors have not fully absorbed.

A study by Nikkei Asia, which has received far less attention than it deserves, analyzed Alphabet, Microsoft, Amazon, Meta, and Oracle. The finding is striking: these five companies hold approximately $1.65 trillion in off-balance sheet commitments. That is more than the roughly $1.35 trillion in obligations already recorded on their balance sheets. Together, the real number is not the one investors are looking at.

Meta stands out even within this group. It holds approximately $420 billion in off-balance sheet debt, partly through special purpose vehicles and private credit arrangements, including deals with firms like Blue Owl Capital. As Neeta notes, if and when those deals go south, the fallout is unlikely to stay contained.

The mechanism funding all of this has scaled at a pace that should raise eyebrows. Corporate debt issuance and special purpose vehicle financing for AI infrastructure surged over 500% since 2024. These structures are designed to keep obligations off the headline numbers. They also make it genuinely difficult for outside investors to gauge how much risk is actually sitting in the system.

Google's Free Cash Flow Is the Canary

Google's week illustrated the bind that hyperscalers are now in. On Monday, news of its new Frozen 2 server chip, which embeds Gemini technology and promises significant energy efficiency gains over current TPUs, sent the stock higher. On Wednesday, the company raised its full-year CapEx forecast from a range of $180 to $190 billion up to $195 to $205 billion. Quarterly CapEx came in at approximately $44.9 billion. Free cash flow turned negative. The stock fell roughly 6%.

The efficiency gains are real. The spending commitments are also real, and they are locked in. Hyperscalers are tied to long-term contracts for chips, data centers, power, and physical infrastructure. Cutting back means handing an opening to competitors. Not cutting back means watching free cash flow deteriorate while open-weight models from Chinese labs and US challengers like Thinking Machines and NVIDIA's Nemotron erode the pricing power that the frontier labs have been counting on.

Where the Two Crises Meet

This is where the oil shock and the AI debt problem become something larger than the sum of their parts.

Rising energy costs hit data centers directly. AI infrastructure is extraordinarily power-intensive, and higher oil prices feed through into electricity costs, financing costs, and operating costs across the entire build-out. That pressure arrives precisely when the revenue case for all of this spending is under the most scrutiny it has faced.

The circular nature of the lending makes the downside scenario harder to model and harder to contain. The big tech firms and the banks and private credit funds providing their capital are deeply intertwined. If utilization rates on new data centers disappoint, off-balance sheet obligations start moving onto balance sheets. If AI pricing fails to hold as cheaper models proliferate, revenue projections that justified trillions in investment start looking like the aggressive, outsized ambition they probably always were.

The AI boom does not just need to grow. It needs to grow enough, and fast enough, to justify a level of capital commitment that is considerably larger than the public numbers suggest, in an energy environment that is getting more expensive and more unpredictable. That is a high bar. Right now, it is not clear who has a convincing plan to clear it.