Article — "The $1.8 Trillion Off-Balance-Sheet Bill Comes Due as Capex Guidance Climbs Toward $1.4T": Alphabet reportedly burned $5.9 billion in cash in the second quarter — its first-ever quarterly cash burn, according to a Bloomberg report. Not a slowdown in cash generation. Cash out the door, net, for the first time in the company's history.
That number is the whole story, and almost no one is talking about it. This week the debate was somewhere else entirely.
Analysts spent the week fighting over whether hyperscaler capex lands at $1.1 trillion or $1.4 trillion in 2027 — reportedly Goldman's range, and now the accepted terms of the argument. Alphabet reportedly just raised its own 2026 capex guide to $205 billion to sharpen the point.
Here is the problem with that fight: both numbers are on the balance sheet. Both are visible. Both are, for now, financeable through the bond market.
The number that actually pre-commits future cash flow never shows up there at all.
Last week we walked through the mechanics of the $1.8 trillion in off-balance-sheet exposure — the purchase commitments, the unactivated leases, the vendor financing. This week the thesis got its confirmation: the market is now pricing the cash drain, and Alphabet's Q2 burn is the tape catching up to the footnotes.
Start with Morgan Stanley's split, because it is the load-bearing figure. Roughly $1.8 trillion in off-balance-sheet operating leverage, per Morgan Stanley: about $1 trillion in long-term purchase commitments, plus more than $800 billion in lease commitments.
These do not appear as debt. They appear in footnotes — the part of the filing no analyst reprices when the stock moves.
But a purchase commitment is a contractual obligation to spend future cash. An unactivated lease is future cash locked up before the asset ever hits the books. The distinction between this and debt is legal, not economic.
That is the sleight-of-hand. The $1.1–1.4 trillion capex number is the one everyone can see, so everyone argues about it. The $1.8 trillion that actually determines whether these companies have any financial flexibility left sits one layer down, unrepriced.
The number that actually determines whether these companies have any financial flexibility left sits one layer down, unrepriced.
Now the plumbing, because this is where it gets genuinely opaque. Morgan Stanley reportedly describes private-credit firms — Apollo and Blackstone specifically — routing leverage down into the supply chain through special purpose vehicles.
The effect: leverage gets pushed off the hyperscaler's balance sheet and layered into the vendors, the data-center developers, the financing entities. It becomes circular, and it becomes hard to trace. You cannot see the whole chain from any single filing.
That structure matters enormously depending on which way cash is flowing. When free cash flow is abundant, off-balance-sheet commitments are a footnote — genuinely cushioned by the operating business.
When free cash flow goes to zero, they are something else entirely. They become fixed claims on cash that no longer exists internally, which means they have to be funded externally — with new debt.
And that is exactly the transition underway.
Gross leverage across the hyperscalers reportedly jumped from 0.9x to 1.8x in two quarters — a doubling. Roughly 0.3x per quarter. Morgan Stanley reportedly notes this is faster than the energy sector, which is not a comparison anyone expected to be making about companies that were, until recently, cash-printing machines.
Capex is growing faster than revenue and faster than free cash flow. And the depreciation from this buildout has reportedly not fully hit the income statement yet, per Morgan Stanley — meaning the reported earnings drag is still ahead, not behind.
Pair the leverage math with the cash math and the picture sharpens.
Alphabet already burned that $5.9 billion. Per Morgan Stanley's forecasts, Amazon and Meta free cash flow trends toward zero or negative in 2026. Meta's cash flow is reportedly forecast to fall 95.7% to $1.85 billion. Microsoft's cash generation is reportedly expected to drop to roughly half the prior year's estimate.
Look at capex as a share of revenue and you see why. Meta is reportedly projected to go from 35.9% to 54.9% of revenue. Alphabet from 23% to 41%. Microsoft from 31% to 45%. Amazon from 18% to 25%.
When capex eats half your revenue and free cash flow approaches zero, financing becomes almost entirely dependent on new debt. There is no internal cash left to fund the commitments.
This is the mechanism that turns the off-balance-sheet structure from cushion into accelerant. In a cash-rich world, those $1.8 trillion in commitments are reportedly absorbable, per Morgan Stanley. In a zero-FCF world, every commitment that comes due has to be met with a new bond issue.
And the debt is already being issued at scale. AI-related bond issuance reportedly hit $236 billion globally by the end of May, up 357% year over year, on pace to break $570 billion for the full year. In April alone, more than $74 billion. Data-center financing made up 85% of high-yield supply and 40% of investment-grade supply that month.
Amazon, Meta, Google, Microsoft and Oracle reportedly now make up roughly 4% of the investment-grade bond index. This is no longer a niche. It is becoming a structural feature of the credit market.
Here is where the leak onto the tape shows up. Meta is reportedly trading wider than the CDX IG benchmark, per Morgan Stanley. That is not noise.
Bond supply pressure has reportedly already dragged hyperscaler spreads from AA to A. Meta trading wider than an investment-grade index of far weaker credits is the market beginning to price something the equity footnotes still ignore. The credit market sees the commitments the equity narrative does not.
That is the tell. When the same company's bonds and stock disagree this sharply, the bonds are usually looking at the balance sheet — including the parts that aren't technically on it.
Now stack the trigger conditions. Free cash flow going to zero. Leverage doubling in two quarters. Depreciation not yet fully in the numbers. Financing shifting almost entirely to new debt. And $1.8 trillion in fixed commitments sitting off-balance-sheet, funded through opaque SPV chains that are hard to trace and harder to unwind.
Each of these is survivable alone. The danger is that they arrive together, and they are arriving together.
The failure mode Morgan Stanley points to is specific: if AI commercialization stalls, or enterprise customers shift to cheaper alternatives, the entire financing chain's fragility gets exposed at once. Because the leverage is circular, stress in one node — a vendor, an SPV, a developer — transmits back up the chain in ways no single balance sheet reveals.
That is what "no balance sheet to absorb it" means in practice. The commitments are real. The cash to meet them internally is disappearing. The debt to bridge the gap is being issued into a market whose spreads are already widening.
The market spent this week arguing about whether the visible, financeable number is $1.1 or $1.4 trillion. It is the wrong argument.
The number that detonates is the one in the footnotes — the $1.8 trillion no one reprices, colliding with free cash flow that just went negative for the first time. That collision has started. The tape is only beginning to notice.