←2026-09-04 · Herbert Simon
Analysis
Think the cash these tech giants are burning on AI is just that CapEx figure on the first page of their earnings reports?
You’ve got to be kidding. That’s just the tip of the iceberg.
Over the past few years, Meta, Amazon, Microsoft, and Alphabet have dropped trillions on AI. But the off-balance-sheet commitments tucked away in the footnotes are even more insane.
Shocked?
Alphabet is sitting on $\$798\text$ in future investment commitments. Meta is tied up to the tune of $\$696\text$—roughly 10 times its net income over the trailing twelve months (TTM).
If AI fails to monetize, these guys are heading straight for a meltdown.
Over the summer, the Wall Street Journal ran a piece exposing the reality: actual AI spending is far larger than the market realizes, easily blowing past the multi-trillion-dollar mark. When I first saw that conclusion, I got chills down my spine. How can these mega-caps sit in the spotlight every single day without most investors catching this hidden risk in the fine print?
As for the financial reports of A-share companies? Never mind, I didn't say anything.
The Point of No Return
A familiar narrative keeps ringing in the market’s ear: in the AI land grab, the danger of underinvesting vastly outweighs the risk of overinvesting.
Meta’s CEO said the exact same thing publicly. His words were basically: Even if we end up squandering hundreds of billions of dollars, that’s obviously a huge misfortune. But to me, the far greater risk is underinvesting. If super-intelligence lands in three years instead of five, we’d completely miss the boat on a core technology that promises to create the most products, innovations, and value in human history.
In short: throwing away hundreds of billions is a blunder, but it pales in comparison to falling behind on a generational tech shift.
Talk about thick-skinned. Meta pulled in $\$68\text$ in net income over the TTM; misallocating hundreds of billions is playing with fire. But that’s a topic for another day—let's set it aside for now.
Over the last two years, the tech titans have doubled down on AI CapEx. This year alone, combined projections hit a staggering $\$600\text$.
People will argue, "What's the big deal with $\$600\text$? These balance sheets are stacked, and they're still growing like small-cap hyper-growth stocks." (Honestly, hearing that trillion-dollar giants are still compounding at $30\%$ to $40\%$ a year sounds downright surreal these days. But hey, I'm getting ahead of myself.)
There's always a flip side: the balance sheets are starting to crack.
The Cash Flow Reality Check
With CapEx pinned at all-time highs, Free Cash Flow (FCF) for these behemoths has effectively flatlined in recent years.
In Q2 2026, both Alphabet and Amazon saw their FCF turn negative—they are literally burning through their own liquidity.
When operating cash flows can't cover project spending, companies typically have two choices: issue debt or raise equity.
Except for Microsoft, the rest are ramping up corporate debt issuance while scaling back or outright pausing share buybacks that ran for over a decade. Alphabet took it to the extreme, opting for equity financing—the single most expensive capital-raising tool and usually a last resort for cash-strapped startups. For a mammoth like Google, this move looks completely out of line.
So I've got one simple question: If you guys are so bulletproof, why are you issuing debt and tapping equity markets?
The Anatomy of Off-Balance-Sheet Commitments
Everyone naively assumes all AI spending is sitting cleanly under CapEx, meaning if returns underperform, they can just pull back anytime.
Read the footnotes of their 10-Q filings, though, and you'll find a completely different reality. The total investment footprint is vastly larger than what's disclosed on the face of the balance sheet. Worse yet, huge sums are already locked into multi-year—and in some cases, decade-long—binding performance obligations.
Does this count as accounting fraud?
Hardly. These giants are shrewd operator-types—they follow GAAP/IFRS to the letter. This money falls under off-balance-sheet commitments. The logic? Neither party has fulfilled the contract yet, so it doesn't hit the balance sheet. These commitments break down into two main categories:
Undrawn/Uncommenced Leases: Signed contracts where the lease term hasn't officially started. These cover servers, real estate, AI hardware, and data centers. The classic setup: the data center is still under construction and not yet operational.
Unconditional Purchase Obligations: Legally binding purchase agreements to buy specific goods or services. For instance, Meta locking in Nvidia GPUs at set prices, or signing multi-decade power purchase agreements (PPAs) with utilities to secure data center electricity.
Digging through the 10-Qs line by line, I compiled these off-balance-sheet investment figures:
Alphabet: Undrawn Leases: $\$91\text$; Unconditional Purchase Obligations: $\$707\text$. (The Wall Street Journal reported $\$811\text$; I couldn't trace the origin of that remaining $\$104\text$.)
Amazon: Undrawn Leases: $\$131.21\text$; Unconditional Purchase Obligations: $\$130.06\text$.
Microsoft: Undrawn Leases: $\$329.1\text$; Unconditional Purchase Obligations: $\$229\text$.
Meta: Undrawn Leases: $\$347\text$; Unconditional Purchase Obligations: $\$349\text$.
Decoding the Numbers
Alphabet faces the heaviest downstream capital pressure, with $\$798\text$ in total commitments—roughly $3.27\times$ its TTM net income. The vast majority must be fulfilled before 2030. In other words, over the next four years, Alphabet's annual CapEx will realistically struggle to stay below $\$150\text$.
Amazon carries the lightest off-balance-sheet burden at $\$261\text$, or $1.93\times$ its TTM net income.
Microsoft holds $\$558\text$ in total commitments, representing $4.19\times$ its TTM net income. Higher than Alphabet in absolute terms, but nowhere near Meta's leverage.
Meta takes the crown for the most aggressive off-balance-sheet leverage: $\$696\text$ in total commitments, reaching a staggering $10.23\times$ its TTM net income.
Looking at these figures, it's easy to see why Alphabet went the equity financing route and why Meta is contemplating following suit. Alphabet faces the highest absolute dollar pressure; Meta has stretched its leverage to the absolute limit relative to its earnings power.
The Counterparty Risk
While signing these multi-hundred-billion-dollar investment commitments, the giants are also sitting on hundreds of billions in Remaining Performance Obligations (RPO)—essentially, revenue backlogs. They are burning cash at a crazy rate, but market demand for their own services is also sitting at historic highs.
It’s like two desperate guys panicking for different reasons.
Alphabet: RPO stands at $\$519.5\text$, with $50\%$ scheduled for delivery over the next two years.
Microsoft: RPO stands at $\$678\text$, with $30\%$ delivering within the next 12 months.
Amazon: RPO stands at $\$496\text$, with a weighted average contract duration of 6.4 years.
Meta: Easily the weakest link here—I couldn't find a corresponding RPO figure. The root cause is its business model: Meta lacks a core cloud business (though that's coming soon) and relies primarily on ad revenues. Advertisers simply don't lock in multi-year ad spend upfront. Once again, Meta stands out as the highest-risk player in the group.
A critical point must be made here: a significant chunk of these backlogs comes from players like OpenAI and Anthropic. Both entities are bleeding cash and heavily reliant on venture equity funding. Whether they can actually honor these long-term contracts remains a massive wild card worth an entire article of its own. If OpenAI or Anthropic hit a funding wall, these tech giants will take a direct hit to their top-line backlogs.
It's a trillion-dollar stress test—how could they not be sweating?
This isn't a call to blindly short these stocks; I'm not like those clickbait permabears who've been predicting the collapse of the US economy for 20 years. My goal is simply to help investors see the true scale of these tech investments. These numbers won't show up on the flashy cover slide of an investor pitch deck—they hide in the financial footnotes. Footnotes usually look like low-priority noise, but this time around, they are everything.
When evaluating the Mag 7's AI push, you can't just look at the face of the balance sheet and current CapEx. That's only a fraction of what they've actually committed to spend. If AI monetization falls short of expectations, these massive obligations can't just be unwound with a mouse click. Taking Meta as an example, fulfilling its total AI commitments would consume ten full years of its current TTM net income. One misstep, and you're looking at a catastrophic wreck.
Of course, we all believe AI represents a generational paradigm shift. But rational investors need to price in the downside risk of AI underdelivering. No investment carries zero risk, and no one's printing money out of thin air.
So when Mark Zuckerberg casually brushes off "hundreds of billions in overinvestment," it takes some serious audacity. Hey Zuck, just remember—back home, those revenue backlogs require a VAT invoice before you can count them.
The Magnificent Seven have marched down the mountain, and there is no turning back.
At this point, keeping the hype train running isn't just strategy—it's a matter of survival.