The Take — The $3T isn't Enron. It's a no-exit clause
The $3 trillion is not a secret, and it is not a fraud. I think it is something more dangerous: a stack of contracts that already spent the next several years of the AI boom, so the industry no longer has a clean way to stop.
This morning's brief walked through the Wall Street Journal's tally — Big Tech's hidden $3T in AI obligations dwarf reported capex. Nine companies. Roughly $3 trillion in AI-related commitments that do not sit on the balance sheet. About $600 billion in reported capital expenditure. The gap is not a rounding error. It is the real size of the bet.
The argument: optionality is already gone
The number is two piles glued together. One is purchase commitments — chips, servers, construction, power — somewhere between $1.52 trillion and $1.9 trillion, depending on how you read the footnotes. The other is leases that have been signed and have not started: roughly $900 billion to $1.2 trillion of data-center space the tenants have locked in but have not yet occupied, so the liability has not rolled onto the books.
Those are not "hidden" in the way a special-purpose vehicle is hidden. They are in the quarterly filing. Anyone who wants to find Alphabet's $811 billion in purchase and contractual obligations as of June 30 can do it. Anyone who wants to notice that the same line was $322 billion three months earlier can do that too. A 152 percent jump in a single quarter is not a company hedging. It is a company prepaying the future.
Meta's Hyperion campus in Louisiana is the cleanest picture of the structure. The company formed a joint venture with Blue Owl Capital, parked the project in a variable-interest entity, and left about $27 billion of project debt off its own balance sheet. Meta's disclosed maximum exposure is $46 billion. Meta is also the sole tenant. Ernst & Young told investors that judging whether Meta controls the entity was "challenging." I believe them. I also think the economic question is simpler than the accounting one: if the only customer walks, who eats the campus?
Nvidia already flinched. Last week it cut a planned $250 billion backstop on OpenAI's Ohio campus to just under $120 billion after investors pushed back — we covered that pullback in Nvidia halves OpenAI's $250B data center backstop to $120B. That is the tell. The boom's biggest winner is being told, by its own shareholders, that unlimited appetite is no longer a strategy. The commitments did not shrink with the headline. The first five gigawatts still get built.
Then there is the circular money. Semiconductor firms have sat in startup rounds collectively valued at more than $250 billion this year. Nvidia alone has shown up in a record 59 financings, including as one of eight leads on OpenAI's $122 billion March round, and it wrote a $5 billion corporate check to Safe Superintelligence in July. When the shovel-seller also bankrolls the mine, the purchase commitments and the customer pipeline are the same loop. That is not a market discovering a price. It is a closed circuit confirming its own demand.
The physical world has already priced the lock-in. We wrote earlier this week that the $1 trillion AI build-out hits a wall cash can't fix — power, labor, permits. The $3 trillion is what happens when companies refuse to accept that wall. They sign the leases and the chip contracts anyway, then treat the signatures as if they were still optional.
The counter-case, fairly stated
The other side is not stupid, and the Enron comparison is mostly a vibe. Enron hid the vehicles. These footnotes are public. US lease and consolidation rules got rewritten after 2001 precisely so that a sophisticated reader can reconstruct the exposure. The companies are not empty shells. They print cash. A purchase commitment is not a bond. A lease that has not commenced is not defaulted debt. If demand holds, these contracts are how you get capacity in a scarce market instead of waiting in a utility queue. Michael Burry pointing at the footnotes does not make the footnotes a crime.
There is also a real accounting defense. Purchase obligations generally hit the books when the goods arrive. Unstarted leases roll on when the term begins. That is the standard, not a loophole invented for GPUs. Calling every off-balance-sheet line "hidden debt" collapses three different instruments into a scare headline. And if you believe AI revenue will grow into the spend — Amazon has said AWS is on a path to a $1 trillion business — then locking supply now is the conservative move, not the reckless one.
Why the take still holds
Because the scandal is not concealment. It is irreversibility.
A disclosed trap is still a trap. Once you have signed $811 billion of purchase obligations, "we will slow capex next quarter" is a press release, not a plan. The cash leaves either way. Once Meta is the only tenant on a campus it does not consolidate, the project debt can sit elsewhere and the utilization risk still sits with Meta. Once Nvidia is simultaneously supplier, investor, and backstop, a demand miss does not stay in one column. It ricochets around the loop.
Nikkei's July cut of the same story put the off-balance-sheet pile at $1.65 trillion across five companies. Six weeks later the Journal's nine-company version is about $3 trillion. That is not a one-time catch-up. That is the number getting worse while everyone can see it. The vehicles look familiar, and that is worth saying out loud. Enron's crime was hiding the special-purpose vehicles. The AI boom's problem is that the vehicles, the leases, and the chip contracts are doing exactly what they were designed to do: convert a flexible investment program into a multi-year obligation that earnings can still pretend is optional.
What would change my mind
Three things, honestly. First: a couple of quarters where the footnotes shrink — purchase commitments rolling off faster than new ones are signed, unstarted leases converting to used capacity that is actually full. Second: AI product revenue, not circular chip-and-cloud deals, covering a clear majority of the cash these contracts will demand. Third: a credible, standardized way to put unstarted leases and multi-year chip buys into the leverage ratios everyone already uses, so "off balance sheet" stops being a valuation advantage. If those show up, this is just how a scarce industrial build gets financed. Until they do, I will treat the $3 trillion as what it is: the boom spending a future it can no longer unspend.
If the footnotes already spent the next five years, what would actually make Big Tech slow the build-out? Tell us in the comments.
