The accounting trick behind Big Tech's $3 trillion AI bet


Everyone tracks Big Tech's AI capex. Roughly $600 billion over the past year across nine companies.

Big number. Wrong number.

The WSJ went through the footnotes of their SEC filings and found ~$3 trillion in additional commitments that don't appear on any balance sheet — about triple what these companies owe in leases and long-term debt combined. And it's growing faster than capex itself.

Watch the mechanism in 25 seconds:

How do you hide a $3 trillion bill? Two accounting rules.

Rule 1: A lease doesn't hit the balance sheet until rent starts.

Meta's "Hyperion" data center in Louisiana covers the equivalent of 1,700 football fields. Meta designed it, builds it, and will run it. But a joint venture majority-owned by Blue Owl Capital funds owns the campus, and a holding company raised the $27B construction debt via bonds. Meta is just the "tenant" — lease starting in 2029, renewal options up to 20 years, plus a guarantee to make bondholders whole if it walks early.

Meta judges that payout "not probable." So it books nothing.

Industry-wide, not-yet-started leases now total $1.2 trillion — 4x what it was a year ago.

Rule 2: A purchase commitment stays off the books until delivery.

Chips, memory, energy — locked in years ahead with binding contracts. Total: $1.9 trillion. Alphabet alone holds $811B, up from $332B in a single quarter, with energy deals stretching to 2054. Even Alphabet's own disclosures don't explain the jump.

Why this matters to you:

  1. The real AI buildout is roughly 5x the headline capex. Anyone planning around "AI slowdown" narratives is reading the wrong line of the filings.
  2. These bills can't be canceled. If AI revenue shows up, fine. If it doesn't, companies already posting negative free cash flow — Alphabet and Amazon — are locked into decades of payments and more borrowing.
  3. Even Wall Street can't see the full picture. Morgan Stanley's own analysts admit investors increasingly can't assess these companies' true leverage.

Either this is the largest infrastructure bet in history paying off — or the largest bill in history coming due. Both scenarios reshape the tools, pricing, and opportunities you'll work with for the next decade.

Position yourself on the right side of the buildout.

AXDAVID AI

👉 Have questions? Just reply to this email or contact us at aihelp@axdavid.com - we reply within 24 hours.


Unsubscribe · Preferences

AI & Automation for Business Growth

Turn AI into revenue. Discover powerful tools, prompts, and strategies to grow your business faster, save time, and increase your income - every week.

Read more from AI & Automation for Business Growth
Meta Settles for up to $16.68 billion

Meta's Settlement Just Changed How Teens Use Instagram A settlement filed today in Oakland federal court will change how teenagers use Instagram and Facebook. Meta agreed to resolve claims from 29 states that it designed its apps to keep kids hooked, misled families about safety, and collected children's data without parental consent. It settled eight days into trial, before any verdict. Meta denies wrongdoing. The number is messier than the headline. Court papers put the payment at up to...

Meta released Muse Glimmer, a new open-weight AI model that's far smaller than rival frontier models. It's built for agentic tasks and runs on a Mac or PC with a single graphics card, targeting demand for on-device AI. The launch follows Meta's formation of an expensive superintelligence team last year to push back into the top tier of the AI race. Alongside the release, Zuckerberg called for lower US barriers on open-source AI to compete with China, arguing foreign labs hold advantages...

New AI rules went live in Europe on August 2. The short version: Chatbots now have to tell you they're AI, deepfakes have to be labeled, and fines run to €15 million or 3% of global turnover. The part most coverage missed: the big high-risk obligations got pushed to December 2, 2027. What actually landed now is the disclosure layer—the rules about telling people when content is synthetic. Why this matters to you even outside Europe: if your content reaches European audiences, these rules...