The Scale of What's Being Built!
Nvidia alone hit a $5 trillion market cap in October
2025 — higher than the GDP of every country except the US and China. Its value
quadrupled from $1T to $4T in roughly two years.
Total AI spending is expected to surpass $1.6 trillion, with
US mega-cap spending alone projected at $1.1 trillion between 2026 and 2029.
The five hyperscalers (Google, Microsoft, Amazon, Meta,
Oracle) had $241 billion in capital expenditures in 2024, with capex spending
expected to rise to half a trillion dollars in 2026.
JP Morgan estimates that more than $6 trillion in funding
will be required between now and 2030 for AI-related data centers, energy
projects, and the AI supply chain.
The Circular Financing Loop (The Ouroboros)
This is the key thing most people don't understand. It's not
just spending — it's a self-reinforcing loop of fake demand.
Rising valuations justify heavier capex; rising capex
signals explosive future demand; the signal itself reinforces valuations. Round
and round.
The specific deals making experts nervous:
- Nvidia
pumped $100 billion into OpenAI to bankroll data centers, and OpenAI fills
those facilities with Nvidia's chips. As one analyst put it: "The
idea is I'm Nvidia and I want OpenAI to buy more of my chips, so I give
them money to do it."
- Meta's
$27 billion data center financing through Blue Owl was structured so the
debt never shows up on Meta's balance sheet — a special purpose vehicle
arrangement that evokes Enron's collapse in 2001.
- Meta
sold $30 billion of corporate bonds and secured another $30 billion in
off-balance-sheet debt through a joint venture, with companies
increasingly resorting to "creative finance."
Michael Burry — the man from The Big Short who
predicted 2008 — put it bluntly: "True end demand is ridiculously small.
Almost all customers are funded by their dealers."
The Fundamental Revenue Problem
Current AI-related revenues are estimated around $15–20
billion, while annual depreciation costs alone from $400 billion in capex come
to $40 billion — roughly double the revenues. To generate a solid return on
invested capital, AI revenue would need to reach $400–500 billion. And that's
just for one year's investment.
Despite $30–40 billion in enterprise investment into
generative AI, an MIT Media Lab report found that 95% of organizations are
getting zero return.
OpenAI, the supposed crown jewel: committed to spending $1.4
trillion over 8 years building data centers against just $13 billion in
revenue, with that long-term spending funded by debt. Deutsche Bank estimates
OpenAI's losses will total $140 billion between 2024 and 2029, with $74 billion
in operating losses in 2028 alone.
The Hidden Debt Bomb
Risk is increasingly migrating away from tech company
balance sheets and into institutions — utilities, insurers, data center
operators, private credit funds, pensions, and retail vehicles — that do not
see themselves as betting on GPU cycles.
Morgan Stanley puts global spending on data centers between
2025 and 2028 at $3 trillion, half of which is covered by private credit.
There's also a hardware obsolescence problem hiding in the
debt structures: a data center filled with Nvidia H100s in 2024 faces severe
competitive disadvantages against one with Blackwell chips in 2025 and
potential obsolescence with the next architecture — meaning depreciation
schedules are too long, collateral values in default are illusory, and cash
flow assumptions are fragile.
Invert the Problem: What Does the Collapse Actually Cost?
At today's valuations, an equity crash like the early 2000s
would wipe out approximately $33 trillion of value — more than the entire US
GDP. The dot-com crash wiped out $6 trillion and took 7 years for the S&P
500 to recover. That was before AI dominated 80% of market gains.
Harvard economist Jason Furman estimates AI-driven
infrastructure investment accounted for 92% of US GDP growth in the first half
of 2025. That means the broader economy is now structurally dependent on this
spending continuing.
A debt-fueled collapse — the worse scenario — echoes 2008:
in 2008, banks discovered they owned far more US housing risk than their
internal reports suggested. They might soon discover the same about data-center
and digital infrastructure risk — only this time, exposures are spread across
corporate, real estate, infrastructure, fund financing, and alternative credit
books.
The Bailout Question
OpenAI's CFO has already floated the idea openly. OpenAI's
CFO suggested the federal government could get involved by offering a
"guarantee" that could "drop the cost of financing" and
increase the amount of debt the firm could take on. The backlash was fierce,
but the intention was clear.
If the AI bubble pops, the US government will likely turn to
the Federal Reserve to stabilize the economy by injecting huge amounts of cash,
as it did after the 2008 financial crisis. But a new bailout would mean another
significant jump in the national debt and increased wealth inequality, because
taxpayer dollars would be focused on stabilizing a sector in which the
wealthiest individuals will benefit disproportionately.
For comparison: the 2008 banking bailout cost US taxpayers
an estimated $498 billion. Today's big AI firms are worth way more than those
banks, with a combined value exceeding $2 trillion, and they are interconnected
through a complex web of deals worth hundreds of billions of dollars.
Circular funding loops imply the same capital appears
several times on different balance sheets, creating hidden fragility and
misleading investors, creditors, and regulators — exactly what authorities
failed to identify before the 2008 Global Financial Crisis.
The Bottom Line
The optimistic case is that this is like railroads in the
1840s — most investors get burned, but the infrastructure proves genuinely
useful long-term. The pessimistic case is 2008 but bigger, because the debt is
more hidden, the valuations are more extreme, the concentration in the market
is more severe, and 62% of American households now own stock (at all-time
highs). The circular financing, the off-balance-sheet structures, the debt
issued against hardware that depreciates faster than assumed, and the explicit
expectation of a government backstop — these are all classical late-bubble
warning signs. Charlie Munger's worry was exactly right: the incentive to keep
borrowing to chase the next model, the next chip, the next data center is
almost impossible to resist when valuations keep going up. Until they don't.
No comments:
Post a Comment
Thank you!