🚨 EVERYONE IS WATCHING WHO OWNS THE GPUs. BUT WHO OWNS THE RISK BEHIND THEM?
First it was GPUs.
Then memory.
Then networking.
Then power.
Then data centres.
But I think another layer is starting to matter.
The financial infrastructure underneath all of it. 👀
Because an AI factory doesn’t just need chips, electricity and land.
It needs someone willing to finance the factory.
And the deeper I looked into how this buildout is being funded, the more interesting the picture became.
This isn’t a prediction of an AI credit crisis.
It’s a different question:
As trillions of dollars move into AI infrastructure, where is the financial risk actually going?
🏗️ THE AI BOOM IS BECOMING A CREDIT STORY
S&P Global says AI infrastructure financing is increasingly spreading across multiple channels:
Bank lending.
Private credit.
Asset-based finance.
CMBS.
ABS.
Corporate debt.
Leases.
Joint ventures.
SPVs.
Guarantees.
On the surface, that sounds like diversification.
More lenders. More structures. More pools of capital.
But S&P identifies something much more interesting.
Apparently different investments can ultimately trace their economic exposure back to the same small group of hyperscalers.
And that’s where this gets interesting.
☢️ What if AI infrastructure is diversifying its financing faster than it is diversifying its underlying economic exposure?
🕸️ FIVE INVESTMENTS. ONE CUSTOMER?
Imagine five different investments.
A data-centre loan.
A private-credit fund.
A CMBS security.
An asset-backed security.
A corporate bond.
Different structures.
Different portfolios.
Different investors.
But somewhere underneath them could sit infrastructure ultimately dependent on demand from the same handful of giant technology companies.
Suddenly the question isn’t simply:
“How diversified is the financing?”
It’s:
“How diversified is the cash flow underneath it?” 👀
That distinction could become increasingly important as the AI buildout gets larger.
💰 THE NUMBERS ARE GETTING ENORMOUS
S&P Global Ratings projects combined hyperscaler capex will exceed $1.3 TRILLION by 2027.
It also expects the six hyperscalers in its analysis to generate negative free operating cash flow during 2026 and 2027, with recovery projected later.
Importantly, that doesn’t mean these companies are running out of money.
Most remain enormously profitable businesses with powerful balance sheets.
But as spending grows, S&P says financing through debt, equity, leases, JVs, SPVs and guarantees is becoming increasingly important.
The machine isn’t necessarily weakening.
The way the machine is being funded is changing.
👀 THEN I FOUND THE GUARANTEES
This is where I went further down the rabbit hole.
The Financial Times reports Big Tech companies have deployed up to $300 BILLION of residual-value guarantees over the past year supporting financing for AI infrastructure including data centres and chips.
One structure can involve an SPV owning the infrastructure while a technology company guarantees some future value of those assets.
That can make projects easier to finance without simply placing the entire funding requirement into conventional corporate debt.
Again, that doesn’t automatically make it dangerous.
But it does make the financial map more complicated.
Because the AI factory now has two stacks.
⚡ THE PHYSICAL STACK
GPU
↓
Memory
↓
Networking
↓
Data centre
↓
Power
💸 THE FINANCIAL STACK
Customer contract
↓
Developer / SPV
↓
Bank / private credit
↓
Debt / lease / guarantee
↓
ABS / CMBS / bond
↓
Investor
The first stack builds intelligence.
The second stack finances it.
⚠️ WHAT HAPPENS WHEN THE DEBT GETS HARDER TO DISTRIBUTE?
Oracle’s Project Jupiter gives us an interesting real-world example.
Around $18 BILLION of loans connected to the New Mexico data-centre project were recently quoted around 89–91 cents on the dollar, according to Reuters reporting on the Financial Times.
Banks reportedly ended up holding more exposure than anticipated as efforts to distribute the debt became more difficult.
There are project-specific issues here, so one project shouldn’t be extrapolated across the entire AI industry.
But it demonstrates something important.
Building the infrastructure is only half the equation.
The financial system also needs to keep absorbing the capital structure behind it.
🏦 AND THE CAPITAL DEMANDS KEEP GROWING
CoreWeave announced another $3 BILLION convertible debt offering as it continues financing its AI infrastructure expansion.
Then there’s Nscale.
Its IPO filing showed first-half 2026 revenue of $140.6M, alongside a $1.02B net loss.
It also reported more than $103B of contracted revenue.
Huge number.
But here’s the part that caught my attention:
52% of current revenue came from one customer.
That doesn’t invalidate the growth story.
It highlights another distinction:
☢️ CONTRACTED DEMAND ≠ DIVERSIFIED DEMAND
🐂 THIS DOESN’T MEAN THE AI BOOM IS BROKEN
This is where I think the bear narrative can go too far.
Today’s AI infrastructure cycle isn’t simply the dot-com bubble wearing an Nvidia hoodie.
The largest hyperscalers generate enormous amounts of cash and possess some of the strongest balance sheets in corporate America.
S&P specifically says this cycle has not yet developed the financing characteristics that made the dot-com infrastructure buildout fragile.
The risk worth watching is what happens if investment increasingly transitions from cash-funded to debt-funded before the returns validate the spending.
That’s a very different argument from:
“AI spending is a bubble.”
The question I’m interested in is:
How does the funding mix change as AI spending keeps growing?
🔄 THE BOTTLENECK KEEPS MOVING
This AI cycle has already taught us something.
Solving one bottleneck often creates another.
GPU scarcity.
↓
Memory and packaging.
↓
Networking.
↓
Electricity.
↓
Grid connections.
↓
Capital.
↓
Risk absorption?
Because eventually the AI boom doesn’t just need investors willing to buy technology stocks.
It needs banks willing to lend.
Private-credit funds willing to finance.
Bond investors willing to absorb issuance.
Structured-credit markets willing to package exposure.
Institutions willing to own it.
And ultimately…
customers generating enough economic value from AI to support the entire machine.
🔎 WHAT I’M WATCHING
👀 Does hyperscaler cash generation keep pace with capex?
👀 Does AI monetisation catch up with infrastructure spending?
👀 Does customer concentration appear across supposedly different pools of credit?
👀 Do guarantees, SPVs and other financing structures keep expanding?
👀 Does data-centre debt remain easy for banks to distribute?
👀 And most importantly, does the return generated by AI justify the enormous amount of capital being deployed to build it?
Because perhaps the next AI bottleneck isn’t compute.
Maybe it isn’t memory.
Maybe it isn’t electricity.
Maybe it isn’t even the availability of capital.
☢️ THE QUESTION NOBODY CAN ANSWER YET
What if the next bottleneck is the market’s willingness to keep absorbing the risk?
Everyone is watching who owns the GPUs.
I’m starting to wonder who owns everything underneath them. 👀
🐯📊
$NVDA $MSFT $AMZN $GOOGL $META $ORCL $CRWV
Personal market analysis only. Not financial advice. Always do your own research.
Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.
