In April 1801, Horatio Nelson found himself in trouble off Copenhagen.
For Nelson, this was hardly unusual.
His ships were under heavy Danish fire, the outcome of the battle was far from settled, and his superior, Admiral Sir Hyde Parker, concluded that it was time to break off the action. He ordered the signal to withdraw.
Nelson saw it.
Then he stopped seeing it.
He raised his telescope to his blind eye and declared that he could not, for the life of him, make out any signal.
Nelson kept fighting.
And won.
It is one of those stories that has survived for more than two centuries because it contains everything a good story needs: danger, disobedience, a touch of madness and — rather helpfully for its protagonist — a happy ending.
More interesting than the telescope, however, is something Nelson had written only a few days earlier:
«The measure may be thought bold, but I am of the opinion the boldest are the safest.»
The measure might appear bold, he wrote, but in his view the boldest measures were the safest.
It is easy to remember boldest.
We should remember safest.
We will come back to that.
Wall Street Has a Problem
Two hundred and twenty-five years later, no one is standing off Copenhagen with a telescope.
Wall Street does, however, have a problem.
Artificial intelligence needs money.
A great deal of it.
Google, Microsoft, Meta, Amazon and others are building data centres on a scale that makes even billions of dollars begin to sound rather small. Those data centres need Nvidia processors, networking equipment, optical connections and cooling systems. And somehow, they also need enough electricity.
America’s largest technology companies have now taken on enormous long-term commitments for data centres and computing capacity. A significant portion of those commitments does not yet appear as conventional debt on their balance sheets.
At the same time, capital expenditure continues to rise.
At some point, even for companies that generate sums of money once associated with small sovereign states, a reasonable question arises:
Who is going to pay for all of this?
For Wall Street, however, a financing constraint is rarely just a problem.
It is a business opportunity.
Take One Data Centre
The principle is elegant.
A technology company could build and finance a data centre itself.
Or it can sign a long-term lease and let somebody else build it.
The owner places the data centre in a special-purpose vehicle. Future lease payments generate a cash flow. Bonds are issued against that cash flow. Insurers, pension funds, private-credit funds and other investors buy them.
The technology company gets its computing capacity.
The developer gets its capital back and can build the next data centre.
The investor receives a long-term return.
Wall Street collects its fees.
Everyone gets what they want.
Financial history suggests that this is precisely where the interesting part begins.
Because the numbers are growing quickly. Data centres are increasingly being securitised, private credit is financing additional projects, and special-purpose vehicles allow part of the capital requirement to be met outside traditional corporate financing.
Nvidia, meanwhile, is working with some of the world’s largest capital providers on structures that could mobilise hundreds of billions of dollars for AI infrastructure.
So the question is no longer merely:
How much can the technology companies invest themselves?
Increasingly, it is:
How much capital can the financial system mobilise on their behalf?
That is a much larger number.
And since the US Securities and Exchange Commission has given certain data-centre securitisations more regulatory room, one comparison almost makes itself.
ABS.
Special-purpose vehicles.
Long-term obligations.
Wall Street.
2008.
At this point, one could already sound the fire alarm.
But that may be the wrong lesson to draw from history.
For the Next Stage, We Need to Go to London
Much of the history of the modern financial system is, after all, the history of one very simple question:
How can existing capital be made to finance just a little more?
One of the great answers emerged in London in the late 1950s and 1960s.
Banks made a remarkable discovery about US dollars:
They did not necessarily have to remain in America.
Dollar deposits outside the United States became the foundation of the Eurodollar market. London developed into the centre of a vast offshore dollar system.
Dollars could be taken in, lent, re-lent and moved across borders without every stage having to take place within the US banking system.
It was one of the most consequential financial innovations of the past century.
Syndicated loans followed, then the recycling of petrodollars, lending to emerging markets, Brady bonds, securitisation and derivatives.
Japanese capital found its way into Asia. European banks financed emerging markets and later built sizeable exposures in places such as Turkey.
The principle was wonderfully simple:
Cheap money looks for yield.
And as long as everyone involved is making money, there is remarkably little social pressure to make the principle more complicated.
Then Came Financial Alchemy
By the turn of the millennium, Wall Street had taken the idea a step further.
Why merely move capital around when you can manufacture new collateral?
Americans and Britons bought houses.
They took out mortgages to pay for them.
Mortgages created payment streams stretching over decades.
Thousands of them were pooled together, the cash flows were divided into tranches, and riskier tranches were placed beneath supposedly safer ones.
One small problem remained.
A mortgage to someone with irregular income and dubious creditworthiness does not necessarily look much like a US Treasury bond.
The financial industry found a solution.
AAA.
Models showed that not all borrowers would default at the same time. Junior tranches would absorb the first losses. The senior tranches appeared extraordinarily safe.
With the right rating, a collection of private mortgages had been transformed into something that increasingly displayed the characteristics of high-grade collateral within the financial system.
It could be traded.
It could be financed.
And, above all, it could be used as collateral.
Now things became genuinely interesting.
The Wonderful World of Repo
A repo transaction is, at heart, remarkably uneventful.
I own a US Treasury bond worth $100.
You lend me $98 or $99 against it for a short period and take the bond as collateral.
Later, I repay you and get my bond back.
So far, so dull.
It becomes more interesting because you do not necessarily place the collateral in a vault and patiently wait for me to return.
You can use it yourself.
You pass the bond on, obtain financing against it, and under certain conditions the next holder can do the same.
A repo becomes a repo chain.
The same security can therefore support several successive financing transactions. Not because more Treasury bonds suddenly exist, but because temporary possession of the collateral is passed through the financial system.
And tomorrow?
Tomorrow, the financing is rolled over.
As long as everyone trusts that the collateral will continue to be accepted, long-term assets can generate a remarkably fluid stream of short-term liquidity.
The collateral is accepted.
Haircuts remain small.
Financing is plentiful.
New positions are built.
Rising prices, in turn, make the collateral appear even more solid.
The result is an exceptionally pleasant feedback loop:
good collateral → more financing → more demand → higher prices → greater confidence → still more financing.
And everyone is relying on two things.
That tomorrow the collateral will be worth roughly what it is worth today.
And that tomorrow someone will once again be willing to lend against it.
With US Treasury bonds, these are two fairly sensible assumptions.
Before 2008, the financial system began applying the same assumptions to something else.
To private securities that, thanks to financial engineering and a AAA stamp, looked almost as safe.
Almost turned out to be a very expensive word.
When Someone Wants Their Money Back
Eventually, one of the participants takes a closer look at the security they have accepted as collateral.
It says AAA on the label.
Very reassuring.
But what is actually inside it?
Mortgages.
Still reassuring. People do, after all, pay their mortgages.
And who are these people?
Now things become slightly less reassuring.
Yesterday, the transaction was simple. Against $100 of supposedly top-quality collateral, a lender might have advanced $98.
Today, the lender is willing to advance only $90.
Tomorrow, perhaps $80.
And eventually comes the sentence nobody in a long financing chain particularly likes to hear:
«I want my money back.»
At first, this creates a problem for the immediate counterparty.
Unfortunately, it rarely remains theirs alone.
They may themselves have borrowed short-term and reused the collateral further along the financing chain.
Now they must raise money elsewhere, post additional collateral or sell assets.
So they turn to the next participant.
Who turns to the next.
And suddenly a whole series of very well-paid people discover at the same time that they, too, would quite like their money back.
On the way up, the chain created liquidity.
On the way down, it creates demands for cash.
That would be unpleasant enough if everyone knew what the collateral was worth.
But suddenly, nobody does.
If a repo chain involving US Treasuries breaks down, somebody is left holding — perhaps rather unhappily — a US Treasury bond.
At least that is exactly what it said on the label.
With a mortgage-backed security, however, somebody was left holding a complicated claim on the payments of thousands of homeowners whose creditworthiness could prove considerably less impressive than the three letters printed on the wrapper.
And with that, the crucial question changed.
Yesterday it was:
How much will I lend you against this collateral?
Today:
Hang on — if you cannot repay me, does this thing become mine?
And immediately after that:
What is this thing actually worth?
Those were two exceptionally bad questions to start asking at the same time.
Because in a system built on the assumption that someone will refinance you tomorrow, the answer “I have no idea” can become remarkably expensive.
The houses had not vanished overnight.
Neither had the mortgages.
What disappeared was something far more elusive.
The certainty that AAA really meant AAA.
One Important Difference
Before we return to 2026, one important difference should not be overlooked.
The old machine had a powerful ally.
Falling interest rates.
Lower financing costs made leverage cheaper. Rising bond prices improved the value of much of the collateral. Refinancing became easier. Rising asset prices, in turn, created additional collateral.
Repo likes many things.
Rising interest rates are not generally among them.
Today, the new financing architecture for AI is developing under almost the opposite conditions.
The demand for capital is rising at a time when long-term interest rates are high and the price of money matters again.
That creates an uncomfortable combination.
Financing becomes more expensive while higher discount rates simultaneously reduce the present value of long-duration cash flows.
And if a lender begins to doubt the collateral as well, there are two levers to pull.
Charge a higher interest rate.
And demand a larger haircut.
For someone using leverage, this is roughly as pleasant as having the rent rise just as the salary falls.
That is precisely what makes the current development so remarkable.
Wall Street is not building this new financing machine in a world in which money keeps getting cheaper.
It is building it in a world in which money has become more expensive.
Perhaps that will impose an early limit on the story.
Or perhaps securitisation, private credit, special-purpose vehicles and new pools of investors will mobilise enough additional capital that higher rates merely slow the boom rather than stop it.
Then we would have something particularly interesting:
A liquidity machine being built into a headwind.
Whether it works is likely to be one of the questions that occupies us in the years ahead.
Back to 2026
Which brings us back to artificial intelligence.
And here is the most important point:
We are not in 2008.
Google is not a subprime homebuyer in Nevada.
Neither is Microsoft.
Demand for computing power is real. Data centres are actually being built. Nvidia is actually shipping processors. Networking and optical-component suppliers are actually reporting extraordinary demand.
The companies using a large share of this infrastructure are among the strongest corporations in the world financially.
And so far there is little evidence that data-centre bonds have already assumed the role of lightly margined, repeatedly re-used repo collateral that private mortgage-backed securities partly played before the financial crisis.
That is a crucial distinction.
Securitisation itself is not the problem.
What matters is what happens to the securitised asset afterwards.
And that is why it may be a mistake to see the first resemblance to 2008 and immediately assume that we know how the story ends.
Financial booms have an inconvenient habit.
Before the collapse, something else often comes first.
First Comes the Party
Suppose insurers, pension funds, private-credit funds and other institutional investors discover AI infrastructure as an attractive asset class.
The tenants are called Google, Microsoft, Meta or Amazon.
The contracts run for many years.
The yields are attractive.
Demand for the securities rises.
Financing costs therefore fall — or at least rise less than they would without that new demand.
In today’s interest-rate environment, that distinction matters.
Projects that were too expensive yesterday can be financed tomorrow.
More data centres are built.
More Nvidia processors are ordered.
More networking equipment is required.
More optical connections.
More electricity.
More cooling.
More of almost everything.
Suppliers report higher revenues and larger order books.
Their share prices rise.
Rising share prices, in turn, reinforce the impression that AI demand is even greater than previously thought.
Capital providers become more confident.
More money flows in.
Eventually the chain looks like this:
more financing → more AI investment → higher revenues → higher valuations → greater confidence → more financing.
That is not yet a bubble.
At first, it is simply a boom.
And booms can be exceptionally pleasant.
The difficulty is that while they are under way, they rarely carry a sign telling you exactly when one has turned into the other.
The Most Important Question
Take Nvidia.
A company needs additional computing capacity and therefore orders $10 billion of Nvidia systems.
That is demand.
Now suppose an AI company receives $10 billion from a financing consortium.
It uses the money to buy Nvidia systems.
Nvidia books $10 billion of revenue.
Profits rise.
The share price rises.
Investors see this as further evidence of the strength of AI demand.
The financial system becomes still more willing to lend against AI infrastructure.
The next buyer receives financing.
They order the next batch of processors.
That, too, may be entirely healthy.
But at some point a question has to be asked:
Who ultimately needs the computing power — and who financed the buyer?
Because one day, that may be where the boundary lies.
On one side, capital finances genuine demand.
On the other, abundant capital begins to create additional demand.
The distinction is easy to describe.
It is much harder to identify in real time.
When We Would Really Start to Worry
Data-centre ABS alone are no reason for us to panic.
Things become more interesting when they turn into highly rated securities against which dealers are willing to lend freely at ever-smaller haircuts.
When leveraged investors buy them.
When the collateral is passed on.
When repo chains begin to form.
When derivatives multiply the economic exposure.
And when progressively weaker projects are financed because the market is desperate for more paper.
At that point, something fundamental has changed.
AI infrastructure would no longer merely consume capital.
It would produce collateral against which the financial system could create additional liquidity.
Then our question would no longer be simply whether Google can pay its lease.
We would want to know what the person at the end of a financing chain actually owns when, one morning, somebody says:
«I want my money back.»
Perhaps it is a state-of-the-art data centre with a first-class tenant in an excellent location.
Perhaps it is a highly specialised industrial facility whose processors, power density, cooling and networking technology have already become obsolete.
The building still stands in both cases.
Its economic value does not therefore have to be the same.
A US Treasury bond remains a US Treasury bond.
A data centre is not a Treasury.
The Telescope
And with that, we are back off Copenhagen.
Nelson’s famous line is easily read as an exhortation to be bold.
That is probably the less interesting half of it.
He did not write that the boldest measures were the most exciting.
Nor the most heroic.
Not even the most profitable.
He wrote that they were the safest.
Nelson believed he understood the situation well enough to see that the apparently cautious retreat posed the greater risk.
Investors should be more cautious about reaching such conclusions than admirals.
The explosive rise in AI investment, new debt, special-purpose vehicles, private credit and securitisation are signals we take seriously.
But they do not yet tell us which decision is the safest.
Perhaps today’s interest-rate environment will prevent the new financing machine from ever developing the dynamics we have seen in earlier credit cycles.
Or perhaps the financial system is building a machine powerful enough to push into that rate headwind and make the AI investment cycle far longer and stronger than the technology companies’ balance sheets alone would suggest.
Then perhaps euphoria comes first.
And, eventually, perhaps financial alchemy.
And perhaps one morning someone returns with the uncomfortable question that made 2008 so costly:
What is the collateral actually worth?
We do not need to know today how this story ends.
We need to recognise how it is developing.
Is money still following demand?
Or is demand beginning to follow the money?
How are the new securities being valued?
What haircuts is the market demanding?
Who is financing the buyers?
Is the collateral being re-used?
And who holds it at the end of the chain?
Nelson had one advantage off Copenhagen: he could see the battlefield in front of him.
We do not have that luxury.
All we can do is look very carefully.
Because the most important word in his famous sentence was never boldest.
It was safest.