When Admiral Horatio Nelson defeated the Franco-Spanish fleet at Trafalgar in 1805, the Royal Navy stood behind his success. Behind the Navy stood shipyards, ports, raw materials, supply chains and industrial capacity.
And behind all of that stood something more fundamental still: a financial system capable of mobilising extraordinary amounts of capital. Credit multiplied the power of that capital — ships were built, fleets equipped and losses replaced.
Behind Nelson stood the Navy. Behind the Navy stood the industrial infrastructure. And behind that infrastructure stood capital.
This combination of technology, industrial capacity and finance was not merely an economic strength. It was the foundation of British power.
More than two centuries later, we are watching something that bears a structural resemblance.
The Machine Behind the AI Fleet
The build-out of artificial intelligence began with semiconductors and has now become a vast industrial complex: memory, optical networks, data centres, power plants and turbines.
Behind all of it stands capital.
NVIDIA, together with BlackRock, Blackstone, Brookfield, Apollo, Goldman Sachs and KKR, has announced plans to mobilise more than $500 billion of third-party capital. BlackRock’s Larry Fink went even further on CNBC, speaking of investment needs measured in trillions of dollars.
The scale cannot be explained by the commercial potential of artificial intelligence alone. For the United States, compute has become a strategic resource. Whoever controls the most capable models, the largest data centres, sufficient energy supply and the associated industrial supply chains will shape a significant part of the next technological order.
The build-out of US AI infrastructure is therefore both an economic project and a means of preserving America’s technological lead — particularly relative to China.
The crucial question is no longer simply where the first $500 billion will come from.
It is:
How much additional investment capacity can the US financial system create from it?
When Compute Becomes an Asset Class
The first step is to stop viewing compute merely as capital expenditure by individual technology companies. Data centres and AI factories are real assets that, when highly utilised, can generate predictable cash flows capable of supporting debt.
For this to become a genuine asset class, however, standardisation is essential. A lender must be able to assess utilisation, useful life and residual value.
This is where NVIDIA’s broader strategy comes into view: reference designs, standardised systems and fleet software can turn individually built data centres into comparable economic units.
Standardisation → Credit → Pooling → Securitisation
The cash flows of individual AI factories could, in time, be transformed into tradable securities — AI Infrastructure ABS.
That is precisely why specialists in private credit, infrastructure and structured finance are at the table. They know how to turn long-term contracts and physical assets into investable financial products and distribute them to institutional investors around the world.
Here lies an American advantage that extends beyond leadership in semiconductors and AI models: the United States has the deepest capital markets in the world and a financial industry capable of standardising, financing and distributing new asset classes.
The competition with China will therefore not be decided in research laboratories and semiconductor fabs alone. The ability to mobilise capital faster and on a larger scale will also become part of the AI race.
Ohio Shows How the Machine Could Work
A major project now announced in Ohio offers a glimpse of what this financial architecture could look like in practice.
SB Energy will build, own and operate a data centre at the PORTS-Pike Technology Campus. OpenAI has committed to using the capacity for 20 years. NVIDIA will exclusively supply the full compute stack — GPUs, CPUs, networking and software.
The first phase comprises 4.25 gigawatts of IT capacity. NVIDIA holds an option for a further 3.75 gigawatts. To support a total of 8 gigawatts of usable computing capacity, SB Energy and SoftBank intend to build at least 10 gigawatts of new power generation and invest $4.2 billion in the regional electricity grid.
The location itself is symbolic. The campus is being built on the site of a former US uranium-enrichment facility. Where the industrial infrastructure of the atomic age once stood, one of the largest infrastructures of the AI age is now intended to rise.
But the most important feature is the financial structure.
NVIDIA is investing $1.5 billion in SB Energy and providing credit support for land, power supply and the building shell. OpenAI provides the expected cash flow through its long-term lease. SB Energy owns and operates the infrastructure. On that basis, additional debt can be raised.
OpenAI provides the demand. SB Energy builds the asset. NVIDIA provides technology and credit support. The capital markets finance the expansion.
This makes visible what Jensen Huang means when he talks about establishing compute as an asset class.
At the same time, the structure creates a US-based nexus between energy, data centres, capital and cutting-edge NVIDIA technology. OpenAI secures long-term computing capacity. NVIDIA ties one of the most important model providers exclusively to its platform. And the United States creates industrial capacity that cannot simply be relocated to another country at short notice.
Preserving US leadership in AI is therefore being translated into concrete, power lines and long-term contracts.
NVIDIA Does Not Need to Own the Trillions
Critics warn of circularity when NVIDIA finances companies that subsequently buy its own GPUs. The more important question, however, is how much of NVIDIA’s own capital is being deployed — and who ultimately bears the economic risk.
NVIDIA is not financing the expansion entirely from its own balance sheet. It is using a limited amount of capital and creditworthiness to make a multiple of outside capital financeable.
NVIDIA is deploying its balance sheet selectively to make the broader build-out financeable.
Jensen Huang does not need to own the trillions. If NVIDIA-based computing capacity becomes standardised and fungible, its cost of capital can fall. Banks, private-credit funds and infrastructure investors can assume an increasing share of the financing.
This is more than conventional vendor financing. NVIDIA is helping secure land, power and buildings for the exclusive use of its own platform over long periods. The data centres can be upgraded with each new GPU generation while the longer-lived infrastructure remains in place.
Alongside technology and software, another moat begins to emerge:
a financing advantage.
For the United States, that creates strategic leverage as well. American innovation is being connected with American capital markets. The financial scalability of the infrastructure therefore strengthens not only NVIDIA, but the entire US AI ecosystem.
The Repo Multiplier and the Lesson of 2002–2008
A securitisation reaches its full financial potential only when the resulting securities are accepted as collateral in the repo market. An owner pledges the security, raises short-term funding against it and redeploys the capital.
Financed Asset → Collateral → New Financing Capacity
The years from 2002 to 2008 showed just how powerful — and dangerous — this mechanism can become when applied to mortgage-backed securities.
The capital created did not remain within the housing market. It moved around the world as hot money, flowing to wherever higher returns were available — risk assets, commodities, emerging markets or carry trades.
This is a crucial point.
The origin of the collateral does not determine where the liquidity created from it ultimately flows.
Repo funding, however, has to be rolled continuously. When doubts arise about the collateral, haircuts increase. Borrowers must post more collateral or unwind positions.
A liquidity multiplier can become a deleveraging multiplier.
The ability to multiply capital is therefore both a strength and a risk. For the United States, it could greatly accelerate the build-out of strategically important AI infrastructure. But it also requires realistic valuation of the underlying assets.
What Would Compute Be Worth as Collateral?
A port or power plant can remain economically useful for decades. Data-centre hardware ages much faster as new generations of GPUs arrive.
A lender therefore has to assess utilisation, contracts, technological obsolescence and residual value over a period of perhaps three to seven years.
The Ohio structure deliberately separates these different useful lives. Land, grid connection and the building shell form the long-lived base. GPUs and networking equipment can then be replaced on shorter cycles above it.
NVIDIA therefore appears not to be supporting primarily the residual value of any one GPU generation. It is supporting the infrastructure on which successive generations of compute can be installed.
The more fungible the overall system becomes, the easier residual values are to estimate — and the more effectively the asset can function as collateral.
An AI Infrastructure ABS is not a subprime MBS. The similarity lies in the mechanism, not in the quality of the underlying assets.
The emerging asset class would probably not be a pure “GPU ABS” either. It would rest on a combination of long-term usage agreements, energy infrastructure, buildings and regularly refreshed compute.
If this mechanism takes hold, the pace of expansion will no longer be determined solely by technology demand.
The price of capital will matter too.
When Capital Gets Cheaper
This is where the real macroeconomic and geopolitical significance may lie.
If standardised compute infrastructure becomes creditworthy, if loans can be pooled and securitised, and if those securities can in turn serve as collateral, then a given stock of capital can generate additional financing capacity.
That can reduce the effective cost of capital.
Lower capital costs would allow the United States to build more computing capacity, power supply and grid infrastructure more quickly. In the technological competition, the depth of US capital markets would therefore become an industrial speed advantage.
The effects could extend far beyond the original AI-infrastructure cycle. Investments that appear uneconomic at higher financing costs can suddenly become viable. Companies can invest more, cash flows can grow — and valuations can receive support.
Here too, the period before the financial crisis offers an interesting parallel. The liquidity created from mortgage lending and securitisation did not remain in the US housing market. It searched globally for returns.
If a new and substantial collateral and financing architecture really does emerge around compute, one of the most interesting questions will therefore be:
Where does the capital go after it has been created?
The answer could reach far beyond NVIDIA, data centres and even the United States.
Cui Bono? A Second Value Chain
Alongside the industrial AI value chain — semiconductors, memory, networks, energy and data centres — a financial value chain could emerge.
Banks earn fees from financing and structuring. Private-credit funds provide capital. Sponsors organise projects. Rating agencies assess credit risk. Asset managers distribute the resulting securities to investors.
The larger the credit market becomes, the greater the need to hedge interest-rate, credit and other risks.
Over time, compute infrastructure could therefore also give rise to a new market for derivatives and risk transfer.
The financial industry would not merely finance the build-out of AI infrastructure.
It would earn money from financing, securitising, distributing and hedging that capital.
This too is part of the American advantage. The United States is home not only to many of the world’s leading technology companies, but also to the banks, asset managers, private-credit specialists and capital markets capable of financing the industrial build-out and distributing its risks globally.
Technological and financial hegemony are beginning to reinforce one another.
Apparently, $500 Billion Is Not Enough
Whether this development actually extends all the way to securitisation, repo financing and a liquid secondary market remains an open question.
Nor do we know whether this is the long-term plan behind the remarkable group of NVIDIA, BlackRock, Blackstone, Brookfield, Apollo, Goldman Sachs and KKR.
What we do know is that these institutions possess exactly the capabilities required to turn an industrial investment cycle into a financeable asset class.
And we can now see what an initial concrete model looks like: OpenAI commits to long-term usage, SB Energy builds and owns the infrastructure, NVIDIA provides the technology and limited credit support, and the capital markets finance the expansion.
Seen in that light, the announced $500 billion takes on a different meaning.
It would not be the final amount.
It would be the foundation.
Just as Nelson’s fleet rested not only on ships, but on shipyards, raw materials, supply chains and ultimately a financial system capable of making all of them possible, the emerging US AI infrastructure could be supported by a second machine:
a machine for mobilising and multiplying capital.
That machine would finance a new industry while helping ensure that technological leadership, industrial capacity and the decisive computing power of the next era remain anchored in the United States.
The most important question may therefore no longer be whether $500 billion is enough.
It is what the US financial system can turn that $500 billion into.