Who Is Paying for the AI Boom? Inside Wall Street’s Race to Finance Data Centers and GPUs

AI feels like a software revolution when you use it. Type a prompt into a chatbot and an answer appears in seconds.

Behind that simple interaction, however, sits something far more physical.

• Data centers.
• Thousands of GPUs.
• Power plants.
• Cooling systems.
• Networking equipment.
• Land.
• Long-term electricity contracts.
• And an extraordinary amount of money.

That last part is becoming one of the biggest stories in artificial intelligence in 2026. For the first phase of the generative AI boom, the world’s richest technology companies mostly looked capable of paying for expansion themselves.  Now the numbers are becoming so large that the financial system is being pulled into the project.

BlackRock, Blackstone, Apollo, Brookfield, KKR and Goldman Sachs are no longer watching the AI boom from the sidelines. They are helping finance it. In August 2026, Nvidia announced partnerships with those six financial groups designed to mobilize more than $500 billion of third-party capital for AI infrastructure.

Broadcom, meanwhile, was reportedly considering more than $60 billion of debt financing around AI-chip projects. The story underneath these deals is bigger than either company.

AI is turning into an infrastructure asset class.

And Wall Street wants a piece of it.

The AI Boom Has a Physical Infrastructure Problem

AI companies can create a new software model relatively quickly.

Building the physical infrastructure required to serve millions of users is much slower.

A hyperscale data center can require years of planning and construction.

Before it earns a dollar, someone may need to pay for:

• Land
• Construction
• GPUs
• Servers
• Cooling
• Electricity infrastructure
• Networking
• Backup power
• Security
• Long-term equipment contracts

IDC expects worldwide AI infrastructure spending to reach approximately $487 billion in 2026, around 53% above 2025. The research firm expects the annual market to pass $1 trillion by 2029. That is why AI financing is becoming more creative. There is simply too much infrastructure to fund everything in the old-fashioned way.

Nvidia Is Building a Financing Machine Around Its GPUs

Nvidia’s August announcement is probably the clearest example. The company signed memorandums of understanding with:

• Apollo
• BlackRock
• Blackstone
• Brookfield
• Goldman Sachs
• KKR

The planned financing platforms are intended to mobilize more than $500 billion of third-party capital over time for AI infrastructure used by Nvidia customers.

This is important to understand correctly. Nvidia did not announce that it had borrowed $500 billion. It is helping create financing vehicles that could allow customers to access enormous amounts of outside capital. Why would Nvidia care how its customers finance infrastructure?

• Because financing can become a bottleneck.
• A company may want Nvidia GPUs.
• It may have customers waiting to use those GPUs.

But it may not have billions of dollars available upfront to construct the required facility. If financial institutions can provide the capital, the customer builds the infrastructure faster—and Nvidia potentially sells more hardware. It’s vendor financing on a much larger and more sophisticated scale.

Jensen Huang Wants Investors to Think About Compute Differently

Nvidia’s CEO Jensen Huang has been making an interesting argument. He doesn’t want financial markets to think of GPUs as ordinary technology equipment. He wants them to think of compute as productive infrastructure. Nvidia argues that modern accelerated-computing systems can produce revenue, serve many customers and be redeployed between workloads. That, Nvidia says, gives compute characteristics that can make it suitable for institutional financing. Wall Street clearly sees an opportunity.

Goldman Sachs CEO David Solomon described the possibility of creating a credit market backed by Nvidia compute.

KKR described compute as critical infrastructure. BlackRock’s Larry Fink emphasized connecting long-term institutional capital with AI infrastructure demand. It sounds like technology language. But increasingly, it is finance language.

What Does “GPU-Backed Finance” Actually Mean?

Imagine a company wants to create an enormous AI cloud. It needs thousands of expensive processors. Instead of paying for all of them upfront, a financing structure could help fund the equipment based partly on the economic value of the compute it provides. Those assets may then support loans, leases or other financing structures.

This is not unusual in the broader economy. Airlines finance aircraft. Telecommunications companies finance network equipment. Businesses lease industrial machinery. Energy projects borrow against future cash flows. The unusual part is the underlying asset. AI chips become obsolete faster than aircraft or power plants. A current-generation GPU may be extremely valuable today. A much faster generation could arrive a few years later.

Broadcom Shows That Financing Needs Aren’t Limited to Nvidia

Just ten days after Nvidia’s announcement, another huge number arrived.

Reuters reported on August 20 that Broadcom was exploring debt financing of more than $60 billion for AI computing projects.

The financing reportedly could include a $30 billion junior tranche alongside senior-secured debt.

Broadcom has become increasingly important in custom AI accelerators designed for major technology companies.

So its financing discussions illustrate another part of the trend.

Companies aren’t merely raising money to construct buildings.

They are financing the chips and compute capacity inside them.

The financial stack underneath AI is becoming almost as complicated as the technology stack.

Why Isn’t Big Tech Simply Paying Cash?

Companies like Alphabet, Microsoft, Amazon and Meta are among the most profitable corporations in history.

So why involve Wall Street?

Scale.

In 2025, Amazon, Alphabet, Meta, Microsoft and Oracle issued around $121 billion of new debt, compared with an average of approximately $28 billion annually from 2020 through 2024.

AI infrastructure is forcing these companies into a far more capital-intensive era.

The traditional internet business could be wonderfully asset-light.

Build software once.

Distribute it globally.

Add users at relatively low incremental cost.

AI changes parts of that equation.

Every large model needs computation.

Every additional generation of models can require larger clusters.

Running those models for hundreds of millions of users requires continual capacity.

That makes AI look less like conventional software and more like telecommunications, energy or industrial infrastructure.

And infrastructure is usually financed.

A $2 Trillion Opportunity for Credit Markets

Apollo thinks the financing requirement could become enormous.

The firm estimates that the AI ecosystem could fundamentally support more than $2 trillion in additional investment-grade debt.

But it doesn’t believe public investment-grade bond markets can absorb all of it.

Apollo estimates those markets may have capacity for less than $1 trillion of additional AI debt through 2030 because investors and ratings systems impose concentration limits.

That leaves a very large gap.

Private credit could fill part of it.

So could:

• Project finance
• Asset-backed lending
• Equipment financing
• Private investment-grade debt
• Infrastructure funds
• Special-purpose financing vehicles
• Long-term leases

This explains why some of the world’s biggest alternative asset managers are now appearing alongside semiconductor companies in AI announcements.

They aren’t just investors in Nvidia stock.

They can become providers of the capital that allows Nvidia’s customers to buy Nvidia’s products.

AI Infrastructure Is Becoming a New Business for Private Credit

Private credit has expanded significantly since the global financial crisis.

Instead of a company borrowing only from a bank or issuing a public bond, private funds can directly finance businesses and infrastructure projects.

AI is almost tailor-made for that market.

The projects are huge.

Traditional lenders may not want the entire exposure.

The financing can be customized.

Assets such as data centers, power contracts and equipment can provide collateral.

And investors are looking for long-duration opportunities.

Apollo says AI-related issuance already represents nearly 40% of longer-duration investment-grade corporate bond supply under its analysis.

That is an extraordinary level of concentration for a technology trend that entered the mainstream only a few years ago.

Data Center Leases May Be Just as Important as Debt

If you only look at corporate bonds, you will miss a major part of the AI financing story.

Tech companies are signing enormous data-center lease agreements.

Reuters calculated in August that five major technology companies—Microsoft, Meta, Oracle, Amazon and Alphabet—had approximately $1.09 trillion of future lease commitments.

Many of those commitments relate to data centers that have not yet entered service.

Accounting treatment makes this especially interesting.

A lease that has not started may be disclosed in company filings but may not yet appear as a lease liability on the balance sheet.

For example, Alphabet disclosed $85.2 billion in future payments for leases, primarily data centers, that had not yet commenced as of June 30.

Meta disclosed approximately $279 billion of such commitments at the end of June and signed an additional roughly $68 billion of data-center leases in July.

These aren’t hidden in the sense of being secret.

They are disclosed.

But casual investors looking only at headline debt figures can easily underestimate the size of future obligations.

The Real Risk Is a Timing Mismatch

There is a simple reason people are becoming nervous about all this financing.

AI infrastructure needs money today.

The revenue expected from that infrastructure arrives later.

That difference matters.

A company signs a 15-year or 20-year data-center commitment because it expects long-term AI demand.

A lender finances GPUs because it expects those assets to produce valuable compute.

An infrastructure fund invests because it expects users to keep paying for AI capacity.

The entire system works beautifully if those assumptions are right.

Problems begin if AI demand grows but monetization doesn’t.

Or if demand moves to newer hardware.

Or if computing costs fall much faster than expected.

Or if businesses decide they don’t need as much premium AI capacity as the industry built.

None of these outcomes is guaranteed.

But this is what lenders are being paid to evaluate.

AI Chips Are Not Buildings

One of the most interesting debates is whether a GPU should really be considered an infrastructure asset.

There is a good argument in favor.

A GPU in a data center can generate revenue every hour it is being used.

It can serve many customers.

It can potentially be moved between workloads.

But there is also an obvious problem.

Technology depreciates differently from traditional infrastructure.

A building may remain useful for 40 years.

A state-of-the-art AI chip faces constant competition from newer, faster hardware.

This creates what finance professionals call residual-value risk.

What is the asset worth when the loan is only halfway through its life?

Nvidia argues that its CUDA software ecosystem helps extend the economic usefulness of its hardware.

Critics remain unconvinced that rapidly evolving compute should be valued like conventional long-lived infrastructure.

That debate will matter more as these financing platforms grow.

Is Wall Street Taking on Too Much AI Risk?

Not yet in a way that automatically signals a financial crisis.

The companies borrowing most heavily are generally strong businesses.

AI demand remains substantial.

Data centers and compute are real productive assets.

But the exposure is spreading.

In August, Reuters reported signs of investor fatigue in the U.S. corporate bond market after AI-related borrowing surged.

AI-related issuance had reached approximately $220 billion in 2026 under the report’s definition, and investors were beginning to demand somewhat greater compensation for absorbing new supply.

That’s worth paying attention to.

The first warning sign may not be a default.

It may simply be that lenders begin saying:

“We’ll still finance it—but the price is going up.”

What the Situational Awareness Crisis Does—and Doesn’t—Tell Us

July also produced one of the year’s most dramatic AI-related financial events.

Situational Awareness, the hedge fund founded by former OpenAI researcher Leopold Aschenbrenner, had built heavily leveraged positions in AI-related equities.

A sharp decline in several holdings triggered severe losses and margin pressure.

The fund eventually sold most of a roughly $16 billion public-stock portfolio to Citadel, which subsequently unwound much of the exposure.

This should not be described as an AI-infrastructure debt collapse.

It wasn’t.

But it is relevant for another reason.

It shows what can happen when an overwhelmingly bullish AI thesis is combined with excessive financial leverage.

A great long-term idea can still become a terrible short-term investment if too much borrowed money is involved.

What Should Investors Actually Monitor?

The headline numbers will continue getting bigger.

• $60 billion.
• $500 billion.
• $1 trillion.
• $2 trillion.

They are eye-catching, but numbers alone won’t tell you whether the financing system is healthy.

Watch the relationships between them.

AI usage versus infrastructure capacity

Are new data centers filling up?

Revenue versus AI usage

Are companies making enough money from the demand they’re generating?

Revenue versus financing obligations

Can those earnings comfortably support debt, lease and interest payments?

Loan maturity versus asset life

Will the hardware remain economically useful throughout the financing period?

Credit spreads

Are investors demanding significantly higher yields to keep lending?

Off-balance-sheet commitments

How much future infrastructure has already been contracted but not yet appeared as a recognized liability?

Private credit exposure

If public markets become saturated, how much risk moves into less-transparent private markets?

Those questions will tell us more than one dramatic debt estimate.

AI Is Changing Big Tech’s Financial Model

There is a deeper transformation happening here.

For years, many of the world’s biggest technology companies were famous for generating enormous amounts of cash while carrying relatively conservative balance sheets.

AI is changing that model.

These companies are becoming more capital intensive.

They are signing longer commitments.

They are borrowing more.

They are working with infrastructure investors.

Their suppliers are helping arrange financing.

And private-credit firms are becoming increasingly important to the technology ecosystem.

That means AI’s next phase won’t be determined only by whether engineers can build better models.

It will also depend on whether financiers believe the infrastructure behind those models can produce acceptable long-term returns.

The Next AI Competition May Be About the Cost of Capital

Every AI company wants faster chips.

Every AI company wants more computing capacity.

But eventually another advantage may become just as important:

Who can finance infrastructure most cheaply?

A company that borrows at attractive rates can potentially offer cheaper compute.

A company with stronger long-term contracts may receive better financing terms.

A company using hardware that lenders believe retains value may gain access to more capital.

That creates a new competitive layer.

The AI winners may not simply be the companies with the best models.

They may also be the companies with the strongest financing structures.

Final Thoughts

The numbers behind the AI boom have entered a different league.

IDC expects nearly $487 billion of AI infrastructure spending in 2026.

Nvidia and six major financial institutions are working on platforms intended to mobilize more than $500 billion of third-party capital.

Broadcom has reportedly explored more than $60 billion of AI-related debt financing.

Apollo believes the ecosystem could support more than $2 trillion in additional investment-grade borrowing.

Those figures don’t prove that AI is a bubble.

They prove something else.

Artificial intelligence has become too capital-intensive to remain only a technology-industry story.

Banks, asset managers, bond investors, private-credit funds and infrastructure investors are now helping build the physical machine underneath AI.

That could unlock the next wave of growth.

It also means that if expected AI returns disappoint, the financial consequences will extend beyond Silicon Valley.

Frequently Asked Questions

Who is financing the AI infrastructure boom?

AI infrastructure is being financed through a combination of Big Tech cash flow, corporate bonds, leases, banks, private credit, infrastructure funds and institutional investors.

What is Nvidia’s $500 billion AI financing plan?

Nvidia announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to establish financing platforms intended to mobilize more than $500 billion in third-party capital over time for AI infrastructure.

Why does AI require so much infrastructure spending?

Training and running modern AI models requires expensive GPUs, servers, networking, electricity, cooling and large-scale data centers. IDC forecasts roughly $487 billion of worldwide AI infrastructure spending in 2026.

What is private credit’s role in AI?

Private-credit and infrastructure investors can finance AI projects that may be too large or concentrated for traditional public bond markets. Apollo estimates more than $1 trillion of future AI financing could potentially migrate toward private and structured credit markets.

Are GPUs being used as collateral?

New financing structures are being developed around the economic value of Nvidia compute and AI equipment. Nvidia describes its compute as an investable infrastructure asset, while some investors question how rapidly evolving chips should be valued over long financing periods.

Could AI infrastructure create a financial crisis?

There is currently insufficient evidence to conclude that AI infrastructure financing will create a financial crisis. The main risks are excessive leverage, weak returns, falling asset values, credit concentration and future revenue failing to keep pace with financial obligations.