How Broadcom's plan to borrow more than $60 billion for AI chips turns compute buildout into a Wall Street asset class.

Chipmakers are no longer just selling silicon. They're financing it with debt structured like a power plant — and that changes who's really underwriting the AI boom.

Broadcom does not need to borrow money to make chips. It is one of the most profitable companies in semiconductors. And yet it is now in talks with lenders to raise more than $60 billion in debt to finance a wave of AI chips destined largely for Anthropic and other AI companies. The structure under discussion could reach as high as $100 billion. That gap — between a company that could pay cash and a company choosing to raise a nation-sized pile of debt instead — is the whole story of how AI infrastructure is being funded in 2026.

The Broadcom AI debt financing is not really a chip deal. It is a financial-engineering deal that happens to produce chips. And it marks the moment when the cost of building AI stopped fitting on the balance sheets of technology companies and started spilling into the debt markets that normally fund highways, pipelines, and power stations.

What Broadcom is actually building

The mechanics are worth slowing down on, because the mechanics are the innovation. According to Bloomberg, the financing under discussion could include roughly $60 billion to $70 billion in senior secured debt, with Broadcom guaranteeing a portion, alongside about $30 billion in junior financing — potentially bringing the total package close to $100 billion. The money would fund custom AI chips and networking gear for Anthropic and, potentially, other AI companies.

This is not Broadcom's first move of this kind. In June, the company partnered with Apollo and Blackstone to finance a $35 billion expansion of Anthropic's computing capacity using Broadcom's custom silicon. The new, far larger raise extends that approach. And the choice of partners tells you what this actually is: Apollo and Blackstone are not chip companies. They are private-credit and asset-management giants. When they show up, the deal has become a financing product.

The logic is that AI infrastructure now costs so much, and demands so much capital so far ahead of revenue, that no single technology company wants to carry it alone. So the industry is doing what capital-intensive industries have always done: it is building special-purpose financing structures around expected future demand. The chips get funded the way a new terminal or a gas plant gets funded — with layered debt, secured against contracted future cash flows, spread across banks, private-credit funds, and institutional investors who want exposure to AI without buying a single GPU.

Why "AI capex as an asset class" is the real headline

For most of the boom, the AI buildout sat on the balance sheets of the hyperscalers — the Microsofts, Googles, and Amazons that could absorb tens of billions in capital spending out of operating cash flow. That model has a ceiling. As the compute demands of frontier models keep compounding, and as pure-play labs like Anthropic need capacity on a scale their own revenue can't yet justify, the buildout needs a bigger balance sheet than any one company can provide.

Debt markets are that bigger balance sheet. By structuring AI compute as senior-secured and junior tranches, Broadcom and its financing partners are turning future AI demand into something a pension fund or an insurer can underwrite. That is what people mean when they say AI capital expenditure is becoming an asset class of its own. It is no longer just a line item in a tech company's spending plan; it is a category of investment with its own risk profile, its own tranches, and its own returns, sold to investors who never touch the technology.

That shift has a profound consequence. It means the AI boom is now underwritten not only by the optimism of technologists and venture capitalists, but by the credit models of Wall Street. And credit models care about one thing above all: whether the future cash flows that secure the debt actually show up.

For the stakeholders

For Broadcom, the appeal is scale without strain. By raising debt secured against AI demand rather than funding chips off its own cash, the company can serve enormous customers like Anthropic — and position itself as the go-to alternative to Nvidia for custom accelerators — without betting its own balance sheet on a single boom. It becomes indispensable to the AI supply chain while offloading much of the financial risk onto the credit markets.

For Anthropic and the other AI labs, this kind of financing is close to existential. Frontier AI requires compute at a scale that outruns current revenue by years. Structures like this let a lab secure capacity now and pay for it against future growth — the only way the numbers work when you need $35 billion of infrastructure before you've earned it. The risk is obvious: the lab is committing to demand it is confident will materialize but has not yet realized.

For private-credit firms like Apollo and Blackstone, AI infrastructure is a new frontier for capital that has been hunting for large, long-duration deals to fund. Lending against contracted compute demand, secured and tranched, is exactly the kind of product they are built to originate and distribute. They get scale, yield, and a foothold in the defining buildout of the decade.

For investors and, eventually, the wider economy, the picture is more sobering. When AI capex becomes a debt-financed asset class, a slowdown in AI demand stops being a story about disappointed shareholders and becomes a story about serviced debt. Senior-secured tranches assume the cash flows arrive. If AI adoption plateaus, or if a lab's growth stalls before it can service the compute it committed to, the losses don't stay inside the technology sector — they ripple through the credit markets and the institutions that hold the paper. The more the boom is financed with debt, the more its failure modes start to look like the failure modes of every other leveraged buildout in history.

The zoom-out

There is a recognizable pattern here, and it predates AI by a century. Railroads, telecom, and fiber-optic networks were all built the same way: enormous, capital-hungry buildouts, funded increasingly with debt against the promise of future demand, sold to investors who wanted exposure to an unstoppable trend. Some of those buildouts paid off spectacularly. Some ended in defaults and glutted capacity that took a decade to absorb. The technology was usually real either way. What varied was whether demand arrived on the schedule the debt required.

AI compute is now entering that phase. The Broadcom AI debt financing is a sign of confidence — you don't structure $100 billion in tranches for a market you doubt — but it is also a sign that the buildout has outgrown the people who believe in it most and now depends on the people who price risk for a living. The chips will get built. The open question is the one every leveraged boom eventually has to answer: does the demand show up on time to pay for them?

For anyone building or investing in AI, the signal is to watch the financing structure as closely as the technology. When a boom shifts from equity to debt, it is maturing — and it is also raising the stakes of being wrong.

Every great buildout is a bet that demand will arrive before the interest does. That's the bet AI just placed, at the scale of a small economy.

Frequently Asked Questions

How much debt is Broadcom raising for AI chips?

Broadcom is in talks with lenders to raise more than $60 billion in debt, with a structure that could include $60 billion to $70 billion in senior secured debt plus roughly $30 billion in junior financing — potentially bringing the total package close to $100 billion. The financing would fund custom AI chips and networking gear for Anthropic and potentially other AI companies.

Why is Broadcom using debt instead of its own cash?

Financing AI infrastructure through structured debt lets Broadcom serve enormous customers at scale without committing its own balance sheet to a single boom. By raising senior and junior tranches secured against expected AI demand, it shifts much of the financial risk onto credit markets while positioning itself as a leading alternative to Nvidia for custom accelerators.

What does "AI capex as an asset class" mean?

It means AI compute spending is being packaged into tranched, secured debt that institutional investors — pension funds, insurers, private-credit firms — can underwrite, rather than sitting solely on technology companies' balance sheets. AI infrastructure gets financed the way power plants and pipelines are, turning future compute demand into a distinct category of investment.

What is the risk of financing AI infrastructure with debt?

If AI demand slows or a lab's growth stalls before it can pay for the compute it committed to, the losses fall on lenders and the institutions holding the debt, not just on shareholders. Debt-financed booms transmit their failures through credit markets, which is why the scale of AI's borrowing raises the stakes of any demand shortfall.

Editor's note — sources: Bloomberg, Yahoo Finance, CNBC, The Next Web.

Share this article: