Why does a company with $50 billion in cash borrow $25 billion anyway?

When a cash-rich company borrows money it does not need, the bond market is voting on its story. The size of the order book tells you what the share price cannot.

Nvidia raised $25 billion in investment-grade bonds this week. It is sitting on roughly $50 billion in cash and threw off about $119 billion in free cash flow over the past year. It did not need the money. So why borrow?

The answer is in who showed up to lend. The deal drew more than $85 billion in orders. That is institutional money queuing to fund the AI build-out, and it is the clearest read on Wall Street conviction you will find this week.

Why a cash-rich company sells bonds anyway

Borrowing when you have cash is normal for mega-cap tech. Apple and Microsoft have done it for years, usually to fund buybacks or for treasury reasons, not need. The benign read on Nvidia is the same: lock in cheap debt while the window is open.

The interesting part is not the borrowing. It is the demand. An order book more than three times the deal size, for a company that did not have to issue at all, says lenders want exposure to this story. That appetite is the signal.

The quiet instrument in the loudest trade: LQD

Most traders watch the Nvidia share price. The bond demand shows up somewhere quieter: the iShares investment-grade corporate bond ETF, LQD.

LQD holds a basket of high-grade corporate debt. When companies like Nvidia flood the market with new bonds, that supply has to be absorbed. If buyers keep paying up for the paper, prices hold and spreads stay tight. LQD is the simplest place to watch whether credit markets still believe the AI build-out, while everyone else stares at the stock.

This matters because the bond market often votes before the stock market finishes arguing. Nvidia shares actually fell on the announcement day and now trade below that level, near $200, even after crossing a $5 trillion market cap. The equity side is jittery. The credit side, for now, is calm.

The build was real. The financing got ahead of it before.

Here is the part the order book does not settle. The Nvidia deal is one slice of more than $300 billion in AI infrastructure debt issued this year, per JPMorgan. The four largest US hyperscalers guide to $700 to $725 billion in combined capex for 2026.

Two history lessons sit underneath that number.

• Fiber, 1999 to 2001. Carriers borrowed hundreds of billions to build internet capacity. Demand did eventually arrive. But the financing ran ahead of the cash flows, overcapacity emerged, and a wave of telecom defaults followed.
• Shale, 2013 to 2015. Energy producers issued heavily to fund drilling. Bond demand was strong while the story was hot. When prices fell, the most leveraged builders struggled and sector spreads blew out.

In both cases the asset was real. The borrowing outran the returns. A heavily oversubscribed order book is exactly what those cycles looked like at their peak.

What the demand proves and what it does not

Strong demand proves capital is willing today. It does not prove the returns will show up. That is the honest split.

Dell makes the case the build is real: a record $51.3 billion AI server backlog and fiscal 2027 AI revenue guidance near $60 billion. The question is whether that backlog converts to margin and cash flow fast enough to justify the debt being raised against it.

The evidence leans toward genuine conviction rather than blind mania, for now. The early crack would not be the stock. It would be a weak order book on the next mega AI bond deal, or LQD slipping as spreads widen. Watch the auctions and watch the ETF. They will tell you before the headlines do.

None of this is a trade call. Credit can stay tight right up until it doesn't, and rates drive both the ETF and the cost of the next round of borrowing.