Big Tech’s AI buildout is moving from cash-flow story to bond-market story, and Kevin Warsh’s first Fed meeting just made that shift harder to ignore.

Nvidia doesn’t need the money. That is what makes its $25 billion bond sale the useful starting point. The chipmaker filed to sell investment-grade debt for the first time since 2021, then upsized the deal from $20 billion after investors put in more than $85 billion of orders, the Financial Times reported. Nvidia still has a balance sheet most companies would envy, and the proceeds are officially for general corporate purposes, including refinancing. But you don’t have to squint to see the bigger message. Even the company selling the picks and shovels of the AI boom wants more dry powder.

That is the part investors should sit with. AI is no longer just a capital spending line buried inside earnings calls from Microsoft, Amazon, Meta, Alphabet and Oracle. It is becoming a financing machine of its own. Morgan Stanley projects AI-linked global debt issuance will reach nearly $570 billion in 2026, according to figures cited by CNBC, after roughly $236 billion had already been sold by the end of May. This isn’t a few weaker companies borrowing because they ran out of cash. It is some of the largest companies in the world deciding that the data center race is too expensive, and too urgent, to fund only from operating cash.

Look at the names. Meta has already tapped the bond market for $25 billion. Oracle has done the same. Alphabet sold $20 billion of debt, including a rare 100-year sterling bond. Amazon has also lined up large new financing for AI infrastructure, with MarketWatch noting a $17.5 billion loan tied to the same buildout. These companies are not equal credit risks, and treating them as one giant Big Tech basket is lazy. Alphabet can absorb more pain than Oracle. Nvidia is in a different position again. But the direction is shared: the old cash-rich technology model is taking on more leverage to keep feeding AI capacity.

Also read: You could be facing an AI chatbot at your next job interview, Lloyds Banking Group bets £100 million on AI as it recruits the executive who rewired DBS, Washington just showed Europe how quickly an AI dependency can become a crisis

Here’s the thing: shareholders who bought these businesses for fortress balance sheets now own something more sensitive to the price of money. CNBC’s analysis points to credit spreads already widening as issuance piles up. Oracle is the warning label. Moody’s rates it Baa2, only two notches above junk, and the company is expected to burn as much as $28 billion of free cash flow in 2026. Even stronger names are not immune. Alphabet and Meta can still borrow on good terms, but their free cash flow is being squeezed as capital expenditure climbs. If you own these stocks and never cared about Treasury yields, you need to start caring.

The timing is awkward because the Fed just stopped sounding friendly. Kevin Warsh, confirmed by the Senate on May 13 and now in Jerome Powell’s old chair, used his first policy meeting to keep rates at 3.5% to 3.75%, according to Kiplinger’s live account of the June meeting. That part was expected. The tone was not. Warsh stripped back forward guidance, said the Fed would deliver price stability, and left investors with less of the hand-holding they grew used to under Powell. The Financial Times called the debut hawkish after Treasury yields jumped and the S&P 500 fell 1.2%.

That matters in a very practical way. A data center financed at easy-money rates is one thing. A data center financed while the two-year Treasury yield is around 4% and the Fed is openly discussing the possibility of higher rates is another. Big Tech can still borrow. The June Nvidia deal proved there is plenty of demand for the strongest credits. But demand is not the same as cheap money, and cheap money is what made many AI spending plans look cleaner than they may really be.

There is also a circularity here that should make you uncomfortable. Cloud companies are borrowing to build data centers. Those data centers buy Nvidia chips. Nvidia is investing across the AI ecosystem and raising debt of its own. Investors then lend more because AI revenue growth looks unstoppable. That loop can work for a long time when cash flows keep arriving. It gets much less forgiving if customers delay projects, if power constraints slow construction, or if the promised productivity gains take longer to show up in actual profits.

Frankly, the market has been too willing to treat AI debt as a footnote because the borrowers have famous names. That is not analysis. A $25 billion bond sale from Nvidia is still a $25 billion bond sale. A company can be wildly profitable and still change the risk profile of its equity when it starts leaning harder on debt. The old technology trade was about growth, margins and cash. The new one adds duration, refinancing and central-bank policy to the same page.

Warsh has made that page harder to read. He has promised less forward guidance, not more, and half the Fed committee’s projections reportedly pointed to at least one rate increase this year. So the AI buildout now has two clocks running at once: the industry clock, where every company wants capacity before its rivals get it, and the bond-market clock, where every new issue depends on what investors think inflation and rates will do next. Nvidia got its deal done. The question is how many more companies can keep doing the same if the Fed refuses to make borrowing easier.