The Fed Raised Rates. AI Infrastructure Faces the Sharpest Test.

The bill that arrived before the decision

Michael Intrator reported $2.6 billion in quarterly revenue in August, alongside $640 million of net interest expense. That expense already equaled about a quarter of CoreWeave's sales before the Federal Reserve moved.¹

The Fed raised its target range by 0.25 percentage point to 3.75% to 4.00% on September 16.^2,18 For AI investors, the consequential question is which companies must keep finding outside money to deliver their promised growth.

CoreWeave ranks first for direct financing exposure. Applied Digital and Nebius follow as businesses whose expansion depends heavily on capital access. Oracle has substantial exposure but stronger protection. SoundHound and Palantir illustrate separate risks to shareholders: dilution and valuation. This ranks financial vulnerability among these companies. It does not forecast tomorrow's stock returns.

CoreWeave: the clearest direct exposure

CoreWeave's business requires buying computing infrastructure before collecting all the revenue it can generate. In the first half of 2026, the company reported $3.7 billion of operating cash flow against $14.1 billion of cash capital expenditure. The resulting $10.5 billion difference explains its dependence on financing.¹

A backlog promises revenue. A balance sheet decides whether you can afford to deliver it.

Its filing for the June quarter disclosed that a one-percentage-point move in floating borrowing rates would change quarterly interest expense by about $30 million. Scaling that mechanically to a quarter point gives $7.5 million, before separate swap protection, assuming unchanged balances and matching benchmark movement. That is an illustration, not a forecast of exact Fed pass-through.³

The immediate increase is manageable against the existing interest bill. The larger risk is repeated borrowing at higher rates, potentially combined with lenders demanding wider credit spreads. Strong demand can coexist with weaker shareholder returns if financing consumes more of each project's earnings. Swaps help control rates. They do not fund the next building.

Applied Digital and Nebius: construction creates dependence

Applied Digital reported about $90 million of operating cash in fiscal 2026 against $2.9 billion of purchases of property, equipment and other assets. Its facilities require substantial spending before tenants produce the intended cash returns. Higher financing costs can therefore make new projects less attractive and overruns more expensive.⁴

There is meaningful protection: much of its financing is fixed. A June bond issue refinanced an earlier floating-rate bridge. Treating that bridge as still outstanding would exaggerate the immediate effect of this hike. The pressure belongs primarily on additional construction and future refinancing, rather than every existing dollar of debt.

Nebius has a similar expansion problem, with different financing arrangements. In the June quarter, cash investment exceeded operating cash by about $3.4 billion.⁵ It subsequently raised $5.75 billion through fixed-coupon convertible notes. That strengthens liquidity while introducing potential dilution.⁶ The relevant question is what its next expansion costs, not whether the June cash gap proves an imminent shortage.

Oracle: enormous spending, stronger shock absorbers

Oracle belongs on the watchlist because its AI infrastructure spending has become enormous. For the quarter ended August 31, operating cash was $23.1 billion and capital expenditure was $28.5 billion, a shortfall of about $5.4 billion.⁷

But debt structure matters more than a frightening total. At May 31, $122.5 billion of Oracle's $130.1 billion principal consisted of fixed-rate senior notes. Most existing borrowing therefore does not reset when the Fed moves.⁸

Customers also supplied $11.4 billion of financing-component prepayments during the latest quarter, already included in operating cash flow. That transfers some funding burden to buyers, although Oracle still owes the services.⁷ Its vulnerability is the price of financing the remaining buildout and refinancing maturities. The existing software business and fixed debt provide buffers that younger infrastructure companies cannot easily reproduce.

SoundHound and Palantir: shareholders face different risks

SoundHound's exposure runs through equity financing. Before the LivePerson acquisition, it used about $60 million of operating cash in the first half and raised $48.5 million gross by selling shares. If tighter financial conditions depress its share price, raising the same money requires issuing more shares.⁹

The company described its combined balance sheet as debt-free after the September 4 acquisition, so its June cash figures are not a combined-company runway.¹⁰ The issue is financing future losses and expansion without excessive dilution.

Palantir presents a different case. It reported $9.2 billion of cash and short-term Treasury securities and substantial operating cash generation. Its business has financial room.¹¹

Its valuation has less room. September 15's $172.56 close and July's disclosed share count put equity value at about $415 billion, or 51 times management's 2026 revenue-guidance midpoint.^12,13 Investors demanding higher returns could pay less for those future earnings even while the company grows.

Why cash generation changes the ranking

Microsoft generated about $183 billion of operating cash in fiscal 2026 against $116 billion of cash capital expenditure. The $67 billion difference, before other uses and excluding finance-lease additions, gives it substantial capacity to finance infrastructure internally.¹⁴ Its shares remain exposed to valuation changes, but its expansion is less dependent on the next financing round.

Even smaller AI names deserve individual examination. C3.ai reported $651 million of liquidity and modest positive free cash flow for its latest quarter.¹⁵ Those facts weaken the case for automatically treating every smaller AI company as a funding casualty.

The strongest counterargument is that AI demand can remain strong enough to justify expensive capital. Fixed borrowing, customer prepayments and successful fundraising already provide protection. Higher rates become most damaging when funding costs rise faster than project returns, or when customer collections arrive later than expected.

Watch financing terms after the headline

A Fed hike does not mechanically determine stock losses. Markets respond to the unexpected part of policy and what it reveals about the economy, and longer-term borrowing costs need not move one-for-one with the overnight rate.^16,17

For Intrator and the other AI builders, watch the coupon on the next loan, the equity issued to fund it, and the time between construction spending and customer payment. Those figures will show whether strong demand is still producing attractive returns for shareholders.