For most of the past two years, the companies building artificial intelligence have raised money on their own terms. Monday was the day the market started naming the price.
Consider the day’s biggest deal. Marvell agreed to buy Lumentum for a headline $115 billion, paid entirely in Marvell stock. The strategic logic is plausible: AI clusters need optics, and optics need tighter integration with silicon. Yet by the closing bell Marvell had lost almost 13% and Lumentum almost 19%. Because the deal pays a fixed number of shares, the sale price shrank in real time. A transaction meant to show confidence in AI demand became a live poll on it, and the vote went against the buyer.
Investment grade, junk prices
Then look at Oracle’s $20 billion raise. Half came from new shares sold below the day’s closing price. The other half came from bonds with coupons running from 7% to 8.65%, levels more typical of high-yield borrowers than of one of the world’s largest software companies. Oracle had promised investors that this would be the last trip to the capital markets for its current spending cycle. Creditors appear to have taken the company at its word and charged accordingly.
Even the most anticipated listing of the year isn’t immune. Anthropic’s updated prospectus shows a company with roughly $17 billion of quarterly revenue and close to $900 billion of compute and infrastructure commitments. Some prospective investors are reportedly arguing for a valuation well below the levels floated a few weeks ago. That is what price discovery looks like.
Competing for the same dollar
The timing matters. Washington plans to borrow $618 billion this quarter and $800 billion next, with 10-year yields hovering near 4.9%. Utilities are in the market too, as Xcel Energy’s $2.2 billion convertible showed. AI borrowers no longer have the field to themselves, and capital that once chased any data-center story now has alternatives that pay well with less risk.
None of this means the buildout stops. Demand for computing power is real, and most of these deals will get done. But the era of nearly free money for anything labeled AI is over. Projects that only work with cheap capital will be found out first. For investors, the question is no longer whether AI companies can raise money. It is how much they will have to pay for it, and who ends up covering the bill.
