For most of the AI boom, the scarce input was chips. This week’s deal flow suggests the scarce input is now credit, and lenders are deciding which AI cloud providers get to keep building.
Consider the spread of outcomes in just a few days. SpaceX lined up $40 billion of debt for chips with an Apollo affiliate, an arrangement we covered on Monday, and Elon Musk credited the company’s investment-grade rating and its Starlink and launch revenue. Nebius secured a $3.5 billion senior secured facility from Citi the week before, though with a 364-day maturity and a floating rate starting at 2.75 percentage points over SOFR. In Australia, AirTrunk on Tuesday assembled 32 lenders for an A$4.25 billion green loan backed by a decade-long build plan.
Then there is the rest of the field. Musk wrote over the weekend that debt markets have almost entirely closed to companies without substantial revenue outside of AI data centers, and predicted that more than 200 neoclouds face a brutal shakeout. Chicago-based Argentum AI is in talks for a $5 billion chip financing with a large private credit firm, according to people familiar with the talks cited in press reports, with the debt meant for chips leased to customers that lack investment-grade ratings. Financing for neoclouds has tightened.
Credit is doing the sorting
None of this means the demand story is broken. General Compute, an inference-focused startup, raised a $100 million Series A on Tuesday at an $800 million valuation, and its chief executive has said that fears of a chip-led collapse are overstated. But even that company’s model is shaped by its lenders: it pairs equity with a $400 million debt facility and expects raising capital to remain a full-time job.
The pattern is that capital is flowing to borrowers that can point to something other than GPU rental income: an investment-grade parent, signed customer contracts, diversified revenue or hard real estate. Lenders are pricing the risk that an accelerator loses value faster than a loan amortizes, and that a customer without a credit rating stops paying. Those are reasonable concerns, and they push the market toward a smaller number of larger, better-capitalized providers.
What to watch
Investors should read financing terms as closely as revenue growth. Short maturities, floating rates and lease structures tied to unrated customers are signs of a provider that will need to refinance in a less forgiving market. Long-dated, syndicated facilities with broad bank participation are signs of one that lenders expect to be around in five years.
The AI buildout was sold as a race for compute. It is turning into a test of balance sheets, and the companies that pass will likely be the ones left to consolidate the rest.
