Musk says SpaceX closed $40B debt for Nvidia chips

A company buying $40 billion of chips just told the smaller buyers they are in trouble. If you rent compute from one of them, that is your problem too.

Elon Musk said on X that SpaceX has closed a $40 billion debt deal to pay for Nvidia GPUs, TipRanks reported. Deliveries are due in the first and second quarters of 2027, and the debt is structured as private bonds issued by an off-balance-sheet entity, TipRanks said. Musk credited SpaceX's investment-grade credit rating and its Starlink and launch revenue.

$40 billion.

Not everyone has the deal done. Bloomberg reported that talks over pricing were still ongoing and the deal had not been priced, according to TipRanks. TipRanks names no lenders, rate or maturity.

Three weeks ago this was a report

The Financial Times first reported the talks in early October, as The Next Web summarized: about $10 billion in bank loans and $30 billion in investment-grade debt, with Apollo expected to lead and Pimco among the lenders in talks. The FT expected a close in 2027. Bloomberg called the talks early-stage and said they could still fall through.

If Musk's account holds, the deal moved a lot faster than that.

Musk's real message was about neoclouds

Then he turned to the rest of the market.

"That means the 200+ neoclouds are going to see guppy-level death and consolidation. Some will make it through, but not many," Musk wrote on X, according to TipRanks.

Neoclouds are the GPU rental companies that sprang up to sell AI compute, often buying chips with borrowed money. Musk said debt markets had nearly stopped lending to companies without major revenue outside AI data centers, TipRanks reported.

SpaceX is not a neutral observer. It rents out its own capacity: Google agreed to pay $920 million a month for access to about 110,000 Nvidia GPUs, The Next Web reported. A rival predicting its competitors' deaths is a forecast with a stake in the outcome.

What it means if you rent GPUs

You don't have to believe Musk's forecast to act on it.

If your training or inference runs on a smaller GPU cloud, check three things this week. What happens to your reserved capacity if the provider is acquired. How fast you can export your data, weights and images. Whether a second provider is set up and tested.

A second provider costs a little more each month. Losing your only one costs a lot more in a week.

What to watch

Watch for SpaceX to name the lenders and the rate, and for the first neocloud to announce a sale or a missed payment.

Have a second provider. Test it before you need it.