Skip to content

SpaceX borrows $40 billion for Nvidia AI chips as Musk warns most of 200-plus 'neoclouds' will not survive

Musk says lenders have all but shut off money to AI compute renters without other income, days after Nvidia floated a $2,000-a-year fee per GPU for its AI agent software.

Published 2 min read
WhatsApp
Elon Musk in a dark suit and black tie at an indoor event
Elon Musk in a dark suit and black tie at an indoor eventPhoto: Gage Skidmore / Wikimedia Commons, CC BY-SA 4.0

SpaceX has raised $40 billion in debt to buy GPUs, the chips that run AI systems, Elon Musk announced on X on Sunday, CoinDesk reported. At today's interbank rate that is about Rs 11 trillion, more than the federal government's whole power-sector subsidy budget many times over.

Reuters had earlier reported that SpaceX was negotiating with banks and asset managers for the money to buy Nvidia AI chips. The debt is mostly callable, which means SpaceX can repay it early. CoinDesk said the money could pay for chip deliveries expected in the first half of 2027, though Musk did not say so directly.

Musk's warning to the 'neoclouds'

Neoclouds are the newer companies that buy AI chips and rent out computing power. Musk wrote that "debt markets have almost entirely shut off access to money" for firms that do not have substantial income outside AI data centres, and predicted that the more than 200 neoclouds will consolidate, with only a few surviving.

He said SpaceX could borrow because of its strong credit rating and steady income from Starlink and rocket launches, neither of which depends on AI demand. Investor Chamath Palihapitiya added that many neoclouds look poorly funded and badly run, and that data-centre delays could leave chips idle in 2027, unable to cover their financing costs.

Nvidia wants a software fee on top

The chip bill is only part of the cost. At GTC Berlin on 22 October, Nvidia chief executive Jensen Huang said it is "about $2,000 per GPU per year to run all the software that Nvidia provides that runs agents on these AI factories", ARY News reported. That is roughly Rs 553,600 a chip each year. Nvidia has not said which products the fee would cover or when it would start, and the figure may change. Its existing AI Enterprise subscription costs $4,500 per GPU a year.

"Software licensing will become a major part of our company's business," Huang said. The fee would cover tools such as OpenShell, which gives AI agents a secure workspace and limits what they can reach, and Nvidia Sentry, which watches agents and isolates any that try to break their limits.

Why it matters in Pakistan

Pakistan's own push for AI data centres, and the local cloud firms that rent GPU capacity to startups and universities, buy the same chips in dollars. A tighter lending market abroad and a yearly software fee per chip raise the cost of every GPU hour sold here. For what a dollar price means in rupees today, use our currency converter or the dollar rate page.

Frequently asked questions

What is a neocloud?
A newer company that buys AI chips and rents out computing power to others, as opposed to the big cloud providers such as Amazon, Microsoft and Google.
Comments
WhatsApp

Sources

Tags:Elon MuskNvidiaAI chipsdata centresSpaceX

More in AI

Reader comments 0

No comments yet. Say something useful.