by

Nvidia turning loans into AI leverage

Nvidia is turning chip finance into another weapon as lenders queue up to bankroll the AI gold rush.

According to the Financial Times, when Nvidia first lent big sums to customers to help them buy its chips, it stirred fears of an AI bubble inflated by “circular” financing. A year later, the chip giant has found ways to limit its exposure while dragging more financiers into the game.

The shift suggests a new use for Nvidia’s balance sheet and another phase in the AI boom. Doubters are still muttering, but the world’s biggest chip company is getting smarter about who carries the can.

One clue came last week, when six financiers from the cocaine-nose jobs of Wall Street agreed on a framework to extend about $500 billion in financing to Nvidia customers. Nvidia would guarantee up to a quarter of some transactions, then it said it would provide an up to $105 billion backstop for a new OpenAI data centre in Ohio.

Those guarantees could still leave Nvidia exposed to large sums, and no one seems sure how much more risk it will accept. Moody’s said the deals had not weakened Nvidia’s “excellent financial profile” and kept its credit rating intact.

Nvidia CEO Jensen Huang says the company’s chips are now so widely used that they have become a new asset class for lenders. The snag is that this asset class lacks the long history that credit analysts usually need before they can pretend everything is fine.

The supply crunch has created freakish conditions. SpaceX recently said it could make enough from renting AI facilities to cover the full cost of building its AI data centres in less than a year.

Nvidia argues that if one customer defaults, another will always take the spare capacity because its chip architecture and CUDA software are everywhere. CoreWeave helped that case last week by signing a deal to rent Nvidia A100 chips until 2029, nine years after that generation appeared.

A downturn would make that logic look a bit ropey. If newer, more efficient chips were in surplus, running older, power-hungry A100s would make less sense, even after depreciation had done its accounting magic.

Moody’s predicts a US data centre capacity shortage for “at least the next four to five years”, so lenders may keep squinting past the risks. Nvidia is now backing neoclouds, sovereign customers and enterprises outside the giant US cloud platforms, while taking a slice of data centre revenues in what Stifel analysts called “an emerging asset-light pseudo-cloud”.

TOPICS:
ai chips  ·  AI finance  ·  coreweave  ·  CUDA  ·  data centres  ·  Jensen Huang  ·  neoclouds  ·  Nvidia  ·  openai

Latest articles

Share

Featured articles

Hot topics

No results found.

Latest reviews