---
title: "Nvidia Has 323 GPU Clouds to Sell To, and Picking One Is the Hard Part"
description: "Hyperscalers, neoclouds, baby neoclouds, bring-your-own-chip plans and SpaceX: a field guide to the compute menu"
author: "Nate Ledger"
published: 2026-10-10T11:00:01Z
modified: 2026-10-10T13:36:44Z
url: https://rews.cc/a/nvidia-has-323-gpu-clouds-to-sell-to-and-picking-one-is-the--306d36
language: en
tags: ["nvidia", "ai", "cloud", "chips", "neoclouds", "tech"]
publisher: "Rews (https://rews.cc)"
---

# Nvidia Has 323 GPU Clouds to Sell To, and Picking One Is the Hard Part

*Hyperscalers, neoclouds, baby neoclouds, bring-your-own-chip plans and SpaceX: a field guide to the compute menu*

By Nate Ledger · October 10, 2026 · https://rews.cc/a/nvidia-has-323-gpu-clouds-to-sell-to-and-picking-one-is-the--306d36

## In brief

- SemiAnalysis counted 323 Nvidia GPU providers in September, up from 209 less than 11 months earlier
- Nvidia expects $108 billion in October-quarter revenue, an 89% year-over-year jump, with a market cap near $6 trillion
- Anthropic and OpenAI have committed over $500 billion to Amazon and Microsoft, which hold 59% of the cloud market
- Oracle now lets clients bring their own GPUs; OpenAI has pledged over $300 billion to Oracle across five years
- SpaceX rents GPUs to Anthropic for $1.25 billion a month through mid-2029 and says payback is under a year

Nvidia’s stock hit yet another record this week, pushing its market value close to $6 trillion, and management is projecting $108 billion in revenue for the October quarter — up 89% from a year earlier. If you are Nvidia, the AI buildout is going wonderfully. If you are one of the companies doing the building, you face a different and stranger problem: there are now so many places to rent Nvidia’s chips that choosing among them has become a chore in itself. [As CNBC’s Jordan Novet lays out](https://cnbc.com/2026/10/10/nvidia-gpus-are-everywhere-heres-how-companies-access-them.html), the menu runs from trillion-dollar cloud giants to startups nobody has heard of, with a few genuinely odd stops in between.

The research firm SemiAnalysis counted 323 providers of Nvidia GPU capacity as of September, up from 209 less than 11 months earlier — a 55% increase, in under a year, in the number of companies whose pitch is *we have the thing you cannot get*. Nvidia CEO Jensen Huang told a Goldman Sachs tech conference in San Francisco last month to expect “a whole new crop of really, really exciting neoclouds with hundreds of billions of dollars backlog together.” Note who the crop is good for: the more clouds competing to buy chips, the more chips get bought. Nvidia’s customer base is broadening to match — five clients each accounted for at least 10% of its accounts receivable in the July quarter, up from three in January, according to a filing.

## The safe choice

Most big companies already hand Amazon, Microsoft and Google tens of millions of dollars a year for cloud services, and since ChatGPT arrived in 2022 they have added GPUs to the tab. The draw is partly reputational: no enterprise procurement department ever got fired for buying Microsoft. “When you’re talking to enterprises, your subprocessor had better be Azure,” said Bindu Reddy, CEO of the AI assistant startup Abacus. In the past year alone, Anthropic and OpenAI have committed to spending over $500 billion between Amazon and Microsoft, which together controlled 59% of the cloud infrastructure market in 2025, according to Gartner.

The catch is supply. Gartner analyst Hardeep Singh says the hyperscalers’ “10-plus years of full-stack capabilities” buy them enterprise trust, but they don’t always have enough GPUs. Amazon CEO Andy Jassy told analysts in July that his company won’t be able to serve all the demand it sees this year, adding: “I believe this dynamic will also be true in 2027.”

## The neoclouds you’ve heard of

If the hyperscalers had enough chips, the neoclouds would not exist. Modal, a startup that runs virtual sandboxes for AI agents, started on the hyperscalers, found it could only ever get “a few hundred GPUs or maybe a thousand,” and now spreads its business across 25 neoclouds, said CEO Erik Bernhardsson. The arrangement gets circular quickly: Google and Microsoft themselves rent from CoreWeave, the biggest neocloud, even while competing with it. “Some of the hyperscalers have approached us about taking care of customers they’re worried about because they don’t have the ability to service those customers when they need it,” said Marc Boroditsky, chief revenue officer of the Netherlands-based neocloud Nebius. Video startup Reactor splits its workloads between Nebius and the big clouds, CEO Alberto Taiuti said, because it needs its GPUs physically close to users so their videos render fast.

There is also a financing quirk worth understanding. The big neoclouds often want upfront payment, and your chips may not switch on for months — because your contract is what the provider raises money against to build the data center you’re waiting for. You are, in a sense, the collateral for your own compute. CoreWeave’s near-term capacity is essentially sold out, CEO Mike Intrator said on the August earnings call, and turning over 10,000 GPUs to a new customer on a day’s notice would be impossible, said executive vice president Chen Goldberg.

## The neoclouds you haven’t

Below the name brands sit hundreds of smaller providers, some built for specific countries, some offering “bare-metal” GPUs that hand customers more control and more of the sysadmin work. They tend to be more flexible on upfront payments and contract length — but discretion is part of the product. “Capacity right now is tight, and your relationships with your suppliers is actually one of the most closely guarded secrets for companies like ours,” said Zhen Lu, CEO of Runpod. In a shortage market, the questions never change — when can I get the chips, and at what price — and Sunny Smith, co-founder and technology chief of Massed Compute, says customers commit to capacity precisely when they expect prices to rise. The long-term contract is the hedge.

## Bring your own

Then there is Oracle, which has more debt than Amazon or Microsoft and a lower credit rating, and so cannot simply outspend everyone on chips. Its answer is elegant: you buy the GPUs, Oracle runs them. “As we’re generally able to preserve and improve margins in the case of things like bring-your-own-hardware, the ROIC for those types of structures will be even higher,” CFO Hilary Maxson said on the June earnings call — return on invested capital being what you get when someone else’s capital does the investing. Guggenheim analyst John DiFucci notes it would make sense for AMD and Nvidia themselves to park chips there. OpenAI has committed more than $300 billion to Oracle over five years, though it hasn’t said anything about bringing its own GPUs, and declined to comment.

## Rent from a rocket company

The strangest suppliers are the ones for whom compute is a side hustle. SpaceX has agreed to hand excess GPU capacity to Google and to the open-source startup Reflection; in April it agreed to supply GPUs to the AI coding startup Cursor and then simply bought the company for $60 billion; in May it signed a deal to rent GPUs to Anthropic for $1.25 billion a month through mid-2029. “The current economics have translated into a less than one-year payback on our new capital deployments for compute,” SpaceX finance chief Bret Johnsen told analysts in August — which is a lovely sentence about what happens when a launch company’s spare chips rent for more per year than most aircraft. CNBC reported in July that Meta is building a cloud unit to sell AI compute too.

A year ago the AI question was who could get GPUs at all. The market’s answer turned out to be: almost anyone, if you don’t mind prepaying a landlord-in-waiting, signing a secrecy pact with a startup, renting from a rocket company or shipping your own chips to Oracle. Nvidia faces no such paradox of choice. It has 323 distribution channels.