News · 2026-07-04

NVIDIA Starts Taking a Share of Its Cloud Partners' Revenue, Not Just Selling Them Chips

NVIDIA has announced a new financial arrangement in which it earns a share of the cloud revenue its partners generate from renting out NVIDIA-supported data center capacity, on top of the revenue it already collects from selling them the chips. The announcement, posted on NVIDIA's own blog on July 1, 2026, names its first two capacity partners.

Key facts

• NVIDIA will earn chip revenue plus a share of cloud revenue partners make on NVIDIA-supported capacity, announced July 1, 2026.
• First named partners: Sharon AI, deploying 40,000 GB300 GPUs, and Firmus, building a 360-megawatt campus in Batam targeting 170,000 GPUs.
• NVIDIA cited AI-native companies Baseten, Fireworks, and Together as evidence of demand for this capacity.
• Source: NVIDIA's own blog post announcing the arrangement.

To see why this is a notable shift, it helps to understand how the AI infrastructure business has worked up to now. NVIDIA designs the graphics processors, or GPUs, that have become the standard engines for training and running AI models. Companies that want to offer AI computing power - the rented capacity that startups and enterprises use instead of buying their own hardware - purchase mountains of these chips, install them in data centers, and rent access to them by the hour. NVIDIA's role in that chain has been straightforward: it sells the GPUs, data center operators buy them, and what happens after the sale - how much the resulting compute gets rented out, at what price, to whom - has been someone else's business entirely.

This announcement changes that boundary. NVIDIA is now positioning itself to earn ongoing revenue tied to how that capacity actually performs as a rented service, not just a one-time hardware transaction. In other words, it collects money when it sells the chip, and then it keeps collecting a share as the chip earns its keep out in the world. That turns a single sale into a recurring relationship, and it gives NVIDIA a direct financial stake in how busy and how profitable its partners' data centers become.

The two named partners give a sense of scale. Sharon AI is deploying 40,000 of NVIDIA's GB300 GPUs, the company's high-end chips built for demanding AI workloads. Firmus is building a data center campus in Batam, Indonesia rated at 360 megawatts of power - roughly enough electricity to run a small city - aimed at eventually housing 170,000 GPUs. Numbers that large are hard to picture, but the point is that these are not modest server rooms; they are industrial-scale installations that consume the kind of power once associated with heavy manufacturing. NVIDIA also pointed to AI-native inference companies like Baseten, Fireworks, and Together as evidence that demand for this kind of capacity is real and growing, not speculative. Those are the businesses that actually run AI models for paying customers, so their appetite for compute is a signal that the capacity being built has buyers waiting.

Think of the old model as a hardware store selling someone a truck. It is one transaction, and what the buyer does with the truck afterward - how many deliveries it makes, how much it earns - is entirely the buyer's business. The new model is closer to the hardware store taking a cut of every delivery fee the truck ever earns. NVIDIA is not becoming a cloud provider itself here; it is still the supplier, and it is not the one signing up rental customers or running the data centers day to day. But it is inserting itself into the ongoing economics of the businesses being built on top of its chips, rather than stepping away once the sale closes. That is a meaningful change in where NVIDIA sits in the money flow.

Why it matters: NVIDIA already sits at the center of the AI buildout as the near-default supplier of the GPUs everyone needs, which gives it extraordinary leverage over an entire industry's costs and timelines (see The $660 Billion AI Buildout for the scale of the money flowing into that build-out, and NVIDIA's Warm-Water Fix for AI-Thirsty Data Centers for how far downstream NVIDIA is already reaching into data center operations). Taking a slice of cloud revenue on top of chip sales pushes NVIDIA further downstream still, tying its fortunes even more tightly to how much AI compute actually gets used, not just how many chips get shipped out the door. A single named partner here, Firmus, is alone building toward 170,000 GPUs on one campus - a reminder of just how much physical infrastructure is being erected to support this shift, and how much of that infrastructure now has NVIDIA's economic fingerprints on it beyond the initial sale.

The honest caveat: this is a newly announced financial structure, and its effect on cloud pricing, competition among cloud providers, and smaller players trying to compete without NVIDIA's blessing is not yet clear. A partner sharing revenue with NVIDIA might pass those costs on to customers, or the arrangement might come with advantages that make NVIDIA-blessed capacity cheaper - it is too early to tell which. It also further concentrates NVIDIA's grip on the AI supply chain at a moment when regulators in multiple countries are already scrutinizing that concentration.

Primary source, verified: read the paper →

Ground Truth. (2026, July 4). NVIDIA Starts Taking a Share of Its Cloud Partners' Revenue, Not Just Selling Them Chips. Ground Truth. https://groundtruth.day/news/nvidia-starts-taking-a-cut-of-its-clouds-revenue.html

@misc{groundtruth:nvidia-starts-taking-a-cut-of-its-clouds-revenue,
  title  = {NVIDIA Starts Taking a Share of Its Cloud Partners' Revenue, Not Just Selling Them Chips},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {jul},
  url    = {https://groundtruth.day/news/nvidia-starts-taking-a-cut-of-its-clouds-revenue.html}
}

Topics: NVIDIA · data centers · cloud · infrastructure · business