NVIDIA to change to cloud monthly rentals? Its first base will deploy 170,000 GPUs, adopting revenue sharing plus credit support.

Nvidia CFO Colette Kress announced a new profit-sharing model under the brand "DSX AI Factory," which no longer requires data center operators to buy out GPUs in one go, instead using revenue sharing and credit support in exchange for cloud usage revenue.
(Previous: Build an AI Customer Service Agent in 2 Minutes! xAI Launches No-Code Voice Agent Builder, Outperforms OpenAI in Voice Benchmarks)
(Background: Dell Partners with NVIDIA to Launch "All-Liquid Cooled" AI Servers! First to Feature Vera Rubin Architecture, Single Cabinet with 144 GPUs Delivering Record-breaking Compute Power)

Table of Contents

Toggle

• Nvidia Switches to Cloud Subscription Model
• 170,000 GPUs All-In on Indonesia
• Capital Loop: Flywheel or Boomerang

Nvidia is no longer just selling shovels; it now wants a monthly cut from the gold miners. In the past, Nvidia sold GPUs to data center operators on a cash-and-carry basis, ending the transaction after delivery, earning one-time hardware revenue.

Now, Nvidia has introduced a new model under the "DSX AI Factory" brand: instead of requiring customers to buy out hardware in one lump sum, it uses "revenue sharing + credit support" to deploy infrastructure, in exchange for a long-term income stream tied to cloud usage.

This changes Nvidia's profit structure: from "earning money once per hardware sale" to "investing once and collecting royalties forever." Nvidia is essentially tying its financial performance firmly to its customers' cloud bills quarter after quarter.

Nvidia Switches to Cloud Subscription Model

On Wednesday, Nvidia CFO Colette Kress published a blog post explaining the new "DSX AI Factory" collaboration model.

The core approach is that Nvidia no longer insists data center operators pay cash upfront to buy GPUs; instead, it uses a combination of "revenue sharing + credit support" to supply infrastructure. Nvidia provides hardware and capital in exchange for a share of the cloud service provider's future revenue.

Another supporting measure is issuing "token credits" to developers, allowing resource-limited research teams and startups to use computing power now and repay later through revenue sharing.

Nvidia is effectively acting as a matchmaker for AI data center operators and cloud service providers, creating a usage-based revenue stream for itself instead of just selling hardware outright.

Kress wrote in the article:

"For model developers, inference service providers, agent platforms, and enterprises scaling AI, this means they can gain access to full-stack accelerated computing faster, without having to wait through the lengthy process of site selection, power procurement, facility construction, and hardware deployment."

This statement highlights the target customers: small and medium-sized players who lack the capital to build large-scale AI data centers but urgently need computing power. In other words, Nvidia has quietly upgraded itself from a "device supplier" to a "computing power landlord," even factoring in whether customers can sustain the business into its own financial blueprint.

170,000 GPUs All-In on Indonesia

The first testbed for this model is in Batam, Indonesia. Local firm Firmus officially announced a partnership with Nvidia to build a "DSX AI Factory" campus with a scale of up to 360 MW, deploying up to 170,000 Nvidia GPUs.

Firmus expects that based on signed off-take agreements alone, the first six years of collaboration could generate $25 billion to $30 billion in revenue.

The business logic behind this is not complicated. Nvidia provides capital and hardware while locking customers into running models on its own hardware stack; every dollar invested in this ecosystem eventually becomes new demand for Nvidia chips.

In simple terms, where Nvidia previously sold GPUs and ended the transaction, it now turns "one-time sales" into "ongoing royalties" through revenue sharing and credit support. The longer and more heavily customers use the service, the more stable Nvidia's profit share becomes, and the better its financials look.

This is also an extension of Nvidia's "computing power equals revenue" strategy.

In the past, Nvidia earned hardware revenue by selling shovels; now, through equity holdings, off-take agreements, and revenue sharing, Nvidia directly enters its customers' balance sheets, becoming a key factor in whether customers can sustain revenue.

For capital-constrained researchers and startups, this is a faster path to computing power. But for Nvidia, it is also a way to tightly bind AI infrastructure demand to itself—demand is no longer solely driven by the market; part of it is created by Nvidia's own investment.

Capital Loop: Flywheel or Boomerang

However, the concern is that this arrangement—where vendors invest in customers, and customers then buy from vendors—has grown so large that it unsettles outsiders.

Bloomberg analysis indicates that by 2026, the total scale of such transactions is estimated to exceed $800 billion. Nvidia's own books also hint at this: non-trading equity investments surged from $3.39 billion at the start of fiscal 2025 to $22.25 billion by the end of January 2026, with roughly $53 billion deployed across 170 deals in just one year.

Analysts compare this arrangement to "vendor financing" during the dot-com bubble, when optical equipment makers like Nortel used financing to prop up orders, creating an illusion of demand that eventually dragged down the entire telecom market.

Nvidia denies playing a capital loop game, but well-known short sellers are not buying it—for a simple reason: when a portion of customer orders is funded by the supplier itself, the quality of that revenue must be questioned.

For reference, heavy users like OpenAI are projected to lose as much as $14 billion in 2026, with both burn rate and demand authenticity under market scrutiny. That's why every time Nvidia unveils a new profit-sharing plan, the market is both excited and nervous—excited because computing power expansion speeds up, and nervous because no one can guarantee that this capital loop won't collapse, leaving someone unable to hold on.

Nvidia has transformed from a shovel seller into a shareholder and landlord of the gold miners. When a significant portion of demand is created by borrowing money and issuing equity, the line between boom and bubble may come down to one unanswered question: Is this capital loop a flywheel that spins Nvidia ever faster, or a boomerang that will eventually swing back and hit it?