← Back to feed

Nvidia is launching a revenue-sharing program for AI cloud operators, giving partners access to large-scale GPU deployments in exchange for a percentage of the inference revenue those GPUs generate. The first two disclosed partners — Sharon AI and Firmus — will operate a combined 210,000-plus Nvidia accelerators under the model.

This is a meaningful departure from Nvidia’s standard hardware-sale business. Instead of collecting cash upfront, Nvidia takes a downstream cut of token economics. It is the GPU maker becoming a compute co-investor.

The Deployments

Sharon AI is taking 40,000 Grace Blackwell GB300 GPUs — the top of Nvidia’s current inference lineup at roughly $250,000–$300,000 per unit list price. At that scale, the hardware value approaches $10B before any discount.

Firmus is deploying up to 170,000 Nvidia GPUs in a 360-megawatt campus in Batam, Indonesia — a site chosen for power availability and proximity to Southeast Asian enterprise demand. Firmus has been scaling its footprint across Asian markets and the Batam campus is its largest commitment.

Combined, the two deployments represent more than 210,000 accelerators — a scale comparable to some of the largest single hyperscaler clusters deployed in 2025.

Why Nvidia Is Doing This

The chip sales model works at the top. Hyperscalers — Google, Microsoft, Amazon, Meta — buy Nvidia hardware at scale regardless of near-term utilisation because they can absorb underused capacity and have diversified revenue to fund it.

Smaller AI cloud operators cannot. They need hardware to generate revenue, but cannot buy hardware without capital, and cannot raise capital without proving revenue. The revenue-share model breaks that deadlock: Nvidia provides the hardware, the cloud operator provides the operations and go-to-market, and they split the upside.

For Nvidia, this opens a new tier of customers that its standard sales motion cannot reach. It also gives Nvidia a recurring revenue stream tied to inference volume — a hedge against the quarterly lumpiness of hardware sale cycles.

Jensen Huang has signalled for several quarters that Nvidia sees its role expanding from chip supplier to “AI factory” builder. This program makes that concrete: Nvidia is not just selling the machinery, it is financing the factory and taking a production cut.

The Competitive Dimension

This move has implications for CoreWeave, Nebius, Lambda Labs, and the broader neo-cloud tier. Those operators currently buy Nvidia hardware outright (or finance it through debt), then mark up the compute to enterprise customers. If Nvidia starts entering revenue-share arrangements directly with cloud operators, it competes with the capital stack that neo-cloud providers were using to differentiate.

CoreWeave, which went public in early 2026, has built its business on being Nvidia’s best downstream partner — taking hardware fast, at scale, and deploying it to AI labs and hyperscalers. The revenue-share model could signal Nvidia wants a more direct economic relationship with that layer of the stack.

Nvidia’s $2B investment in coherent interconnect infrastructure (photonics) and its commitment of $150B annually to Taiwan manufacturing are parallel moves in the same direction: vertically integrating down from chip design toward the physical and economic substrate of AI compute.

What It Means for Southeast Asian AI

Firmus anchoring 170,000 GPUs in Batam, Indonesia, is a notable data point for regional AI infrastructure. Southeast Asia has been underserved by frontier GPU capacity relative to its AI demand growth. Indonesia’s government has pushed for domestic AI compute sovereignty; Firmus deploying at 360MW in Batam fits that policy environment.

The move also signals that Nvidia’s revenue-share model is not limited to US or European markets. If the Firmus deployment proves the economics, expect Nvidia to extend the model to additional operators across the Asia-Pacific region where capital-constrained cloud operators face the steepest GPU access barriers.

The net effect is Nvidia moving from chipmaker to infrastructure financier — a transformation that mirrors what happened when AWS turned server hardware into cloud services, but from the supply side of the GPU stack.