NVIDIA pitches “AI factory” revenue sharing as Blackwell demand turns into usage-linked business

NVIDIA is trying to move beyond simply selling the chips that power artificial intelligence data centers. In recent discussion circulating in markets coverage, the company’s “AI factory” approach was framed as a way to monetize demand for its Blackwell platform more directly, potentially converting large infrastructure build-outs into a revenue stream tied to how much computing capacity customers actually use.

The idea, as described in market coverage, centers on partnerships with AI cloud companies and large-scale infrastructure operators. Under the model, AI customers sell Nvidia-powered cloud services while Nvidia still collects standard product revenue and also takes a share of cloud revenue on supported capacity, according to the market analysis. That structure is meant to make Blackwell-linked demand less dependent on hardware timing alone and more connected to utilization over time.

The market focus comes as analysts adjust expectations for NVIDIA’s next fiscal period. One market note pointed to 44 upward and 4 downward EPS revisions for the upcoming fiscal year over the last three months, alongside 46 upward and 4 downward revenue estimate revisions. The same note described the latest catalyst as the monetization angle, specifically how Nvidia could tie AI infrastructure demand to a recurring, usage-linked revenue structure.

Partnership examples cited in the coverage include a deployment by Sharon AI in Australia, where the company plans to bring online up to 40,000 Grace Blackwell GB300 GPUs. Another example described in the same write-up involves Firmus building an AI factory campus in Batam, Indonesia, with an expected scale-up to 360 megawatts and up to 170,000 GPUs. The larger theme is that Nvidia is encouraging the creation of “multi-tenant” AI factories, which can support more than one customer’s workloads and, in theory, make utilization tracking and revenue sharing more meaningful.

Separately, a Business Wire release republished by StockTitan outlined a compute collaboration in Australia between Nvidia and Sharon AI that includes a multi-year deployment and a named target of up to 40,000 Grace Blackwell GB300 GPUs. The release also used the “AI factory” framing, underscoring how Nvidia’s system-level approach is meant to scale by pairing accelerators with datacenter capacity planning.

NVIDIA’s investor story in data-center AI has increasingly depended on demand for its newest GPU architectures, and Blackwell is the key next step in that roadmap. The company’s ability to secure long-duration infrastructure commitments matters because the training and inference workloads that drive purchases are capacity hungry, and the industry is racing to build power, networking, and cooling as much as it is buying accelerators. Usage-linked terms, if they are meaningful in magnitude, would align Nvidia’s revenue with actual consumption of AI services running on its hardware.

Still, important details are not clear from the market coverage alone. Neither the Yahoo Finance post nor the secondary write-up provides specifics on the size of Nvidia’s usage-based revenue share, how “supported capacity” is defined contractually, or what happens if utilization ramps slower than expected. The operational and financial performance of these arrangements will likely depend on customer economics, tenancy levels, and the durability of demand for the compute services sold by partners.

For investors and analysts, the next announcement to watch will be whether usage-linked revenue appears in disclosures around segment performance, margins, or any specific commentary on the AI factory model’s contribution. Updates on additional AI factory deployments, the pace of scaling (measured in megawatts and number of GPUs brought online), and any disclosures on contract economics would help determine whether the concept turns from a strategic initiative into a measurable financial tailwind.